An India media startup got the internet talking after adopting AI enabled it to double sales in a year to cross $500 million in projected annual revenue, but at what cost? Its platform appears full of slop, which raises questions over who its customers are and how much AI generated content they will put up with.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Can AI slop lay the pathway to a billion-dollar funding round and IPO?

AI is helping one of India’s media businesses supersize its production and reach global markets like never before. Pocket FM, a platform for audio content, made waves last week when it claimed to have doubled its annual revenue rate (ARR) to over $500 million in just one year, but at what cost does embracing AI for media come at?

The company started out in 2018 to fill the time commuting to and from work with podcasts and other audio entertainment shows, raising close to $200 million from investors like Tencent and Lightspeed. Its last round in 2024 came at a $750 million value, and it claims to have more than 250 million listeners from across over 20 countries.

Pocket FM monetises using advertising and a pay-as-you-go model requiring users to buy credit to access shows, as it found an all-encompassing subscription like Netflix and Spotify didn’t land. Pocket FM claims more than 550,000 writers registered to its service. Writers release content and take a cut of revenues made. If the company sees a writer with promise, they back them with resources and help.

That was the model, but adopting AI has enabled it to massively increase its library of content, which it says are driving audience growth, retention and monetisation. Previously, making 100 hours of content took a year, but that can now be done in just one day, CEO Rohan Nayak told TechCrunch. AI now “powers” 93% of the platform’s content, he added.

Indeed, a deal with AI audio startup ElevenLabs helps Pocket FM writers make audio content “with one click.” This all screams AI slop, and quite justifiably. I am not the target audience, but Pocket FM’s service does not appeal to me in the slightest when I open it.

Programme tiles are clearly AI generated. When I did open an episode, the narration was in a voice easily identifiable as not human with video that was also obviously not made by a person. Pocket FM has joined the bandwagon of AI-generated microdramas, which do all of the above in vertical frames with episodes of up to 90-120 seconds.

This is not my taste, although the company claims it is working. According to the TechCrunch story, its Pocket Saga app has reached a $15 million ARR just three months after launch with the US its only market right now. The US accounts for 70% of revenue, with revenue in Germany and France reaching $21 million in 11 months and $10 million in three months, respectively, Nayak wrote on LinkedIn.

“Our Bet [is that] In ~3 years we will be able to produce [a] Naruto-like series for $1000,” he added. “Netflix has to spend $17 billion to find 100 Blockbusters per year. What happens when you can produce 1 million high quality shows for $1 billion?”

Pocket FM may have found a way to make money from AI slop videos and audio, but there’s no valid comparison to be made to Netflix, which is cinematic in contrast to Pocket FM. Its revenue success is hardly panacea to the creative industry, which has been polarised by generative AI, despite recent agreements like ByteDance and the Motion Picture Association inking a copyright pact.

It also hasn’t been plain sailing for Pocket FM. The company let go of more than 100 employees in May, that’s around 10% of its workforce, with reports that more cuts would come.

So why is the company speaking up now? If you want to get cynical, it has been streamlining its costs to look good for investors.

It was linked with a $100-$120 million funding round in April. In June, it was reported to have begun discussions to shift its holding structure back to India (from the US) to enable a domestic IPO. Nayak told TechCrunch that there’s no immediate plan for a listing in India.

The founders will want to position their business as a global and AI-first company to maximise their valuation, and potential for a US IPO. The data looks promising on paper, until you open the app and use the product.

Subscribe now

America’s AI giants are poaching from big tech to grow in Asia

OpenAI and Anthropic are the new big tech, and that’s showing with its hiring in Asia as the duo rush to expand in the continent.

We wrote before about how OpenAI is diving into the opportunities in Southeast Asia, working closely with the government and public sectors in Singapore and opening a startup accelerator programme alongside the Thai government. Anthropic has taken a different approach by focusing on the more mature market opportunities in South Korea and Japan, but it is now increasing its efforts in India and Singapore, Rest of World reports. It is poaching from the likes of Google, Microsoft and Amazon in proof of what is a new tech world order in Asia.

The playbook is very much classic US corporate tech, with a focus on public policy and other relationship building roles, as well as sales. OpenAI is diving into the experimental potential, through its work with the Singapore government with training opportunities and public-private partnerships, not to mention the Thailand programme, which is something it hasn’t tried anywhere else in the world.

Deals

Bengaluru-based Swish raised $24 million to expand its own kitchen and delivery network to compete with 10-minute food-delivery services [Economic Times]

US belt and road? The Trump administration will lend nearly $100 million to Africell, Africa’s only American-owned telecom operator, to deploy mobile-network equipment from US and allied suppliers [Reuters]

Markets

Tencent-backed Chinese AI chipmaker Enflame surged 206% in its Shanghai debut after its retail offering drew orders for more than 6,000 times the shares available [CNBC]

Chinese AI developer Z.ai is raising about $5 billion through a Hong Kong share placement and convertible bond sale [Reuters]

AI and Chips

We recently critiqued Moonshot AI’s $50 billion valuation with $300 million ARR but now Bloomberg reported that figure has jumped to over $1 billion as of August. A very timely leak as the company plots an IPO [Bloomberg]

We also know Chinese AI firms need to make money, so here’s Bloomberg again looking at how China’s AI sector is moving from competing on large models to commercialising AI agents [Bloomberg]

Newly-public Unitree Robotics grew humanoid-robot revenue eightfold to $129 million in 2025 [Ars Technica]

Flipkart’s Super.money is rolling out AI agents to its 20 million users to shop on Flipkart and buy gold at user-set prices [Bloomberg]

Policy

South Korea expanded its espionage laws to cover theft of sensitive economic and technology information, including chip secrets, with those acting for foreign countries now facing up to 30 years in prison [FT]

China set a 2030 target for large-scale deployment of autonomous vehicles on highways, urban expressways and selected city roads [Reuters]

The European Commission is preparing legislation to close enforcement gaps that let illegal goods sold on e-commerce platforms reach EU consumers [Euronews]

In other news:

Tokyo startup MW plans to sell its first robot-equipped home by 2028, embedding robotic arms in ceilings and walls rather than building humanoids [Bloomberg]

A misconfigured database exposed more than 220 million passenger and crew records, including passport numbers and flight details, from travellers flying to, from or through Vietnam between January 2017 and April 2026 [Bleeping Computer]

Amazon Pay India is considering travel insurance and other products to push beyond payments after initially launching health insurance [Reuters]

Hugging Face, OpenRouter, these two AI infrastructure platforms were just sold in deals that combined reach nearly $20 billion. Now Asia is seeing its own versions rising with UniPat, a startup created by a former Alibaba AI employee, linked with raising $300 million led by Alibaba itself. It may be Friday, but today’s newsletter is packed with deals across AI, a Grab rumour and more.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Alibaba is getting into the AI infrastructure game with a former intern

Deals around AI infrastructure are a major focus right now. Following Stripe’s $7.5 billion acquisition of OpenRouter and Nvidia’s $12.9 billion purchase of Hugging Face, Alibaba is making its own play. Bloomberg reports that the Chinese firm will lead a $300 million investment in training and benchmarking platform UniPat at a $2.5 billion valuation.

UniPat CEO Li Kuan is a former intern at Alibaba’s Tongyi AI division so there are already plenty of links between the two companies. But, more broadly, it’s clear that focus has moved into the infrastructure and platforms that underpin how AI is measured and trained.

Hugging Face and OpenRouter are very much along these lines. Both operate as central points through which users can easily switch between hundreds of AI models in an instant and with the same account and credit balance. Since they are repositories for the world’s top AI models, they generate vast attention and traffic which also makes them valuable.

UniPat isn’t exactly the same, but there are plenty of similarities, as Bloomberg explains:

Founded in late 2025, UniPat hosts benchmarks evaluating AI coding agents on software engineering tasks, browser agents executing everyday web actions, and frontier multimodal models on visual reasoning problems. Li’s research lab — with backing from Monolith and ByteDance Ltd.-backed Jinqiu Fund — also builds its own smaller models for purposes like prediction and scientific research.

Aside from a reliable read on the power of AI models, UniPat is also providing human-generated internet data, a resource that many companies training models are desperately seeking and struggling to find.

Alibaba has its highly-regarded Qwen open weight model, and it has invested significantly in bringing AI into its core business and enterprise customers as we’ve noted a number of times this year. There’s no doubt this platform is already useful to the company, but it makes sense that it would be even more interesting as Alibaba looks to extend its reach and touchpoints with potential customers and the wider community of AI users.

Alongside Alibaba, the speculated investors include Tencent and HSG, HongShan Group aka Sequoia China.

Subscribe now

Indonesia’s Zankore secures $3.1 billion to build out Nvidia-powered AI data centre

On a related AI note, we don’t often see major companies coming out of Southeast Asia but Indonesia-based AI infrastructure platform Zankore, which is backed by Nvidia, just secured a $3.1 billion loan to purchase advanced chips.

That deal is one of the largest financings for a company of this type in Asia.

The money will go towards buying and deploying Nvidia GPUs (this is why Nvidia backs these companies) for a new AI factory that is planned to reach 1 gigawatt in capacity. Aside from Nvidia, Zankore is backed by Nokia, telco Indosat and its parent company Ooredoo.

We’ve written about the challenges of making sovereign AI, i.e. AI systems that are not reliant on international companies and products, a reality in Asia, and particularly Southeast Asia. A compromise of sorts looks to be having local infrastructure like Zankore as the data centre layer. That is just one part, of course, and the lack of local AI models remains an issue.

Grab eyes BNPL as the next leg of its fintech growth story

We know Grab is doubling down on its financial services business, as it finally bids to turn its fintech potential into clear growth, and now there’s a hint at what further direction that might take. Bloomberg reported that the ride-hailing firm is in talks to take a majority ownership stake in Atome Financial, the Singapore-headquartered buy-now-pay-later platform, in a deal that would value it at upwards of $2 billion.

Grab has more than $5 billion in cash to spend and, as we’ve said repeatedly, it will use that for strategic M&A. It made a series of investments in autonomous vehicle tech last year, acquired US-based investment platform Stash and it is moving into a new market with the $600 million purchase of Foodpanda Taiwan. Adding BNPL would complement its existing payments and lending push.

Disclosure: Atome is an Ellerton & Co. client. This item is based on Bloomberg reporting.

Deals

We ran out of space to analyse this fascinating deal, but TikTok will invest $980 million in a 50-year logistics project in Ho Chi Minh City, with its Tokgistic unit initially contributing $196 million before operations begin in November [TechNode]

Uber invested $10 million in Indian fleet-management startup Carrum Mobility at a $168 million valuation, up from $63 million when the US firm made an investment in January [TechCrunch]

Bengaluru-based QNu Labs raised Rs 200 crore ($21 million) in a round led by India’s National Quantum Mission and Speciale Invest, taking its total funding to Rs 375 crore ($39 million) as it develops quantum-safe cybersecurity [Economic Times]

South Korean AI startup Upstage is considering a pre-IPO funding round as it seeks capital for its hardware-intensive AI push, after raising 560 billion won ($420 million) in its Series C this year [Bloomberg]

Abu Dhabi’s Mubadala invested in China’s Luckin Coffee as part of a $1 billion transaction with controlling shareholder Centurium Capital [FT]

Markets

Chinese regulators are tightening approval standards for humanoid-robot IPOs after Unitree Robotics’ volatile debut, requiring startups to show recurring revenue, narrowing losses or genuine technological innovation [The Information]

Investors are paying steep fees for access to special-purpose vehicles holding DeepSeek stakes, pushing the Chinese AI lab’s implied valuation to about $71 billion from $52 billion a month earlier despite five-year lock-ups and no voting rights [FT]

Moonshot AI is considering a Hong Kong IPO followed by a Shanghai Star Market listing as the Kimi maker looks to raise capital and broaden its investor base [SCMP]

AI and Chips

The US (has again) accused six Chinese AI companies including DeepSeek, Moonshot AI and Alibaba of using distillation to copy American models [Reuters]

In more detail: Anthropic itself directly accused Chinese AI labs DeepSeek, Xiaomi and Moonshot of using Claude conversations to train their own models [Anthropic]

DeepSeek released V4.1 Flash, a 552-billion-parameter model that it says is faster and cheaper to run than its previous flagship [SCMP]

Samsung Electronics opened a 6,600-square-metre advanced-chip-packaging research centre in Yokohama [Nikkei Asia]

Malaysia faces a 9-gigawatt gas-fired power shortfall by 2032 as data centres consume more electricity, with their share of demand reaching a record 9.3% in August [Reuters]

Underemployed Chinese lawyers, architects and engineers are said to be taking low-paid gig work training AI models in specialist fields as the country’s weak economy squeezes incomes [Rest of World]

China’s AI chipmakers, including Huawei and Cambricon, are raising prices as high-bandwidth-memory costs rise, with Huawei’s Ascend 950DT accelerator now quoted at more than 250,000 yuan ($37,255) [Reuters]

Crypto

Citigroup plans to offer Japanese companies instant overseas remittances using blockchain-based tokenised deposits as early as this year, including transfers outside banking hours [Nikkei Asia]

Singaporean Malone Lam pleaded guilty in the US to helping steal and launder $245 million in Bitcoin, one of the largest cryptocurrency thefts on record [BBC]

India plans to push BRICS members to use central bank digital currencies for cross-border trade settlements while opposing a bloc-wide payments network that could challenge the dollar [Bloomberg]

Policy

Tokyo’s Shinjuku ward plans to ban Airbnb-style short-term rentals in residential areas and near schools and cultural sites as complaints about tourist behaviour fuel a backlash against overtourism [The Guardian]

India’s Serious Fraud Office recommended an investigation into Xiaomi over alleged irregularities in its business model and compliance with foreign-investment laws [Reuters]

The US Treasury sanctioned Xinbi Guarantee, a Chinese-language Telegram marketplace accused of facilitating cyber scams, money laundering and human trafficking, and seized $52.8 million from 52 crypto wallets [The Record]

In other news:

Flipkart and Amazon are expanding into quick commerce, streaming video, microdramas and food delivery as their combined share of India’s monthly active users fell from 67% in 2022 to 53% in August [Economic Times]

Indeed, Flipkart launched a standalone app for its Minutes quick-commerce service and began piloting food delivery in Bengaluru ahead of October’s Big Billion Days sale [Economic Times]

At least four cyber-espionage groups, most linked to Chinese state intelligence, have exploited a previously unknown Chrome vulnerability since late August, targeting US defence contractors, NGOs and Southeast Asian governments [The Record]

Ant International, Mastercard and Visa are developing common standards to verify AI agents that can make purchases for users, creating a framework that payment networks and platforms can use to recognise trusted agents [Reuters]

Southeast Asia just got a major acquisition as Singapore’s Tazapay got scooped up by stablecoin giant Circle in a $400 million deal. But the most interesting part of the company is that it is a global business not simply one from Southeast Asia, as we explore today.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Tazapay’s move into crypto pays off handsomely

Asia has another stablecoin exit after US-listed Circle announced it will buy Singapore-based Tazapay in an all-stock deal worth $400 million. The price itself wasn’t announced, but it is mentioned in a filing—god bless public company reporting…

Circle and Tazapay are well acquainted as Circle Ventures took part in the startup’s Series B round one year ago, before leading a $36 million extension in March of this year. Tazapay raised more than $60 million in total and beyond early backers like Peak XV and Saison Capital, it also counts crypto firms Ripple and Coinbase’s VC arm on its cap table.

This deal comes just months after Payward, the parent of US crypto exchange Kraken, agreed to buy Hong Kong-headquartered stablecoin platform Reap for $600 million. While Reap helps companies bring stablecoins into their financing and operations, Tazapay comes from the opposite direction. Its payment rails handle cross-border transactions and in recent years it has added stablecoins.

Six-year-old Tazapay started out as a pure payments play. That’s unsurprisingly given CEO Rahul Shinghal is a 10-year veteran of PayPal and previously led Stripe’s revenue and growth in APAC, but that focus changed in recent years.

Tazapay became a Circle partner in 2025, enabling payments in Circle’s USDC stablecoin, and it added those aforementioned crypto investors as backers around the same time. That appears to have been transformative. Today, around 60% of its $25 billion of annualised payment volume comes from stablecoins, according to the press release announcing the deal. That’s more than double the $10 billion volume it disclosed when it announced its Series B in August 2025.

This deal ranks among the largest exits for a Southeast Asian startup (it is notably Circle’s first acquisition since its NYSE IPO in 2025), but Tazapay’s real strength and appeal is that it has built a global business that goes well beyond its home region. It claims to have more than 60 banking and fintech partners, and local payout rails that span 100-plus markets, a licence from the Monetary Authority of Singapore, and registrations with FINTRAC in Canada, AUSTRAC in Australia, and FinCEN in the United States.

This is exactly why Circle is splashing the cash (stock) to bring the business in-house. Buying Tazapay instantly gives it payment rails for USDC that it can grow immediately to help its stablecoin adoption expand beyond the typical crypto market and user base.

“Over time we expect this [acquisition] to mean more routes, wider coverage, and the ability to move money in more corridors and outside local banking hours. When those are available, we will share the news,” Shinghal wrote in a blog post for Tazapay.

So the deal isn’t selling Southeast Asian startups per se, but it is an endorsement for why Singapore is an ideal location to build a global fintech service that can be acquired by the world’s biggest companies.

Subscribe now

Paytm launches enterprise AI business

AI agents are everywhere, with even Meta launching its own today, and India’s Paytm is zoning in after it launched a dedicated enterprise business called Paytm Intelligence, or Pi. Nice alias, and the business could do decently too. Bloomberg reports that the focus is on offering AI agents for financial institutions in India and the UAE, so AI agents for sales, customer service, operations and more.

Paytm is getting back to its roots here. The company first emerged as a services business long before fintech was a thing, and upselling agentic AI is a wholly logical play given the connections it has. Barely a day goes by without a new enterprise AI business emerging, Anthropic has its own such is the opportunity. Enterprise sales are about relationships, trust and connections which Paytm should tap into here.

After losing its payment bank arm in 2024, Paytm has gone after value added services and earlier this year it forecast stronger annual revenue growth.

Deals

Indian satellite-imaging startup Pixxel raised $100 million in a round co-led by Temasek and Seraphim Space, valuing it at an estimated $400 million to $450 million [Economic Times]

Samsung led a €3 billion ($3.48 billion) investment round in French AI startup Mistral AI at a valuation of more than $24 billion [WSJ]

Singapore’s Circles Group is spinning off its telecom-operator software business and could raise several hundred million dollars from a strategic or financial investor [Bloomberg]

Markets

In an effort to boost its IPO pipeline, Thailand’s stock exchange will lower the minimum market capitalisation for new-economy listings, including electronics and robotics, to 5 billion baht ($151.9 million) from 7.5 billion baht ($227.9 million) [Nikkei Asia]

Chinese memory-chip maker Longsys raised HK$7.08 billion ($903 million) in its Hong Kong IPO, but opened flat after pricing the shares at a 44% discount to its Shenzhen-listed stock [Bloomberg]

AI

OpenAI signed a multiyear deal with Australian firm Firmus for capacity at two Malaysian data centres [Reuters]

The Philippines unveiled a seven-year $34.4 billion plan to lift AI data-centre capacity from 50 MW to 1.5 GW, with $21 billion expected from private investors [Nikkei Asia]

ByteDance is developing a real-time spatial video-generation model that could be used in robotics and autonomous systems, with founder Zhang Yiming personally overseeing a potential launch next month [Bloomberg]

China is pushing computing capacity into its north and northwest, with Inner Mongolia’s Ulanqab emerging as a data-centre hub and more than half the national pipeline now outside the crowded east [Bloomberg]

DeepSeek plans to hire about 150 senior backend engineers to rebuild systems strained by its rapid user growth and compute-heavy AI agents [SCMP]

Moonshot AI, MiniMax and other Chinese AI firms are selling subscriptions through Alibaba’s new Tmall AI Space Station in a new way to find consumers [SCMP]

Researchers used advanced AI models to build a computer worm that can hijack WeChat accounts and spread across iOS and Android devices without victims opening a link or message [NYT]

Chips

Malaysia is considering Huawei chips for a 2 billion ringgit ($494 million) sovereign AI project [Bloomberg]

Huawei says its new Kirin 9050 Pro delivers 42% higher performance through its LogicFolding architecture, testing whether it can boost chip performance without lithography equipment blocked by US sanctions [SCMP]

Huawei is backing Chinese suppliers developing semiconductor equipment and components, including an advanced DUV lithography machine being tested by SMIC and Huawei [FT]

China aims to build AI computing clusters of up to 100,000 accelerator cards and targets 9,800 exaflops of capacity by 2030, alongside 3.8 trillion yuan ($532 billion) of information-infrastructure investment [SCMP]

TSMC plans to adopt ASML’s high-NA EUV lithography machines in 2030, following Samsung’s 2028 target and Intel’s earlier production plans as chipmakers race for more powerful AI processors [Nikkei Asia]

Policy

Huawei has gone on trial in Brooklyn over US accusations that it stole trade secrets and confidential information from competitors while building its global business [FT]

Belgian prosecutors detained a Chinese-Belgian former chip executive on suspicion of spying for China in a case involving gallium nitride semiconductors [Reuters]

Chinese export controls on indium are creating a supply risk for semiconductor makers, whose high-speed data-transmission products depend on indium phosphide substrates [Bloomberg]

In other news:

India’s China Plus One gains remain concentrated in a handful of sectors and regions despite booming iPhone production, limiting the broader employment impact of its manufacturing push [India Dispatch]

Logitech turned China into its fastest-growing market by building a dedicated local operation that designs and markets products for Chinese consumers before taking successful ideas global [Semafor]

Japan’s Terra Drone is developing autonomous interceptor drones designed to destroy low-cost attack drones without using multimillion-dollar missiles [WSJ]

Southeast Asia’s content-commerce GMV is expected to reach $77.9 billion in 2026, up from $49.7 billion last year as live and short-video shopping become core ecommerce channels [TechNode]

South Korean battery-materials maker EcoPro is developing high-nickel cathodes and solid-state batteries for humanoid robots, with commercial production of its sulphide-based solid electrolytes planned for next year [Bloomberg]

China is about to get its first heavyweight AI IPO validation as Moonshot, the maker of Kimi, lines up a listing for a valuation of at least $50 billion that could raise as much as $5 billion. We’ve seen crazy memestock growth from Zhipu this year, but now Chinese AI firms will need to justify their high valuations and show real revenue growth. That will take time.

Finally, I started a new role last week with PR firm Ellerton. I updated my disclaimer, but the focus of this newsletter will continue as it did when I worked in media, venture capital and consulting. Do get in touch if you need great communications support!

Moonshot brings a new era for Chinese AI IPOs

Hot on the heels of the first earnings reports from Zhipu and MiniMax, China’s first listed AI companies, comes the news that they’ll soon be joined by Moonshot AI. The company behind the Kimi models has confidentially filed for its Hong Kong IPO, according to the Wall Street Journal.

Beijing-based Moonshot was last valued at $50 billion, and that’s said to be the target for its IPO which could raise between $3 billion and $5 billion. These numbers would make it comfortably Hong Kong’s largest listing this year.

The figures are achievable. Zhipu (which runs Z.ai) is trading at a market cap of just over $63 billion with its share price up more than 7X this year to date. But Moonshot is coming into a market that isn’t just going up and up. There’s a more realistic feeling and that’s reflected in valuation.

Early backers of Zhipu have logged incredible returns, but the stock has lost more than half its value since the second half of June. Zhipu, meanwhile, is up just 5% in 2026, it currently trades at a $16 billion valuation which is nearly three-quarters down on its March 2026 highs.

The difference has always been that Moonshot wasn’t in a rush to go public. When it raised a $500 million Series C in January, it had more capital on hand than both Zhipu and MiniMax. They rang the IPO bell out of necessity, as their cash in the bank was dwindling. But Moonshot went on to raise a further $3.5 billion in July to further strengthen its balance sheet.

At the turn of the year, Moonshot was valued at a mere $4.3 billion and now it has pulled in at least $8 billion from its backers at a much higher valuation. That’s a very different proposition for public markets. Those two early AI IPOs were a real test of the waters. Nearly nine months later and Chinese AI models have gained recognition and respect, with Moonshot’s Kimi particularly becoming a known brand in global tech circles thanks to impressive releases.

Reputation of models is one of the clear gaps between Zhipu and MiniMax, thanks to its GLM model which has won plaudits outside of China in a similar way to Kimi. Zhipu has also changed its business mode to focus on cloud-based services that can reach a global audience. That wasn’t the case at the start of the year when it listed.

We don’t know what Moonshot’s financials look like, other than the company recently crossed $300 million annual recurring revenue. That’s not substantially more than Zhipu and MiniMax, which did $136 million and $117 million respectively in the first half of the year. It remains to be seen how Moonshot will justify a much higher listing price. Potentially, it has grander plans given its cash position and the stage of its business development outside of China, but I’d imagine it isn’t that different to where Zhipu is right now.

Still, this is a new phase for Chinese AI IPOs. With DeepSeek, which is valued at over $70 billion, also preparing to go public over the next twelve months, Hong Kong retail investors are going to get exposure to more mature and crucially more expensive stocks.

That will bring raised expectations. To hit higher valuations, investors are going to want to see a real jump in revenue compared to what they see right now from other Chinese AI firms. Particularly with US duo Anthropic and OpenAI headed to public markets. If Moonshot lists at $50 billion, what will it take for its valuation to appreciate significantly enough for investors to see returns? Given the fierce competition and tough pricing conditions, I don’t know whether that’s realistic at this point.

India Quotient is learning the importance of realised exits

A fund’s returns are only as good as realised profit, as one fund in India is finding out.

Quotient struck gold when it backed ShareChat, a domestic-focused social networking platform in India, with the deal generating a 77X gain on paper. That put the fund on the map and helped generate dealflow and business as a fund backing local winners at a time when numerous Chinese services were banned and Western tech imports faced pressure, as Newcomer explains:

After that spectacular launch on the back of ShareChat, India Quotient went on to raise five funds totaling $320 million. Despite being small by modern VC standards, it has developed a reputation of spotting contrarian themes that proved profitable.

However, ShareChat has undergone layoffs, down rounds and there’s been no big exit yet. That will impact Quotient’s second and potentially third fund, and overall its reputation as an investor. It’s an issue that impacts all funds, but this is a particularly acute example. Newcomer’s post is paywalled but even the preview gives a gist of a topic that is equally as applicable in Southeast Asia and other emerging markets.

Subscribe now

Deals

ByteDance secured a $29.6 billion loan, bigger than the $20 billion it first sought, giving the TikTok owner more balance-sheet firepower as it keeps pouring money into AI [Bloomberg]

Indian spacetech startup Astrobase is said to be in talks to raise $40-$50 million in a round led by Peak XV Partners, valuing the company at about $250-$300 million [Economic Times]

Indian fintech Slice raised $100 million at a valuation of about $450 million, a steep reset from its $1.4 billion peak [Moneycontrol]

India-based smart-ring maker Ultrahuman raised $70 million from investors including Qualcomm Ventures at a $365 million valuation [Entrackr]

Gaming giant Krafton plans to invest another $250 million in India over the next 3-4 years to take its total commitment past $500 million and into AI, robotics and deep tech [TechCrunch]

Markets

Philippines fintech Mynt got the greenlight for its IPO to raise $1.5 billion, becoming the first local company to use the regulator’s lower public-float rule for large issuers [DealStreetAsia]

A study shows China’s tech-sufficiency push has seen the number of IPOs on Shanghai’s Star Market this year that target technology bottlenecks has risen from 8.1% in 2022 to 20% [SCMP]

South Korea’s semiconductor exports surged 209% year-on-year to a record $46.65 billion in August, or nearly half of the country’s goods exports [CNBC]

Vietnamese electric taxi operator Green SM plans to expand into the US and Europe ahead of a targeted 2028 Hong Kong IPO [Reuters]

Shein shares fell as much as 10% on their third day of Hong Kong trading, extending the firm’s weak debut [FT]

AI and Chips

Despite being blacklisted by Washington over national security concerns, China’s Inspur has reportedly continued to gain access to some of the most advanced US AI chips through a Silicon Valley subsidiary that appears to exploit gaps in export controls [NYT]

China remains about 15 years behind in cutting-edge chipmaking equipment, according to Zeiss executive Frank Rohmund [Bloomberg]

KEPCO, South Korea’s largest utility provider, proposed that Samsung Electronics pay 20 trillion won ($15 billion) upfront for power through 2031 and SK Hynix pay 5 trillion won ($3.8 billion) [Reuters]

Meanwhile, Korea plans to merge five public power companies under KEPCO in October 2027 to accelerate investment as AI and chipmaking drive electricity demand [Nikkei Asia]

ByteDance is reportedly in preliminary talks with data-centre operators to add 5-6 gigawatts of computing capacity in Inner Mongolia over the next two years [SCMP]

ASML Japan plans to increase its workforce from 500 to 700 people by 2030 to expand support for advanced chip production from local clients [Nikkei Asia]

DeepSeek plans to deploy at least 160,000 Huawei Ascend 950DT AI chips at a new Inner Mongolia data centre [Bloomberg]

Samsung Electronics and SK Hynix are expected to add about 600,000 wafers a month to South Korea’s memory-chip capacity by 2028, while China’s CXMT is also forecast to double capacity to 600,000 wafers a month [Bloomberg]

Japan’s AI data-centre capacity is set to more than quadruple from 1.1 GW in 2025 to 4.9 GW by 2033 [Nikkei Asia]

Tata Consultancy Services and HyperVault committed 700 billion rupees ($7.4 billion) to a 1-gigawatt AI data-centre campus in Hyderabad, one of India’s largest AI infrastructure bets [Bloomberg]

Crypto

Singapore’s central bank proposed legislative changes for stablecoin issuers, moving its regulatory framework closer to launch while planning to ban interest payments and require stress tests and wind-down plans [PYMNTS]

Policy

A year after India banned real-money gaming, offshore betting platforms are still pulling in users and money while the Enforcement Directorate widens a money-laundering probe linked to Parimatch [Economic Times]

India has stalled Alipay+’s proposed link with UPI for cross-border payments over concerns about Chinese ownership, data security and storage/usage of customer information [Reuters]

The Trump administration imposed tariffs of up to 100% on drones, targeting China’s dominance in the market while trying to push more production into the US [NYT]

South Korea’s labour ministry ruled that profit-linked employee bonuses are not subject to mandatory collective bargaining [Nikkei Asia]

Vietnam plans to make parents register and supervise social-media accounts for children under 16, and bar under-13s from posting, commenting or sharing content [TechNode Global]

The US and China will hold their first official bilateral talks focused solely on AI in mid-September, with Treasury Secretary Scott Bessent leading the US delegation [Reuters]

Thailand suspended construction of 49 data centres while it drafts new rules for the power-hungry facilities, with regulations expected within a month as local resource concerns grow [Straits Times]

In other news:

India’s cycle-rickshaws are rapidly giving way to electric models as registered e-rickshaws surge from 78,700 in 2021 to 1.8 million [Bloomberg]

Uber is cutting 200 to 250 jobs in India as part of a global restructuring that will eliminate about 3,300 roles [Economic Times]

PayPal will cut about 220 jobs in India as part of the broader multiyear turnaround plan [Reuters]

A Chinese state-linked hacking group compromised executive laptops at an agricultural conference in Hainan by physically entering hotel rooms and booting the machines from USB drives, bypassing the usual phishing and network-intrusion routes [VentureBeat]

Amazon Prime Video will more than double its investment in original Japanese content over the next two years [Nikkei Asia]

Luckin Coffee is expanding quickly in New York City five years after a fraud scandal nearly derailed the company [FT]

Chinese companies are experimenting with ways to package AI usage tokens for consumers, including credit-card rewards and telecom bundles tied to services such as DeepSeek [Rest of World]

When Zhipu, aka Z.ai, and MiniMax went public back in January it was notable that neither made much money at all. Fast forward to their first earnings reports, and revenue growth has blown up thanks to a focus on selling direct to consumers right across the world. Few saw that coming, and the next six months will also be telling as the duo’s business models develop further. What’s for sure is R&D and other spending will continue to accelerate, too.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

China’s AI startups find growth but costs spiral

China’s AI pioneers are growing up with surging revenue, but at a cost. That’s the verdict from the first earnings reports for Zhipu (Z.ai) and MiniMax, the companies that went public in the world’s first AI IPOs back in January.

Zhipu has been the more impressive of the two with revenue growing by nearly 400% year-on-year to reach $136 million during the last six months. When I say the companies grew up, Zhipu is the perfect example as it switched its core money-making focus from on-premise deployment to cloud-based products.

That’s reflected in its GLM AI models, which have gained traction in recent months in particular for offering a quality experience at a fraction of the cost of Anthropic, OpenAI and others. Cloud-based revenue went from just 15% of revenue one year ago to 86.5% today. On-premise revenue actually fell as the focus on launching global products that anyone in the world can use, and crucially pay for.

It’s a similar story for MiniMax, which grew its platform and enterprise service revenue by over 700%. Those sales cover its LLM, music AI product and consumer apps around video and image generation, and they went from one-third of all MiniMax sales to nearly two-thirds.

Overall, MiniMax logged an impressive 283% annual revenue jump, coming in at $116.6 million for the last six months, slightly behind Zhipu. January, when we wrote that neither company makes much money at all, seems a long time ago.

The rapid development of AI models this year has allowed Chinese companies to reach global audiences for the first time, in a way that wasn’t possible before. Sure, Tencent launched games that had been hits outside of China and Alibaba had success with e-commerce in parts of Europe, but that’s about the extent of it. Even ByteDance had to alter its core product to find success overseas.

Zhipu and MiniMax, though, offer the same product locally and internationally. Plus, unlike the successes of TikTok, Temu or Shein, international uptake has taken months not years and much less lobbying and localisation. They haven’t reached consumer tech scale, for sure, but there seems to be plenty of growth potential remaining.

This progress, however, has come at a cost. Playing on a global stage means a big increase in spending to keep its products competitive.

MiniMax’s adjusted net loss more than doubled, from $138.7 million to $293.0 million as admin costs like staffing doubled and R&D spending surged 139%. Zhipu’s R&D bill rose a more modest 34% but its cost of revenue ballooned 635%. The narrowed its net loss to $296 million, from a $336.8 million loss a year earlier, but it remains deeply in the red.

Neither company provided financial guidance or forecasts, they instead focused on product development with a focus on enterprise/API growth, building more agentic capabilities and, of course, continued investment in models.

You’d expect that their next results bring more of the same. Spending will increase but revenue should increase at a more rapid rate, given the progress that both have made selling across hundreds of countries worldwide. It will also depend on what trends run through the industry, agentic AI wasn’t as established in January as it is today, for example.

Zhipu and MiniMax will be the first of many public AI companies. Chinese rivals Moonshot, DeepSeek and even Manus have been linked with IPOs in the coming year or so. Plus, in the US, Anthropic and OpenAI have completed initial applications to list, too.

More public companies mean more data, more potential to scrutinise growth and company development, etc. But already, these two firms are producing very unique financial results the likes of which haven’t been seen before.

Markets

Yotta Data Services plans to go public by March 2027 to raise as much as $1.5 billion to fund its AI infrastructure expansion, which includes sovereign cloud products for India. The company recently raised $150 million at a valuation of $3.9 billion. [Reuters]

Deals

Adobe acquired India-based marketing intelligence startup Rilo, bringing in the team’s technology to automate marketing workflows and help brands track their visibility in ChatGPT, Gemini and Claude [TechCrunch]

Beijing-based Tripo AI, which develops 3D-native AI models, raised about $445 million in a Series B round [FinSMEs]

AI and Chips

Kioxia’s chip plant is reshaping Kitakami in rural Japan as it adds a third facility with government backing, though the benefits have been uneven as local businesses compete with the company for workers [Bloomberg]

Hua Hong Grace Semiconductor is investing $2 billion in a new 12-inch fab in Wuxi, adding 55,000 wafers a month as China expands capacity for AI infrastructure and cuts reliance on foreign technology [SCMP]

Policy

Taiwan has intensified its crackdown on Chinese companies accused of concealing their links while recruiting chip talent and seeking sensitive technology, with previously unpublished data showing the scale of its investigations [Rest of World]

China has removed more than 5.61 million pieces of content and 49,000 accounts in a crackdown on AI-generated “slop” across platforms including Douyin, Kuaishou, RedNote and WeChat [SCMP]

In other news:

Reliance Jio is expanding JioPC beyond its broadband subscribers, allowing any Indian internet user to access a cloud-based virtual PC on an existing computer and gain the capabilities to run AI applications [TechCrunch]

JioHotstar is replacing Hotstar in the UK, Canada and Singapore in its first expansion beyond India [TechCrunch]

China-based hackers compromised Cisco routers to monitor organisations, steal credentials and enter other networks, using a campaign dubbed Fire Ant to exploit trusted infrastructure and evade detection [The Record]

Kazakhstan’s Kaspi.kz is expanding its Turkish offering through Hepsiburada, adding shopping loans now and more fintech services next year as it looks to replicate its e-commerce and financial-services model [Nikkei Asia]

Walmart-owned Flipkart is bringing short-form scripted shows to its e-commerce app as it pushes further into content-to-commerce, with Terribly Tiny Tales and Pratilipi among its production partners [Economic Times]

Today we look at how time hasn’t been kind to some of Covid’s hottest businesses. Shein is rebuilding its fast fashion business model out in the open, after it went public in Hong Kong this week in a rush. In India, two of the country’s once-fast-growing online learning startups joined forces to plot their own journey to IPO.

A quick mention for Culpium, which has moved to paid subscriptions. Long-time ATR friend Tim Culpan is one of Asia’s longest-serving and most thorough tech journalists. His newsletter will make you smarter.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Shein is rebuilding its business in public

The time to go public is not when things are up in the air, but no matter for Shein. The Chinese fast fashion company rang the bell when it listed in Hong Kong in an IPO that raised $1.74 billion and valued the business at just under $27 billion.

This was anything but a convincing debut from the company, which was once valued as high as $100 billion. There was no pop as we’ve seen with memory maker CXMT or robotics firm Unitree. Instead, Shein’s shares are trading slightly up after recovering from an initial 10% drop.

The company was the hottest property in tech just a few years ago, and an early example of how Chinese companies could go global with real impact and results. But this IPO rush tells a story of mistakes and misfortune when Shein was riding high.

The listing looks rushed, and it is. In going public, Shein has removed a $4.4 billion cash obligation to early investors that it would have faced had it not IPO’d this year. In recent years, the company had tried unsuccessfully to list in the US and then UK, but just came in before the deadline.

The company has plenty of cash on hand, more than $14 billion for that matter, but handing more than a quarter of it in redemptions wouldn’t exactly be prudent.

Typically, companies go public when they’ve long figured out their product-market fit, growth and financials, but Shein is in a unique situation of still needing to turn things around. All while it is a newly publicly traded business.

It tried for years to shed its Chinese tag, playing up its headquarters in Singapore after relocating in 2021 and splashing cash to sponsor lifestyle events, gigs and more in the US and other Western markets. But since pinning its hopes on a Hong Kong IPO, it has embraced its roots in China and even its reclusive founder, Sky Xu, has made more appearances and been quoted more than before.

Beyond the change of narrative, Shein is also shifting its business model from simply being a provider to being an enabler. It made its name as a cheap supplier of fashion items, but post pandemic consumption habits changed and the import tax loopholes it took advantage of in the US and Europe were sealed.

Bloomberg reports that executives describe this new ambition as becoming the “Amazon Web Services of the fashion industry.”

“The AWS for X” is a tortuous simile you hear across many parts of the tech industry, and in this case it appears to mean that Shein will open its supply chain and operations to fashion brands to source from China. An in-house example is coming with Everlane, the US fashion brand that Shein acquired for a discounted price of around $100 million.

The deal is being probed by the US government but, assuming it is cleared, Everlane would tap into Shein’s network of more than 7,500 designers, manufacturers and merchants to streamline its supply chain operation and lower its costs. Shein has more than 20 brands of its own so it has plenty to work on.

We’ve seen ‘China manufacturing via API’ businesses emerge before, and fail, but none have the scale, cash or ambition of Shein which means this shouldn’t be dismissed. But, a change this significant is made even tougher when it’s done as a public-facing business.

It isn’t unprecedented. PDD, which operates Shein rival Temu, built its fast fashion and goods business after it went public. Sea has grown its Shopee business since being listed. But this uncertainty explains why the market is taking a step back from Shein stock.

There are lessons here in how capital is raised, not to mention the complicated and ever-evolving tensions between the US and China. Now, at least, our analysis of Shein’s business can be firmly rooted in its numbers. There will be nowhere to hide.

Subscribe now

Two faded Covid unicorns join forces in India in $200M acquisition

During Covid times, online learning startups were among the hottest properties in India. The hype faded significantly since then but there are still some category winners. A major piece of consolidation has made that even clearer after UpGrad completed the acquisition of rival Unacademy in a $200 million all-stock deal.

That price would have been unthinkable back in pandemic times, when Unacademy was worth upwards of $3 billion. Both companies raised close to $1 billion from investors, with UpGrad’s valuation reaching a (slightly lower) peak of around $2.2 billion.

The deal was announced in March, but it took some time to get over the line. TechCrunch reports that Unacademy angel investors are cashing out following the completion. Early backers will still have done well, but the likes of SoftBank, Tiger Global, General Atlantic who came in later will be hoping that UpGrad can come good on their bet on online education in India.

Its Airlearn language learning app is showing progress and the newly acquired business isn’t burning cash as it did before. Combined, the two businesses are at north of $250 million annual revenue, with TechCrunch providing more details:

The deal comes despite Unacademy having about ₹9 billion (about $94.8 million) in the bank and annual revenue of roughly ₹4 billion (around $42.13 million), Munjal wrote. He said most of the startup’s businesses were profitable or close to profitability, and said the company had the option to continue operating independently.

Things could, of course, be worse. Take the example of Byju’s, which has effectively become worthless and dogged by lawsuits and court action between investors and its founder.

UpGrad’s goal is to reach an IPO, and combining the forces of these two is the best bet to get there.

Deals

Nvidia is investing $3.5 billion in Taiwan’s MediaTek through convertible bonds, deepening a partnership aimed at expanding its data-centre chip ecosystem [Bloomberg]

500 Global is reportedly shelving dedicated Southeast Asia funds, meaning future regional bets are likely to come from its global vehicles as the venture market keeps consolidating [DealStreetAsia]

Bengaluru-based Alteon raised $2.5 million to build autonomous aircraft that can stay airborne for more than a year by harvesting ocean winds [TechCrunch]

Magna International is investing another $35 million in Bengaluru-based battery swapping startup Yuma Energy, raising its stake above 51% [TechCrunch]

Markets

Hong Kong’s IPO market has taken off around Chinese company listings. There’s been a lot of speculation that Southeast Asian companies are looking at it for their own IPOs, we’ll get a little taste (pun intended) if and when Jollibee picks Hong Kong for a listing of its overseas assets. Obviously not tech, but might start to clear a pathway [Nikkei Asia]

Chinese memory-chipmaker Longsys is seeking to raise up to HK$6.28 billion ($801 million) in a Hong Kong IPO, extending the rush of AI supply-chain companies tapping public markets [Bloomberg]

SB Energy, the SoftBank-backed AI infrastructure platform, filed for a Nasdaq IPO as it prepares for a listing that could raise $5 billion to $7 billion [WSJ]

Related: OpenAI received warrants worth an estimated $5.5 billion in SB Energy as part of a deal to become a data-centre tenant, tying the ChatGPT maker more closely to SoftBank’s power-infrastructure ambitions [WSJ]

Shanghai-based Enflame Technology, backed by Tencent, is seeking to raise about 6.12 billion yuan ($911 million) in a STAR Market IPO, making it the last of China’s “four little dragons” of AI chipmakers to go public [Bloomberg]

Earnings

Broadly, China’s AI chipmakers are starting to benefit from Beijing’s self-reliance push, but first-half results show a widening gap between profitable leaders and loss-making rivals [SCMP]

In particular, Z.AI’s first-half revenue rose to 954 million yuan ($142 million), missing analyst estimates as China’s AI price war weighed [Bloomberg]

Huawei isn’t public, but it still releases earnings figures: it says first-half net profit fell 36% to 23.81 billion yuan ($3.54 billion), as higher costs and R&D spending outweighed a 9.6% rise in revenue [Reuters]

AI and Chips

Manus has resumed independent operations more than four months after Beijing blocked its $2 billion acquisition by Meta with its original team in place. There’s no official confirmation of which investors were involved in the deal to release it from Meta. [Manus]

Semianalysis looks at how Korea’s trillion dollar efforts to develop sovereign AI have gone, concluding that Nvidia is winning [Semianalysis]

China’s CXMT has begun small-scale production of HBM3E, putting it one generation behind Samsung, SK Hynix and Micron in high-bandwidth memory for advanced AI chips [The Information]

Sony Semiconductor Solutions is working with Saudi Aramco on an AI model that uses image sensors to predict equipment failures at industrial plants [Nikkei Asia]

Japan’s Miyagi prefecture is trying to lure SK Hynix into building an advanced memory chip plant, although the company has denied reports that it is considering the site [Nikkei Asia]

Ajinomoto and Daikin Industries are expanding in Taiwan to work more closely with AI chip customers, as demand for advanced packaging and materials surges [Nikkei Asia]

Micron unions in Taiwan are moving toward possible strike action, demanding changes to the US chipmaker’s bonus system and a larger share of profits for workers [Reuters]

Policy

Big news here with Reuters reporting that India is preparing a framework that would let AI agents make small UPI payments without requiring approval for every transaction [Reuters]

South Korea proposed a record 821 trillion won ($596.92 billion) government budget for 2027, including a sharp push to strengthen its technology sector and compete in the global AI race [Reuters]

Singapore is committing to spend S$220 million ($173 million) over three years on fintech and innovation, including support for AI and at least 1,000 internships [Bloomberg]

India will offer chip-design startups up to 50% of project costs, capped at ₹150 million ($1.7 million), under its new Semicon 2.0 programme [Bloomberg]

A Chinese court has frozen up to 2.14 billion yuan ($300 million) of assets held by Dutch chipmaker Nexperia and its equipment arm [Reuters]

France will begin imposing a penalty on ultra-fast-fashion products this week, targeting platforms including Shein, Temu and AliExpress under a law aimed at reducing fashion waste [France24]

In other news:

China is rapidly electrifying its heavy-truck fleet, with new-energy models making up nearly 30% of sales last year, far ahead of Europe’s roughly 4% share [Semafor]

Apple could lift India’s share of global iPhone production to 30-35% within five years, supported by a new ₹62,500 crore ($7 billion) mobile manufacturing incentive package [Business Standard]

Prosus-backed iFood has asked Brazil’s antitrust regulator to investigate Meituan’s Keeta over allegedly unsustainably low prices [Bloomberg]

Chinese automakers are using AI to target 18-month development cycles, but the push is drawing tougher regulatory scrutiny over safety and testing as competition squeezes margins [Bloomberg]

Chinese rocket maker Galactic Energy successfully launched its Pallas-1 reusable rocket into orbit [SCMP]

Vietnamese automaker VinFast has reportedly suspended plans to manufacture three electric vehicles in India while it reassesses costs, telling suppliers to halt work on the VF3, VF6 and VF7 [Reuters]

Today we’re looking at how Thailand and OpenAI are teaming up to help 10 health and educational startups in the country grow through a new initiative. And we have the details of a massive new round for Ajaib, a billion-dollar valued Indonesian investment app that’s similar to Robinhood in the US.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Thailand turns to OpenAI to revive its startup ecosystem

Thailand’s government enlisted an unlikely partner after it announced a partnership with OpenAI that will see the two launch an eight-week incubation programme to develop startups and bring AI growth into its economy.

Ten startups have been selected for the inaugural batch, which marks the first time OpenAI has worked with a national government in this way. The first companies selected span healthcare and education.

Accelerators are typically run by investment firms, or at least with them in close proximity, so this particular one will be quite different. OpenAI isn’t exactly opening up its vast coffers, but it is providing hands-on support:

OpenAI will provide each team with US$2,000 in API credits, one-on-one technical guidance, and access to OpenAI’s latest frontier models. Each startup will also have a dedicated mentor throughout the program. Weekly sessions will cover product design, engineering, automated testing, evaluation, responsible AI, privacy and security, cost management, growth, and fundraising.

There will be a demo day in Bangkok in November, but it isn’t clear whether the initiative will be repeated.

Speaking at the Techsauce Global Summit event on Friday, Thailand’s deputy prime minister Dr Yodchanan Wongsawat (who leads the ministry of education, science, research and innovation) suggested that the government is keen to turn this into a regular series that could eventually attract startups from overseas. A follow-up event would need more participants, and investors especially, to have meaningful impact beyond a publicity stunt.

Beyond the programme, Dr Wongsawat laid out an ambition for Thailand’s economy to tap into technology, and AI in particular, around areas such as food, education and health. Politicians often paint pretty pictures, but making into a reality is not easy.

Startup ecosystems tend to blossom organically without government stimulation, but Thailand’s startup scene has wilted in recent years.

The country counted more than half a dozen early-stage programs before Covid, with telecom operators, banks and conglomerates getting in on the act. That may be overkill or a signal of startups becoming trendy, but fast forward to today and the accelerators have disappeared and few domestic VCs are left.

We wrote earlier this year about the funding decline when looking at why Thai startups wanting to go public overseas:

Thai companies raised nearly $5 billion in 2020, a figure that placed it behind only China. But the impact of the Covid-19 pandemic and Thailand’s volatile political landscape has seen the numbers dwindle to just $408 million last year.

This initiative with OpenAI shows that the government is trying to put startups, AI and general tech on its economic map, perhaps taking cues from South Korea and others. Time will tell if will be anything more than ceremonial, and whether OpenAI will seek out similar opportunities elsewhere.

In parallel, OpenAI also announced a “ Learning Accelerator” programme in India.

The Thailand tie-up appears to be led by Sandhya Devanathan, a former Meta VP who just joined OpenAI to lead consumer growth, enterprise adoption and regulatory engagement. Based in Singapore, she is a Thai national.

Subscribe now

Ajaib raises $270 million to buck Indonesia’s funding winter

Indonesia’s startup market may be in the doldrums following fraud and the sentencing of Gojek’s founder, but that won’t stop growing startups from getting funded. Proof of that came late last week when Ajaib, a digital investing app modeled on Robinhood, closed a $270 million investment. That’s the country’s largest deal since 2022.

The investment is led by Japan’s SBI Holdings and it values Ajaib at over $1 billion. That’s all the specifics we have. Ajaib has previously been valued at over $1 billion before, and it has now raised more than $500 million.

The obvious question is how investors find a meaningful return at that valuation, and that’s one that continues to dog startup investing across Southeast Asia. Singapore has pushed hard to make its stock exchange a viable option, and SGX has an agreement with Nasdaq to enable dual listings, while Hong Kong has also been mooted as a possible option for startups in the region.

We await that elusive first listing to pave the way. Certainly, the fact that Sea and Grab, Southeast Asia’s most prominent public figures, are only tradeable in the US is indicative of the situation.

SBI, meanwhile, has been busy doing deals. In recent months, it bought a majority stake in Singapore-based crypto exchange Coinhako, was part of stablecoin platform Fasset’s recent round and backed financial instutitonal focused startup Pints AI. It seems to see this relative down market as a good time to be bold on financial platforms in Southeast Asia.

Share

Deals

China’s Didi plans to invest more than $200 million in Argentina this year as it expands into smaller cities and builds out safety technology for its Latin America push [Reuters]

Markets

Shein will debut in Hong Kong on Tuesday after failing to win Chinese regulatory approval for planned New York and London listings, giving the Singapore-headquartered retailer a more politically acceptable route to market [Reuters]

India’s market regulator approved Jio Platforms’ $3.8 billion IPO, teeing up what could be the country’s biggest listing during a sharp rebound in new offerings [Reuters]

Earnings

Meituan returned to profit in the second quarter as easing price competition allowed it to pull back on subsidies, with adjusted net profit reaching 2.5 billion yuan ($372 million) and revenue up 14.4% to 105 billion yuan [SCMP]

China’s CXMT reported first-half revenue of 150.3 billion yuan ($22.4 billion), nearly 10 times higher than a year earlier, as AI demand helps turn the memory chipmaker profitable [Bloomberg]

AI and Chips

Malaysia’s semiconductor industry chief raised his 2026 electronics export forecast to more than 900 billion ringgit ($223 billion) as AI demand runs ahead of expectations [Bloomberg]

Tencent released Hy4 Preview, a 770 billion-parameter foundation model with a 1 million-token context window that it says beats Z.AI and Moonshot AI in internal tests [Tencent]

South Korea plans to give every citizen free access to generative AI services, turning homegrown chatbots into a public utility as Seoul pushes for AI sovereignty [WSJ]

Vietnamese leaders urged Qualcomm and Samsung Electronics to expand investment in AI, semiconductors and R&D as the country tries to move further up the regional tech supply chain [Reuters]

Z.ai released GLM-5.3’s weights on Hugging Face under a new licence that requires companies with more than $10 billion in aggregate revenue to pass a security review before hosting the model commercially [The New Stack]

Beijing’s AGI Bar has become an unofficial clubhouse for China’s AI industry, with entrepreneurs using it to trade gossip, recruit talent and court investors over beers [Bloomberg]

CXMT will supply LPDDR6 memory chips for Xiaomi’s upcoming 18 Fold smartphone, marking the first mass-market use of the new DRAM standard less than a year after the Chinese chipmaker began LPDDR5 production [Reuters]

Chinese AI companies including Alibaba, MiniMax and ByteDance are leading AI video-generation benchmarks, helped by huge short-video datasets, looser copyright rules and aggressive pricing [SCMP]

AI video tools such as ByteDance’s Seedance 2.0 are replacing Chinese actors and influencers in parts of the content industry, with 89 of Douyin’s top 100 animated dramas generated by AI by May [FT]

Nvidia is pushing robots, cars and drones as the next frontier for AI chips and software, with China’s humanoid robot supply chain giving it a large physical-AI market despite wider US export controls [WSJ]

Policy

CXMT sued the Pentagon over its inclusion on a US list of companies alleged to support Beijing’s military [Reuters]

Google opened its first Pakistan office in Karachi as it pledges to support the country’s $30 billion IT export target and expand beyond advertising in the market [Nikkei Asia]

Meta and Roblox agreed to strengthen child-safety measures in the Philippines, including age checks, parental controls and faster removal of harmful content [Reuters]

Cambodia says it has inspected nearly 800 online scam sites and shut more than 600, claiming no large-scale fraud compounds remain operating as international pressure on Phnom Penh grows [Nikkei Asia]

The Trump administration is drafting new AI chip export controls that would target Chinese firms’ access to advanced chips through data centres in countries including Thailand and Singapore [The Information]

Taiwanese prosecutors searched two Unimicron Technology sites and questioned 18 people over allegations that the company falsely labelled China-made printed circuit boards as Taiwanese [Reuters]

In other news:

Chinese drone makers are developing low-cost, 3D-printed interceptors for kamikaze drones, with companies at Shenzhen’s main drone fair pitching AI-assisted accuracy and speeds above 300 km/h [FT]

This Friday we are diving into e-commerce charts that show a number of Southeast Asian countries ranking among the world’s fastest growing for e-commerce. The region’s online buying behaviour remains less mature, but also impacted by a major battle between the likes of Shopee, Lazada and TikTok Shop to grow fastest.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Fierce competition drives Southeast Asia’s e-commerce growth

It’s easy to forget how early e-commerce is in Southeast Asia, but a new report shows countries across the region are among the world’s fastest growing when it comes to mobile.

Singapore, Indonesia and Vietnam are among the world’s fastest growing markets for e-commerce app usage and downloads, according to a new report from mobile analytics company Adjust. The company found that regional growth was highest in North America with installs up 46% and session length up 26%, but it didn’t break out figures for Southeast Asia while slower growth in parts of Asia dragged the region down.

With Shopee, Lazada and TikTok Shop among the firms battling for consumer attention and wallets, Singapore ranked highest with installs up 67% and average session length increasing by 58%. Other Asian markets stood out:

• Vietnam: installs up 42%, sessions up 21%

• Indonesia: 36% and 62%

• Malaysia: 14% and 38%

• India: 37% and 42%

The report also found that markets in Southeast Asia, and Asia more broadly, are attracting more paid users via advertising and other channels such as YouTube. That reflects the fierce battle taking place in the region between those three e-commerce platforms and others.

Shopee, for example, continues to post impressive annual sales growth of nearly 50% each quarter, but the cost of sales is increasing as is its budget for logistics. Transporting goods in Southeast Asia is a complicated process, and it explains why Amazon decided to pull the plug on the region without ever expanding beyond Singapore.

But there is still plenty of progress to be made in Asia. Despite the impressive growth, actual time users spend in mobile shopping apps across Asia is lower than the global average, according to the report.

Amazon and others in Europe and North America still suck more time than their Asian peers. That could be down to the style of the shopping experience, in addition to the relative lack of maturity in Southeast Asia and beyond. But the gap continues to close.

US government investigates logistics firm over alleged chip smuggling

Southeast Asia has emerged as a cheat code for Chinese companies wanting access to US chips and hardware that’s banned on the mainland. We know some companies use facilities in the region to legally train their AI models on Nvidia-powered setups, that was the case with Kimi K3. But the US government is now cracking down on companies that are actively shipping these parts into China from Southeast Asia.

Apex Logistics, a Singapore-based transportation company, is being looked into over suspicions that it brought Super Micro servers built on Nvidia hardware onto Chinese soil. Bloomberg reports that Apex, which is owned by Switzerland-based Kuehne+Nagel, has no relationship with Super Micro, it simply handles shipments on behalf of customers. Nvidia CEO Jensen Huang has previously insisted there’s no covert shipping of the company’s products, particularly since they’re challenging to transport.

Still, if Apex is prosecuted, this would be the first time such charges have been raised over AI chip smuggling. That could open the floodgates for other investigations in the future.

Subscribe now

Deals

SoftBank is in talks to acquire a majority stake in OpenAI-backed humanoid robotics startup 1X Technologies at a valuation of about $6 billion, giving the Norwegian company fresh backing after it missed a $1 billion fundraising target last year [The Information]

Singapore-based Flashlight Capital has offered 906.6 billion won ($655 million) for a 20.6% stake in S-1, South Korea’s largest security services provider, in the first activist takeover attempt involving a Samsung affiliate [FT]

Singapore enterprise AI company FileAI secured investment from SMBC Asia Rising Fund and Singtel Innov8 to expand fileForge, its platform for automating finance, compliance, procurement and customer-onboarding work [TechNode]

Chinese brain-computer interface startup Gestala raised 570 million yuan ($84 million) across two rounds in its first six months, as China’s race to commercialise less invasive neural-interface technology heats up [Gestala]

Markets

SenseTime posted its first profit since going public in 2021, with first-half revenue up 23.4% to 2.91 billion yuan ($407 million) and profit of 617.3 million yuan ($86 million), though its core business remained loss-making on an adjusted basis [TechNode]

Shein is looking to acquire or partner with fashion brands as growth in its own label slows, leaning on its China supply chain after a Hong Kong IPO that valued the company at just over $26 billion, down from $100 billion in 2022 [Bloomberg]

AI and Chips

Nvidia resumed H200 shipments to China under a new US licensing scheme, but the business remains tiny for now, with China sales accounting for less than 1% of its $89 billion data-centre revenue in the second quarter [SCMP]

Kioxia and SanDisk plan to invest more than $31 ‌billion in Japan by 2032 to drive semiconductor technology and expand production capacity [Reuters]

Taiwan’s Powertech Technology plans to begin production at the world’s first panel-level packaging facility for AI chips as early as 2027, potentially moving ahead of TSMC on the next-generation packaging technology [Nikkei Asia]

Chinese chip suppliers may attend Semicon India 2026 as trade delegates for the first time, as India’s expanding electronics and smartphone manufacturing base draws more interest from across the border [Economic Times]

OpenAI launched ads for logged-in adult users on ChatGPT’s free and Go tiers in India, with Plus, Pro, Business, Enterprise and Education users remaining ad-free [Economic Times]

Alibaba launched its first data centres in South America, opening two facilities in Brazil to provide cloud infrastructure and agentic AI services [SCMP]

SK Hynix broke ground on a more than $4 billion advanced packaging and research facility in Indiana, expanding its US footprint as AI-driven memory demand keeps rising [Reuters]

Policy

Google is hiring a senior partnerships manager in Seoul as it prepares to launch full-functioning Google Maps in South Korea, including turn-by-turn directions [The Korea Herald]

South Korea’s media regulator said Meta’s youth-protection changes should apply worldwide [Reuters]

Private credit is on the agenda today. Granite Asia cut the ribbon on its Libra Hybrid fund that continues to diversify its focus beyond VC deals. Then Switzerland-based ResponsAbility Investments announced an Asia fund that comes in at $461 million. Today’s investment landscape is more varied than traditional equity deals as we explain.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Granite, formerly GGV Asia, closes $500 million private credit fund

Southeast Asia has been challenging for venture capital investors in the last few years with limited exits and few realised returns, factors that make raising new funds more difficult than ever. Alternative options are increasing among funds, with private credit one particularly popular route. This week Granite Asia, formerly the Asia-based unit of GGV Capital, launched its first private credit fund as an independent entity.

Its Libra Hybrid fund is $500 million and it includes commitments from sovereign-linked funds in Singapore (Temasek’s Aranda private credit unit), Malaysia (Khazanah) and Indonesia (Indonesian Investment Authority), alongside Singapore’s DBS and others.

The fund completed a first close last year, and during that time it has completed eight deals with two exits, the firm said. Its focus has spanned healthcare, consumer and advanced manufacturing.

Private credit funds are interesting for deals since they typically provide loans or debt financing as an alternative to equity investments. So founders don’t have to trade significant equity for capital, and that particularly suits growth businesses with steady revenue. Investors can anticipate steadier returns. This isn’t replacing the core VC model for Granite, and it is part of the fund’s expansion into different investment areas since it split from GGV more than two years ago.

Back in February of this year, it partnered with DBS to launch a dedicated IPO fund of $110 million to back AI companies wanting to go public. The fund is distributed exclusively to the Singaporean bank’s wealth clients.

Climate fund ResponsAbility Investments closes $461 million Asia fund

We’re in the mood for private credit today, not only did Granite Asia’s Libra Hybrid fund close but climate-focused responsAbility Investments (yes, they don’t capitalise that R) announced the closure of a $461 million fund for Asia.

The Switzerland-headquartered fund said it has already committed just over $200 million of the fund to deals with 17 companies across renewable energy, electric mobility, energy efficiency and circular economy verticals. World Bank’s IFC is among the backers of the fund, which was expected to reach $500 million.

ResponsAbility Investments has more than $6 billion in assets across 70 countries, but it has identified Asia as a key opportunity.

“As economies expand, urbanise and electrify, demand for renewable power, distributed energy, electric mobility and efficiency solutions is expected to increase substantially. At the same time, much of the region remains undercapitalised relative to the scale of its infrastructure and transition needs, creating a significant opportunity set for experienced private market investors,” the firm wrote.

Its Asia deals this year have included Battery Smart in India, green lending in Sri Lanka, Singapore’s Skye Renewables and an India-based energy-efficient wastewater treatment company.

Climate-focused investing is on the up in Asia. We wrote in June about Singapore-based climate fund 100x100 which takes a different approach by essentially venture building new startups focused on the environment. The firm, which spun out of tech VC Wavemaker, closed a $100 million fund.

Subscribe now

Deals

Indian drone startup Airbound raised $37 million in Series A funding, taking total capital to nearly $50 million [TechCrunch]

South Korean AI platform Wrtn Technologies raised about 100 billion won ($72 million) in Series C funding at a 1.2 trillion won ($870 million) valuation to fund its global expansion [Bloomberg]

Indian voice-AI startup Ringg raised $10 million from Peak XV, taking its Series A to $15.5 million as it expands AI agents that automate customer calls [TechCrunch]

DeepSeek is nearing a $7.4 billion funding round at a roughly $74 billion valuation, ahead of a potential Shanghai STAR Market listing [SCMP]

DeepSeek generated 475 million yuan ($70.7 million) in revenue in the first seven months of 2026, around 10 times its full-year 2025 total, while recording a 715 million yuan loss [The Information]

Indian startup Runable raised $21 million in Series A funding at a $65 million valuation to build tools that help businesses find customers and grow as AI lowers the cost of making software [TechCrunch]

Markets

Indonesia plans to invest up to $28 billion in international financial centres in Jakarta and Bali as it tries to compete for global capital with Singapore and Hong Kong [Nikkei Asia]

SK Hynix’s union rejected a tentative wage deal that included pay rises and share-based bonuses by the narrowest of margins, sending both sides back to negotiations [Bloomberg]

What goes up, often comes down: shares in Unitree fell 45% after surging more than 5x on their Shanghai debut [Reuters]

Alibaba founder Jack Ma bought more than HK$600 million ($77 million) of the company’s Hong Kong-listed shares, while chairman Joe Tsai and chief executive Eddie Wu also added stock [SCMP]

AI and Chips

Z.AI, formerly Zhipu, has been revealed to be behind Ox Alpha, the free AI model that has risen sharply in usage charts [Bloomberg]

China’s supply of advanced chips is projected to grow 46% a year through 2035, although equipment bottlenecks mean domestic foundries will still fall short of demand [SCMP]

Indian AI infrastructure company AM Intelligence ordered 9,000 Nvidia Vera Rubin systems, which it plans to bring online in southern India next year for cloud providers, AI labs and local-model developers [Bloomberg]

Xiaomi is expanding its in-house chip portfolio for smartphones and cars, with its 3nm Xring O3 processor due in September and new AI and autonomous-driving chips expected next year [SCMP]

AI startup Rogo Technologies is making Singapore its Asia-Pacific hub and hiring across Tokyo and Sydney as it automates tasks for investment bankers [The Straits Times]

Huawei proposed building AI data centres in Egypt using more than 2,000 Ascend chips for government and military uses, putting the plan at the centre of a growing US-China technology contest [Bloomberg]

Vietnam’s FPT began building a 10.5-hectare education complex focused on semiconductors, drones and advanced technology, with the first phase due to open in 2028 [TechNode]

Moonshot AI is negotiating revenue-sharing deals with Microsoft, Amazon and Google to host its Kimi K3 model, potentially marking a rare major pact between a Chinese AI firm and US cloud providers [Reuters]

MiniMax reported first-half revenue up 283% to $116.6 million, although its top-line AI model releases continue to trail rivals including Z.AI [MiniMax]

Zhipu launched GLM-5.3-Flash, its first natively multimodal GLM-5 model, with 320 billion parameters and 18 billion active parameters [Z.ai]

Crypto

Visa joined Singapore’s BLOOM initiative, led by the Monetary Authority of Singapore, to test whether regulated dollar- and euro-backed stablecoins can support settlement every day of the week [PRNewswire]

Vietnam will require crypto-asset service providers to report suspicious transactions to the State Bank under new anti-money-laundering rules [TechNode]

Japan plans to develop blockchain infrastructure for 24/7 settlement of stocks and government bonds, with a development plan possible by early 2027 [Nikkei Asia]

Policy

Malaysia is fast-tracking an e-commerce law to tighten consumer protection, regulate digital platforms and give small businesses a fairer position in cross-border trade [Bloomberg]

Google, OpenAI, Meta and TikTok joined Japanese authorities and businesses in a new anti-scam alliance aimed at criminal networks linked to Southeast Asia [Nikkei Asia]

The US said it disrupted a China-linked hacking operation that breached the Justice Department, NASA, the Federal Reserve and the Senate, seizing domains tied to two hacking platforms [Reuters]

We don’t often see Chinese technology adoption from US companies but HP signed a multi-year agreement to license Wi-Fi patents from Huawei [CNBC]

It feels like a long time ago when Shein was valued at $100 billion, and now more than four years later it is finally set to go public at a much discounted valuation of less than $30 billion. But an exit is better than no exit. We also have details of a Japanese startup that’s helping make fusion power part of the electricity grid.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

For news as it happens, follow ATR Bulletin, our real-time news site that keeps you in the know.

Subscribe now

Shein finally finds a stock market home in Hong Kong

Shein’s IPO in Hong Kong looks good to go and the company will begin trading on September 1 at a valuation of up to $27 billion.

The listing is set to raise as much as 13.9 billion HKD ($1.8 billion), according to a stock market filing that shows the company will offer 280 million shares priced between 47.60-49.50 HKD. It’s been reported that Shein has dabbled with including cash payouts and additional shares for investors who previously backed it at a higher valuation.

This IPO valuation is a big deal for a fashion business, but significantly below the $100 billion that Shein was valued at when it last raised venture capital in April 2022. It puts it some way from the world’s top fashion and apparel companies, which include LVMH ($266 billion), Hermès ($167 billion) and Uniqlo ($146 billion).

Shein had a tough time trying to go public in the US and then the UK in recent years as regulators pushed back. It was locked out of India, too, but it returned in 2025 after a five year absence through a partnership with conglomerate Reliance.

The firm’s trajectory was hit hardest when the US removed the ‘de minimis’ exception for tax on items valued less than $800, and the EU followed. These changes made importing to these key regions significantly more expensive for Shein and rivals like Temu, Alibaba and TikTok, and that continues to impact its financials.

“Shein’s IPO prospectus shows it swung to a loss of $99 million in the first quarter of 2026 from a $395 million profit a year earlier, while revenue has also been declining,” Bloomberg reported.

There might yet be more intervention as US authorities are reviewing Shein’s $80 million acquisition of Everlane, the US-based sustainable online fashion brand, which was announced earlier this year. Shein voluntarily initiated the review, according to another Bloomberg report. That is typically done before a deal closes, but there’s no indication that the deal will be undone at this point.

Shein’s investors may not be getting their huge payday, but they will have a partial exit opportunity. Shein, as a public company, will be able to use public markets to grow its business and value. Hong Kong has had plenty of exposure to AI investments and Shein might be an option for those seeking diversification.

Related: Shein rival PDD, which operates Temu, said it missed its earnings target due to tough competition in China and global regulatory pressure. What a time to go public, indeed.

Kyoto Fusioneering raises $162 million to bring fusion power to the grid

We don’t often hear about deep tech companies growing from Asia but Japan’s Kyoto Fusioneering is just that. It announced it raised $162 million across a mix of equity, credit and loan facilities to scale its infrastructure business to enable fusion power to become mainstream.

Founded in Japan, as the name suggests, Kyoto Fusioneering develops equipment for commercial fusion power plants. That’s not to be confused with nuclear power, which uses fission, or splitting atoms. Fusion power instead combines small atoms into large atoms, with fewer environmental concerns like radioactive waste or meltdown risk.

Fusion is, however, experimental, but Kyoto Fusioneering is taking a strong position in being a key enabler when it becomes ready for the power grid. The company provides infrastructure that includes plasma-heating, fuel-cycle and heat-management systems, and its operations span Japan, Germany, the UK, Canada and the US. TechCrunch recently looked at Unity-3, the newest prototype it is developing to test breeding blanket technology used to generate tritium fuel and validate data for fusion reactor developers.

Kyoto Fusioneering has raised 32.98 billion JPY ($207 million) to date from the likes of the Development Bank of Japan, East Japan Railway Company, MM Japan Korea-Japan Fusion Fund and venture capital fund Coral Capital. Kiyoshi Seko, who is President and COO of Kyoto Fusioneering, actually worked at Coral and was part of the deal before moving over to the company so there’s plenty of startup DNA in the business.

Subscribe now

Deals

Xpeng’s robotics unit raised more than $900 million in its first funding round, valuing the business at over $6.3 billion as the Chinese EV maker pushes deeper into humanoid robots [Reuters]

Vietnamese hospitality startup M Village, now rebranded as Modern Village Lifestyle, has raised $26 million from investors including Mizuho Asia Partners and Trip.com [DealStreetAsia]

Markets

Singapore and Thailand are loosening listing rules and courting technology companies as the AI boom gives Southeast Asian exchanges a chance to reverse years of weak tech IPO activity [Nikkei Asia]

AI and Chips

August continued to be a busy month for Alibaba as Alibaba Cloud launched Wan3.0, a video-generation model that creates clips of up to 30 seconds from text and document inputs [TechNode]

ByteDance is merging its Trae coding and Coze agent-building teams into its Doubao chatbot and personal assistant app to battle rivals such as Tencent [Bloomberg]

Nexperia’s China unit plans to shift its entire product line to a domestic 12-inch silicon wafer foundry as it pursues independence from overseas supply chains [SCMP]

Xiaomi launched its Xring O3 processor for flagship devices, extending its push into in-house chip design and challenging Qualcomm and MediaTek in the Android premium market [Bloomberg]

Crypto

India will next month test tokenised corporate bonds, using blockchain technology to enable instant settlement. State-owned power financier REC will be the first issuer, putting India alongside markets including Europe and Hong Kong that are using blockchain for bond issuance and settlement [Reuters]

Policy

Taiwan prosecutors indicted nine people, including employees of Nvidia and Super Micro, over the alleged illegal export of AI servers to China [Reuters]

In other news:

Chinese state-affiliated hackers have more than doubled attacks after adopting AI models such as DeepSeek to automate tasks and develop malware, Taiwanese cybersecurity firm TeamT5 said [Bloomberg]

A 600-megawatt generation failure in Kyrgyzstan on August 14 triggered outages across Kazakhstan, Uzbekistan and Tajikistan, testing Central Asia’s effort to integrate power grids as data-centre demand rises [Nikkei Asia]

KakaoTalk is one of Korea’s most successful tech products. It has beaten WhatsApp and iMessage in the country and even has decent usage in overseas markets. It spawned a $10 billion business with services including ride-hailing and payments, but its efforts to bring AI into its core have divided the market as we explore.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

For news as it happens, follow ATR Bulletin, our real-time news site that keeps you in the know.

Subscribe now

Investors aren’t buying Kakao’s AI Talk

Kakao is one of Korea’s most distinctive new internet companies, but its newest manoeuvre is upsetting shareholders after it planned a split to hive off its AI-focused businesses.

The company plans to hive off its messaging, mapping, advertising and commerce businesses into a new entity that will be called KakaoAI. The remaining units, banking/payments, content, mobility and investments, will run through a second entity called KakaoX.

Kakao is valued at $11.5 billion, but its shares are down by over 75% in the last five years, so this is a major proposal to modernise its value in today’s AI economy.

The split could give Kakao a cleaner structure to raise its AI spending and investment without impacting its more conservative areas of business. Given Korea’s market alignment to AI, another goal may be to make its stock more appealing.

Kakao is putting the consumer products that are best suited to growing with AI related investment or features into one basket. That includes KakaoTalk, its breakthrough and most recognisable product today, and its maps business, which beats out Google. Google Maps was cleared to operate in Korea in February after being held back on security grounds.

It plans to put the more operational heavy assets into KakaoX, but a number of these units are already listed in their own right, including KakaoBank, KakaoPay and KakaoGames. That would allow investors to take varying levels of exposure to the assets.

The proposal is subject to a shareholder vote in December, but that is far from guaranteed.

Many investors aren’t convinced by KakaoAI’s potential to grow its revenue significantly, and see the move as a risk despite the company’s assurances. Kakao forecasts overall sales will nearly double over the next five years, with the AI unit growing at a compound annual growth rate of 20%. That’s despite the AI businesses growing just 3.6% between 2023-2025.

“These mumbo-jumbo words sound nice but they are not believable, especially because they are not backed by enough evidence. The company only had a YoY sales growth of 9.4% in 1H 2026. What is going to cause the company’s total sales growth to suddenly double in the next several years? Due to AI? Many investors are not buying this story,” wrote South Korea-focused analyst Douglas Kim.

Kim wants more evidence before the vote, and others appear to have similar sentiment. The Korean market has been blessed with an AI stock boom that’s focused on the here and now, with in-demand memory firms like Samsung and SK Hynix, not big number projections. How does Kakao square this?

Meanwhile: Kakao Mobility, the company’s ride-sharing business, filed for a US IPO that could raise $1 billion. The company, which is backed by US PE fund TPG, scrapped a planned domestic listing in 2022, but there’s no timeframe or further details at this point.

China’s Bilibili plans global expansion to rival YouTube

China’s open weight AI success might be giving the country’s tech companies reason to believe they can crack international markets after video site Bilibili reloaded its global push.

Semafor reports that the company has relaunched its international app, which it shuttered a year ago, and it is now actively posting job openings for business development and country managers in markets including Brazil, Mexico, the US and the UK. The roles are focused on areas such as deals with local creators, independent studios and IP holders, as well as building the company’s presence at festivals and other industry events.

Like other Chinese media companies, Bilibili has focused on being a front door for Chinese internet users wanting to learn about international topics through its collaborations with creators like Mr Beast, who has more than 500 million YouTube subscribers. Mr Beast has nearly 9 million subscribers on Bilibili after posting more than 30 translated videos to the site. Going after non-Chinese audiences outside of its domestic market will require a very different approach, and plenty of capital.

We might hear more this week when the company’s newest Q2 earnings drop. The previous quarter saw revenue rise 7% to nearly $1.1 billion with advertising, memberships and games the top earners.

Deals

Nvidia is in early talks with South Korean AI chip startup Rebellions about a technical partnership, investment or possible acquisition, giving the US chipmaker another route into Asia’s custom AI silicon market [Bloomberg]

Japanese space startup Letara raised ¥2.6 billion ($16 million) to expand from satellite propulsion systems into larger rockets for space, defence and security markets. Headline Asia, JIC Venture Growth Investment and Incubate Fund co-led the round [TechCrunch]

Malaysia’s Johor state claims to have attracted more than 5 billion ringgit ($1.2 billion) in potential investments after a visit to Singapore, with proposed projects spanning optical technology, health tech and semiconductor manufacturing equipment [Bloomberg]

Markets

Shein is seeking to raise up to HK$13.9 billion ($1.8 billion) in a Hong Kong IPO, valuing the fast-fashion retailer at as much as $26.8 billion ahead of a planned September 1 trading debut [Bloomberg]

China’s YMTC moved closer to a 33 billion yuan ($4.9 billion) Shanghai listing after the STAR Market accepted its IPO application, as AI demand lifts the country’s memory chip sector [Reuters]

Samsung Electronics expects shareholder returns of 90 trillion won to 110 trillion won ($65 billion to $80 billion) in 2026, capping a big week for South Korean chip payouts after SK Hynix’s buyback plan [FT]

Alibaba is seeking to raise about HK$80 billion ($10 billion) through a share sale as it ramps up spending to compete for global AI leadership [Bloomberg]

AI and Chips

Nvidia plans to use its $6 billion Poolside deal to help build one of the world’s most powerful open-weight AI models, taking aim at Chinese rivals including DeepSeek and Kimi K3 as well as OpenAI and Anthropic [WSJ]

Elsewhere, Nvidia is reportedly preparing small-batch shipments of a China-focused AI chip by year-end, with local buyers already placing orders for a processor designed to work alongside GPUs and speed up chatbot responses [The Information]

Nvidia denied a report that it plans to ship a new China-focused AI chip by year-end, keeping the company’s China strategy under scrutiny as export controls reshape the market [Reuters]

Harvey, the legal AI startup backed by OpenAI, built its first in-house AI model on Moonshot AI’s Kimi K3 [SCMP]

Brazil will invest about 2.3 billion reais ($444 million) in its AI ecosystem, splitting projects between US and Chinese companies while putting more than half the funding into a Rio de Janeiro supercomputing project with Huawei and iFlytek [Reuters]

Super Micro said an independent probe found no evidence that current senior management knew of an alleged scheme to divert Nvidia-powered servers to China, following US charges against co-founder Wally Liaw and two others [Bloomberg]

Japan will provide another ¥150 billion ($944 million) to Rapidus, deepening state support for the chip startup’s attempt to challenge global foundry leaders [Bloomberg]

South Korea plans to create a Future Response Fund that channels semiconductor tax windfalls into youth support programmes and investments in AI and other growth industries [Reuters]

DeepSeek launched V4 Flash Vision Exp, a multimodal addition to its V4 family that initially runs on its paid developer platform and beat V4 Flash on six of seven benchmark tests [SiliconANGLE]

The company has begun charging off-peak API rates on weekends to drive more usage of its models [Bloomberg]

China’s three state-owned telecom operators are making billable AI tokens a growth metric as computing demand rises [SCMP]

Inner Mongolia is becoming China’s AI infrastructure hub, with nearly 100 data centres announced or under construction in Ulanqab and companies including DeepSeek, ByteDance and Alibaba investing in local compute capacity [Wired]

South Korean students are turning to semiconductor cram schools as Samsung Electronics and SK Hynix compete for chip talent, helped by record AI-driven profits and richer employee bonus schemes [FT]

Policy

Former Google engineer Linwei Ding had part of his AI trade-secrets conviction overturned after a US federal judge tossed portions of the verdict tied to allegations he benefited two Chinese companies [Reuters]

Chinese visa restrictions on Indian executives are slowing robotics and hardware startups in India, delaying R&D and business development while forcing some teams to look at Hong Kong, Taiwan and Singapore for supply-chain alternatives [Economic Times]

Suspected China-based military hackers used AI to develop malware for a cyberespionage campaign against Central Asian governments, with researchers identifying five previously undocumented strains in the SilkParasite operation [The Record]

A US Department of Energy laboratory is probing whether Chinese-made lidar sensors could create security risks if widely deployed in American vehicles [TechCrunch]

India ordered Google to shut hundreds of Firebase accounts after finding the platform was being used to impersonate major banks and run online scams [Reuters]

China ordered its biggest-ever automotive recall over retractable door handles, covering nearly 3 million Teslas and over 1.3 million vehicles from eight Chinese automakers [NYT]

A 600-megawatt generation failure in Kyrgyzstan triggered outages across Kazakhstan, Uzbekistan and Tajikistan, exposing pressure on Central Asia’s grid as the region builds more power links and energy-hungry data centres [Nikkei Asia]

In other news:

AI is accelerating fraud and other cross-border crime in Southeast Asia, cutting costs for scam networks and forcing authorities into a cat-and-mouse game as crackdowns push groups to relocate [Nikkei Asia]

China approved its first private-sector commercial trial for satellite IoT services, giving Geely-backed Geespace a two-year permit as Beijing starts opening satellite communications to private capital [SCMP]

Flipkart Minutes is delivering 1.1-1.2 million orders a day, up from roughly 400,000 in November, narrowing the gap with Swiggy Instamart while still trailing Blinkit and Zepto [TechCrunch]

China’s Tiangong Ultra humanoid robot ran 100 metres in 9.39 seconds at the World Humanoid Robot Games in Beijing, beating Usain Bolt’s men’s world-record time on paper and giving state media a ready-made robotics headline [Reuters]

We are going payments back-to-back to close out this week as we look at why Apple Pay has moved so slowly across Asia, and still not entered major markets like India, Indonesia or Thailand. QR codes and national payment systems have made Apple a late mover in countries where its brand of cashless payments might not be needed at all.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech. And check out Asia Tech Review Bulletin, a real-time news feed for Asia’s technology industry for those who are as obsessive about the news as we are.

Have a great weekend!

Apple doesn’t have it all its own way when it comes to payments in Asia

After what seems like years of waiting, Apple Pay is quickening the pace of its expansion across Asia after it launched in the Philippines this month.

Western markets are accustomed to tap-and-pay services from the likes of Apple and Google, to the point where it has gone beyond point of sale and into rewards, transport and even ID. But in Southeast Asia, India and other markets, the reality is different. QR payments picked up the cashless payment baton, even before Covid in some countries, and Apple has faced battles to introduce its payment service for numerous reasons.

Apple Pay is supported in 12 countries across Asia, including China, Japan, Korea, Singapore, Malaysia, Vietnam and now the Philippines. The latter is Apple Pay’s newest market after it went live on 4 August with four partner banks. That’s a relatively modest launch squad but indicative of the challenge in a market where QR codes and local services pushed digital payments to nearly two-thirds of all transactions. The Philippines is telling of the payment challenge.

It may be available in more than 90 countries worldwide, but conspicuously absent from the list is India, Thailand and Indonesia. That’s surprisingly given that these are three large markets where Apple is ramping up business.

India incoming

Apple appears to be edging closer in India. We wrote back in May that its efforts there had stalled simply because it is not needed by many banks. The success of UPI, India’s government backed digital payment service, has made cashless payments the norm. That makes the situation very different from say the US or UK where Apple Pay came along and established that behaviour for consumers.

From our May issue:

The proposition is different there [India] because Apple is no longer offering a premium and differentiated offering in a busy market of cards. It is trying to offer a service that banks don’t really need, or want since it involves them losing the already tight margins they take from cards. Smaller banks are more keen to join the Apple Pay party, as you’d expect given their need to be disruptive.

That really is the crux of it. Apple’s iPhones might have 10% market share in India, at best, but card and UPI access covers almost the entire population.

Even if Apple Pay comes to India in October, as is being reported in the media, it appears that it will not launch with UPI support since it needs permission and a sponsor bank. That would massively dent its potential usage and adoption.

Late to the game

The situation in Thailand bears some comparison, too, since QR code payment adoption reached mainstream consumers a number of years ago. Retailers in cities like Bangkok have moved from manual QR payment (I pay, then I show the cashier the proof) to integrating Thailand’s PromptPay system into point-of-sale and even ATMs, so I can withdraw my cash without a card.

Banks in Thailand are also a traditional bunch. That’s to say that they don’t welcome outsiders often, though their stance has loosened in recent years.

Turkeys don’t vote for Christmas, but when cashless payments seemed imminent, Thailand’s banks teamed up to launch a system in 2016 that would keep them at the centre of the universe and not impact their revenue.

Their creation, PromptPay, remains a closed system that’s open to domestic banks, but some recent exceptions carved around tourism. Thailand has built interoperability with neighbouring QR code systems and there’s a tourist app that’s run by the national tourism association and a bank. Popular apps like Alipay, WeChat and Shopee have differing levels of connectivity to allow merchant payments while Grab is due to add support this year.

That makes sense given Thailand’s economic reliance on tourists. But against that backdrop and 10 years of PromptPay, you imagine banks don’t want Apple Pay, Google Pay or another third-party to sit in between them and their customers. Currently, visitors can pay at most merchants with Apple Pay but Thai banks themselves don’t support it for their customers.

The story is similar in Indonesia, where the national QRIS cashless system has grown in adoption and super apps like Grab, Gojek and Shopee are dominant when it comes to quick and easy payments.

With iOS adoption at around 15% in Indonesia, Apple won’t move the needle on the mainstream market but it might be able to argue that it can help more affluent customers spend more. But, again, this is subject to banks ceding some control in the customer relationship and handing over some of their payment margin to Apple.

East vs West

Apple Pay’s adoption is another of the classic examples to show that what works in the West, doesn’t necessarily work in the East. Apple has played the long game with a number of markets, and it is sure to do this same as it ramps up in the Philippines, moves into India and works on entering other markets.

But the local conditions are entirely different, and moving later isn’t an advantage in countries where banks have built a position on cashless payments already. The flip side for Apple, of course, is how valuable can many of these countries be compared to the US or Europe?

Related: Indonesia expanded Kartu Kredit Indonesia, its domestic credit card network, to consumers in a push to curb the dominance of Visa and Mastercard.

Markets

Kakao Mobility has confidentially filed for a US IPO that could raise around $1 billion, four years after shelving a domestic listing [KED Global]

Samsung and SK Hynix are preparing record shareholder payouts as investors fret over AI hardware demand. Samsung is expected to announce a return plan worth more than $72 billion, while SK Hynix could deliver more than $130 billion after a $29 billion buyback [Bloomberg]

TikTok’s international revenue grew 46% to $9.1 billion in 2025 despite its US sale and looming European fines [Forbes]

AI and Chips

Chinese AI companies are reworking software as US export controls limit access to Nvidia’s most advanced chips. Domestic hardware can increasingly handle inference, but complex work such as coding still depends on Nvidia processors [SCMP]

Earnings

Alibaba’s June-quarter profit fell 75% as it ramped up AI infrastructure spending. Capital expenditure rose 75% to 67.7 billion yuan ($10 billion), while revenue grew 9% to 268.95 billion yuan [CNBC]

Crypto

Binance launched a platform that lets AI agents analyse markets and execute trades for users, putting autonomous AI directly in charge of real money [TechCrunch]

Policy

China has tightened scrutiny of aerospace and optical-material exports to Taiwan, disrupting supply chains for the island’s technology manufacturers as customs checks delay shipments [Nikkei Asia]

In other news:

T-Mobile thwarted a Chinese cyberattack in 2024 by physically cutting a network cable after tracing suspicious activity to equipment connected to its network. The intrusion was linked to the wider Salt Typhoon campaign against US telecoms companies [Bloomberg]

LandSpace has become the first Chinese private company to recover a rocket’s first stage after an orbital launch, state media reported, pushing China’s reusable-rocket effort forward [NYT]

Hangzhou has used robots as traffic police since May, part of China’s push to move robots beyond showroom demonstrations and into everyday jobs [Reuters]

We’re inundated with AI models, open weight releases and other software from US or Chinese companies, but we hear little from those in the rest of Asia. But top Indian fintech Razorpay bucked that trend by releasing a payment model that leans on its experience as the country’s top payment processor. We dive into what and why, and the timing. We also recap China’s new blockbuster IPO and the rest of the news.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

And check out Asia Tech Review Bulletin, a real-time news feed for Asia’s technology industry for those who are as obsessive about the news as we are.

Subscribe now

Razorpay launches foundational model to make India’s payments smoother

AI model development isn’t typically heard about in Asian markets outside of China, but Indian fintech Razorpay has stepped up to the mark by launching an AI foundational model for payments.

The model is called Vulcan and it is designed to enable financial institutions to detect fraud, improve payment routing, make risk assessment more accurate and improve checkout personalisation. That’s promising given the company’s position as a payment gateway which processes $180 billion in annualised TPV, total payment volume.

The main details released so far are that Vulcan can help generate 8-10% higher payment success rate, and that it can detect eight times more international fraud. Razorpay claims it has been trained on more than 3 trillion data points that include more than just transactions. For transactions alone, it learns from around 3,000 signals. These numbers all feel fairly subjective, but you can’t really release an AI model without some numbers to quantify its utility.

“Like large data models can understand text, this model understands the complex language of money movement in India,” said Harshil Mathur, Razorpay CEO and co-founder, in a launch video.

Now technically Vulcan isn’t an LLM. It’s a transformer. (Yes, there’s a difference.) But essentially Vulcan helps other companies use some of the infrastructure that Razorpay has in place, and by offering it openly, additional inputs and data can refine it further to make it more useful. That’s the goal at least.

It’s not initially spanning all areas of business. “Razorpay said it eventually intends to extend the model across more payment decisions, including authentication, routing, fraud detection and lending,” according to Analytics India Magazine.

Razorpay has done a fair amount of work around AI this year, despite the Indian ecosystem not jumping into open source and general development as we might expect. It added agentic payments to Claude and Codex, introduced agents for business partners in June and was an early partner for Sarvam, the India-built AI model. But releasing its own material, it is upping its game further.

That’s well timed given the company has reportedly filed for an IPO at a valuation that could reach $6 billion. AI is the theme of the year for the public markets, so there’s no doubt that beefing up your AI capabilities and narrative is good for business.

Elsewhere in India AI news, a startup called Murf AI launched a text-to-speech model that it claims beats products from market leaders OpenAI and ElevenLabs on performance and cost. And US AI firm Perplexity revealed its free deal with operator Airtel generated 56 million downloads in the first seven months. There’s still five more to go but this is the first quantifiable data point from the rush of free promos that AI companies rolled out in India.

Unitree keeps China’s blockbuster IPO run going

First it was CXMT, now Unitree is the newest Chinese AI stock to see its price surge right after listing.

Unitree raised 6.1 billion yuan ($905 million) when it went public on Shanghai’s STAR board for tech IPOs yesterday. The firm’s share price closed the day up more than 450%, but before a late pullback the company’s valuation reached $66 billion. A story on links between Unitree and the US army, published on the eve of its listing, didn’t suppress the appetite for its stock.

Just weeks ago, memory maker CXMT’s share price surged over 500% after it, too, listed on STAR with an IPO that raised a colossal $8.6 billion. That pump saw the company, which is the world’s fourth largest memory provider, overtake Tencent as China’s highest valued business with a market cap of nearly $590 billion.

We’ve written plenty about China’s push for tech self sufficiency and the growth of its IPO destinations, in particular the STAR board (launched in 2019) and the revamped Hong Kong Stock Exchange. These two examples show how the two are coming together.

These two companies are standouts, for sure, but there’s more to come. Robotera, a robotics startup backed by HSG among others, is reportedly planning to raise as much as $1 billion through an IPO in Hong Kong.

Deals

HSG, the venture capital firm formerly known as Sequoia China, is seeking at least $1.2 billion for its first dollar-denominated fund since splitting from its US counterpart, as it weighs a growth fund and a yuan vehicle alongside the raise [Bloomberg]

Indian HR software startup Darwinbox is attracting acquisition interest from Salesforce, Microsoft and ADP while exploring a potential IPO, giving the KKR-backed company options after its last valuation of about $1 billion [Livemint]

Indian fintech Navi has raised $100 million from Prosus in its first institutional funding round, valuing the Sachin Bansal-founded startup at about $1.3 billion after earlier seeking external capital at roughly $2 billion [TechCrunch]

Markets

SoftBank is preparing a record 1 trillion yen ($6.3 billion) retail bond sale in Japan to fund its AI push, in what would be the largest bond issue aimed at individual investors by a Japanese company [Nikkei Asia]

Shein is targeting a valuation of about $25 billion to $28 billion for its Hong Kong IPO, a steep comedown from the nearly $100 billion valuation the fast-fashion company commanded four years ago [Reuters]

SK Hynix will buy back and cancel 40 trillion won ($28.6 billion) of treasury shares and put more than half of its free cash flow between 2025 and 2027 toward shareholder returns [Reuters]

AI and Chips

Alipay has launched China’s first full-stack agentic commerce platform, giving merchants tools to turn web pages, products and workflows into AI agent-ready services [TechNode]

Alibaba released a beta version of HappyShrimp 1.0, an AI music model that can generate lyrics, vocals, melodies and arrangements from a single text prompt [Bloomberg]

South Korea has narrowed its sovereign AI contest to LG AI Research, SK Telecom and Upstage after eliminating a team led by Motif Technologies, with two winners set to be chosen next year [Bloomberg]

China is trying to reshape the data behind AI by building high-quality datasets across more than two dozen strategic industries by 2028, part of Beijing’s push to reduce reliance on English-language training data and Western systems [NYT]

CXMT’s rise past Tencent as China’s most valuable company captures the shift from the internet giants of the mobile era to AI and semiconductor champions at the centre of Beijing’s self-sufficiency drive [Bloomberg]

Pakistan launched its first state-backed data centre for critical national data, a $90 million AI-ready facility built with technology from China’s ZTE that is stirring questions about digital sovereignty [Nikkei Asia]

China has allowed small batches of Nvidia H200 chips into the mainland, with ByteDance and Tencent each receiving about 10,000 processors as Beijing balances support for national AI champions with its domestic chip agenda [FT]

Z.ai launched its GLM-5.3 model on OpenRouter at $0.60 per million input tokens and $1.92 per million output tokens, pitching it as a lower-cost option for coding, cybersecurity and long-horizon agent work [OpenRouter]

Samsung raised prices for some advanced contract chipmaking services by up to 15% for new orders as demand for AI chips, especially from China, strains foundry capacity [Reuters]

ByteDance and the Motion Picture Association agreed a framework to protect film and TV intellectual property across the Chinese company’s AI generation tools after Hollywood studios accused Seedance of enabling copyright infringement [Variety]

Policy

Indonesia expanded Kartu Kredit Indonesia, its domestic credit card network, to consumers as it tries to reduce Visa and Mastercard’s grip on card payments [Nikkei Asia]

Singapore will require WhatsApp, Telegram and other large messaging services to make it harder for unknown contacts to reach users, while major social platforms face new duties to detect and remove scam ads [CNA]

China ordered some state-linked entities to remove a government-approved version of Microsoft’s Windows 10 months ahead of schedule, accelerating Beijing’s push to reduce reliance on foreign technology [Bloomberg]

China accused the European Union of overreach in its review of JD.com’s €2.2 billion acquisition of German electronics retailer Ceconomy, blocking compliance with information requests it says amount to extraterritorial jurisdiction [Bloomberg]

Earnings

Baidu missed second-quarter revenue estimates as a 19% drop in online advertising outweighed growth in its AI cloud business, with revenue falling 4% to 31.33 billion yuan ($4.65 billion) [Reuters]

Xiaomi reported better-than-expected quarterly profit despite a memory chip shortage that has lifted component costs, with net income down 21% to 9.46 billion yuan ($1.4 billion) as its electric vehicle push keeps pressuring earnings [Bloomberg]

Pony.ai said it has a pipeline of more than 4,000 planned and potential overseas robotaxi deployments, while second-quarter revenue rose 68.8% year-on-year to $36.2 million [Reuters]

In other news:

Google plans to move production of all Pixel smartphones, smartwatches and wireless earbuds out of China next year as it accelerates efforts to reduce manufacturing exposure to the country [Nikkei Asia]

Japanese animation giant Studio Ghibli is entering a new chapter under Nippon Television ownership as Hayao Miyazaki, now in his mid-80s, continues to shape the studio’s future [Bloomberg]

TikTok is exploring peer-to-peer payments through direct messages, extending the platform’s financial-services push with a feature that would reportedly use TikTok Pay in Southeast Asia [Bloomberg]

Tencent Cloud will launch its first cloud region in Malaysia with up to three availability zones in Johor, alongside plans to train more than 1,000 people in AI and cloud computing [The Edge Malaysia]

Uber India partnered with travel platform ixigo to add train ticket booking to move beyond ride-hailing into broader intercity travel [Entrackr]

Nokia plans to cut most of its mainland China workforce and close sites by the end of the year, marking a sharp retreat after more than four decades in the country [SCMP]

Binance provided Russian authorities with data on crypto donations to Ukrainian fundraising campaigns, material later used in a terrorism financing case against a Russian IT specialist despite the exchange’s claim that it had fully exited Russia [Reuters]

It’s been a big week for Chinese open weight AI, and especially Alibaba which released a landmark new model as it crossed 3 billion downloads to date, more than any other company on earth. This week we also look at AI growth for the economies of Singapore and Malaysia, and our news includes a Unitree IPO explainer, a failed Uber merger in India and China has a new tech king.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Finally, in ATR news, say hello to Asia Tech Review Bulletin. Bulletin is a real-time news feed for Asia’s technology industry that our team of two has quietly built in recent months. If you’re as obsessive about the news as we are, you might enjoy being able to easily scroll headlines and company updates. Stay tuned for more!

Subscribe now

Qwen is the world’s most downloaded open weight AI

It was a good week for Alibaba after Qwen was crowned the world’s most-popular open-weight AI with more than 3 billion downloads of its models to date. Qwen overtook Meta, Google and its other Chinese rivals in the last six months to take the top spot but, beyond that, it also released a potential game-changer for open-weight AI.

Qwen3.8-27B is the open weight version of the newest model it introduced earlier this month (Qwen 3.8 Max). That means you can download it locally and run it, train it, etc. With Qwen3.8-27B, though, the model can run on some laptops since all its minimal requirement is just 17GB of RAM. Granted that doesn’t work for all laptops (like the MacBook Air I’m using right now) but it’s a major step forward for a frontier-level model.

And the newest Qwen very much is frontier-level. It outperforms Qwen3.7-Plus, Alibaba’s previous flagship, and benchmarks above Anthropic’s Opus 4.6 Max, which was released in February and powered a lot of early agentic AI products.

Putting what was state-of-the-art AI on anyone’s laptop for free six months late is a feat. It is obvious but worth saying that laptop-downloadable open weight models will not only become more powerful, but their footprint will get smaller and available for more devices.

Already, Qwen3.8-27B has clocked over 250,000 downloads and it is likely to help Alibaba grow that 3 billion number at a much faster rate. This success doesn’t generate revenue itself, however. Alibaba seems to have a long-term plan for AI, but it also restructured its Qwen division and has put focus on driving revenue from its AI businesses. Things are much better than they were as recently as 2024 when it underwent a painful rebuild.

The timing of the newest Qwen launch is apt since Hugging Face, which hosts open weight model downloads, found that Chinese open models are increasingly now the infrastructure layer that developers are building on. In Europe, for example, open weight models are growing in popularity as ‘sovereign’ options even though they are built in China.

The Hugging Face report on the state of AI in the first half of 2026 found that Chinese labs are setting the pace on open models, with Qwen the most dominant model ahead of compatriots DeepSeek, Z.ai, Moonshot (behind Kimi) and MiniMax. Qwen has more derivatives than any other company, nearly double that of Google in second, and its 39 million downloads per month are nearly double second place, that’s Google again on 20 million.

These are insights anyone following the market will know, but the sheer scope of Chinese open weight models is wild when visualised. As is the growth of models from AMD and Nvidia.

“Hardware vendors have realized that open models are a way to sell chips: a model optimized for your hardware and freely available is the clearest proof that the hardware works,” Hugging Face explains.

It wasn’t just Qwen in the spotlight this week.

DeepSeek launched a new team to build AI agents. The company also quietly released an updated version of its flagship model (DeepSeek-V4-Pro-0813) and, in a move that was anything but low key, raised prices by as much as 4X during peak periods. Its pricing remains competitive but this is seen as a monetisation step before an IPO.

Not to be outdone, Z.ai released GLM-5.3, an open-weights AI model it claims is its strongest coding model ever and MiniMax launched version 3 of its music generation model.

AI powers Singapore and Malaysia GDP growth, but long term questions remain

I’ve often asked how Southeast Asia is going to monetise the AI boom, and we are getting our first pieces of confirmation that it is through selling picks and shovels, aka data centres. Singapore and Malaysia last week released encouraging GDP data showing economic growth powered by AI spending.

Singapore’s Ministry of Trade and Industry reported that national GDP grew 6.1% in the first half of the year. It highlighted “robust global AI-related demand” that boosted electronics, engineering and equipment supplies.

We know Singapore is a regional (maybe even global) hub for technology, but Malaysia is also bringing in the money with its data centre play. Its central bank data showed 6% growth in the first half of 2026 thanks to 7.5% manufacturing growth and 6.6% growth for construction.

Singapore has upgraded its growth forecast for the year to 4.5-5.5% from a previous 2-4% estimate. Malaysia is looking at around 5%, which is on par with last year’s 5.2%.

Southeast Asia is well positioned to be a data centre destination for companies across China, the US and beyond, while it is also a manufacturing outpost for companies like SK Hynix. There’s sure to be revenue around those opportunities, but countries in the region still lack the infrastructure to build their own national and sovereign AI and compute strategies, as I wrote in May.

A reliance on OpenAI, Anthropic, DeepSeek or the countless others is a problem when access to models, or even pricing, can change. When we see broader adoption of AI-based services, this can impact governments, key companies and the economy at large unless there are local alternatives. There’s no easy answer, but it’s a valid question to ponder as these countries are celebrated for impressive economic growth.

It isn’t just Singapore’s economy that’s growing impressively, its stock market is also making a comeback, as The Economist notes.

Deals

We know Manus is about to be independent but reports reiterate that Tencent is set to become its largest shareholder by acquiring Meta’s stake along Hongshan Capital and ZhenFund. The deal is likely to be at the same valuation Meta paid, around $2.5 billion, and the Manus founders will have their travel restrictions lifted. [Nikkei Asia]

It has been revealed that Uber and Rapido held talks in May to merge their India ride-hailing businesses, but the deal fell apart over its proposed structure [Economic Times]

Alibaba has reportedly sold gaming unit Lingxi Games to PE firm Trustar Capital for more than $1.5 billion [Bloomberg]

Singapore-based retail commerce company Graas raised $17 million in a Series B led by LemmaTree, the investment firm founded by Temasek. It also acquired product-data platform Trustana [TechNode Global]

Mirae Asset Venture Investments India raised Rs 1,125 crore ($134 million) in the first close of its second flagship India fund, which targets early-growth startups. The goal is to raise a total of Rs 1,800 crore ($214 million), that would be nearly 5X the size of its first fund [Economic Times]

Markets

CXMT overtook Tencent as the world’s most valuable Chinese company, it reached a market value of nearly $550 billion last week, ahead of Tencent’s $510 billion [Bloomberg]

Robotics company Unitree is raising ¥6.1 billion ($904 million) in a STAR Market IPO at an implied ¥61 billion ($9.04 billion) market cap, after selling 5,215 humanoid robots last year. Here’s a good guide to the listing and Unitree’s financials. [RoboStrategy]

Japan’s SRI International and LY Corp’s Z Venture Capital are launching a $19-32 million fund to import Silicon Valley technologies by backing deep-tech startups in AI, space, defence, robotics and advanced manufacturing [Nikkei Asia]

Indian contract manufacturer Zetwerk is reportedly preparing to publicly file for an IPO that could raise up to $400 million potentially as soon as September [Bloomberg]

China’s newest listing will be chip designer Kiwimoore after it confidentially filed for a Hong Kong IPO, targeting a valuation of about $2 billion [Reuters]

Earnings

Didi returned to profit in the second quarter as the ride-hailing firm posted net income of 869 million CNY ($129 million) thanks to record transaction volumes in China and international growth [Reuters]

China’s top chipmakers SMIC and Hua Hong saw second-quarter profits surge 262% and 386% respectively, as demand for domestic AI chips accelerates amid US export controls [SCMP]

JD.com reported its first quarterly revenue decline since its 2014 listing, with second-quarter revenue falling 2.9% to 346.4 billion CNY ($51.4 billion) as it pulls back from price competition amid weaker Chinese consumer demand. Net income rose 15% to 7.1 billion CNY ($1.05 billion), beating expectations [Bloomberg]

AI and Chips

China’s YMTC entered the world’s top three NAND flash suppliers for the first time, taking 14% of global NAND bit shipments in the second quarter and edging past Japan’s Kioxia. Samsung led with 25% and SK Hynix followed with 22% [SCMP]

SK Hynix is betting $720 billion that the AI boom will keep memory demand surging, launching one of the world’s largest chipmaking expansions as it races to double production capacity and cement its lead in the high-bandwidth memory market [CNBC]

Japanese autonomous-driving startup Tier IV plans to develop open-source AI semiconductor designs for automakers, aiming to reduce reliance on foreign chip technology [Nikkei Asia]

NXP Semiconductors is expanding its assembly and test facility in Malaysia, a move expected to more than double the plant’s output when fully operational [GlobeNewswire]

Foreign investment in Japan’s semiconductor industry reached ¥6 trillion ($37 billion), boosted by government subsidies and the latest image sensor venture between Sony and TSMC [Nikkei Asia]

Apple developed a large language model specifically for China in partnership with Alibaba, marking a shift from its reliance on third-party AI models for the market [Reuters]

Crypto

Hong Kong-based stablecoin payments company RedotPay has reportedly delayed plans for a US IPO as it seeks regulatory approvals and addresses legal issues, it could now list next year at the earliest [Bloomberg]

DigitalX, a Mirae Asset affiliate that runs South Korean crypto exchange Korbit, plans to raise about 50 billion KRW ($35.3 million) through a new share issuance [The Block]

Policy

South Korea is introducing mandatory investment classes for first-time single-stock investors after a frenzy among retail traders raised concerns about speculative trading and financial losses [FT]

Also in South Korea, the government is planning an underwater data centre off the coast of Ulsan to create space for AI infrastructure at an estimated cost of 51.1 billion won ($35.4 million) [Nikkei Asia]

A former US National Science Foundation official oversaw computer science research grants while participating in Chinese government talent programs without disclosing the ties, NSF documents show [Bloomberg]

The Trump administration is imposing 100% tariffs on Chinese drones it deems a national-security risk, escalating its efforts to curb China’s dominance of the fast-growing commercial drone market [FT]

The Trump administration is also opposing Apple’s plans to source memory chips from China, as soaring AI-driven demand creates shortages and pushes up component prices [WSJ]

Meanwhile, the US is preparing to pressure dozens of countries to choose between rival American and Chinese AI blocs, warning they could be excluded from Washington’s Pax Silica coalition if they also join Beijing’s competing framework [Reuters]

In other news:

Indian workers are increasingly being hired to train the robots that AI companies hope will replace human labour [Bloomberg]

Microsoft has shut at least 15 branch offices and joint ventures in China over the past five years as it scales back its presence in the market, corporate filings and company sources show [Reuters]

And finally, here’s a mad stat: more than 53 million Chinese now work as ride-hailing drivers or food couriers, that’s up 10 million in just two years [FT]

We are in earnings season, and that means more good news for Sea as the Singapore-based firm registered record numbers for Shopee (again) and promising growth for its fintech arm. Elsewhere, Manus confirmed it will unwind its $2 billion sale to Meta, and there’s lots of other news.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Subscribe now

Another big quarter of growth for Shopee and Sea Monee

Sea is Southeast Asia’s only true internet giant based on the company’s newest earnings.

We wrote about the company’s impressive growth in May when it published Q1 results and Q2 is equally as impressive. It brought even stronger growth across the board with the exception of adjusted EBITDA.

• GAAP revenue: $7.8 billion, up 48.1% year-on-year.

• Gross profit: $3.5 billion, up 47.3%

• Net income: $458.1 million, up 10.6%

• Adjusted EBITDA: $917.2 million, up 10.6%

Its Shopee e-commerce business hit record highs for GMV, order volume and revenue. Cost of revenue, i.e. what Sea spends to make money and fund growth, increased 48.7% year-on-year but that’s down from 51.7% in the previous quarter. Costs for Shopee went up by 51.9% year-on-year due to the increased cost of logistics, Sea said. It said it is also investing in expanding its logistics capabilities.

The company’s fintech arm (Monee) looks like being the next major growth lever. It pulled in nearly double the revenue of Sea’s Garena games business ($1.4 billion vs $747 million) but with 58.9% annual growth, compared to 48.9% and 33.5% for Shopee and Garena, respectively.

It’s been a good month for Sea, but the stock is still down this year to date

Shopee continues to account for over 60% of all revenue, and it is clearly the big focus, but Monee has the scope to grow strongly.

“Only a fraction of the users across our ecosystem are using Monee’s financial products today, and credit penetration remains low across our markets,” Sea chairman and CEO Forrest Li said in a statement.

This fintech play is similar to that being pursued by Grab and Gojek. But their financial results, released earlier this month, don’t have anything like the same traction. Grab tried payments when it first went public in 2021, but it has since turned to M&A to build out a savings and financial services playbook. Sea’s focus on lending sits better since Shopee is Southeast Asia’s largest online sales platform.

Sea has used Garena cash flow to fund Shopee’s continued growth and Monee can do the same. That’s deliciously ironic since Monee will grow off the back of Shopee’s growth.

Manus confirms it will soon be independent of Meta as $2B unwinds

It’s official, Manus is going independent.

It wasn’t a well-kept secret, but the Chinese company confirmed it “will soon return to operating as an independent company” via a notice informing users that data created following its Meta acquisition will be deleted.

Meta bought the Manus business at the tail end of last year for just over $2 billion but, as is now famous, it was forced to unwind the deal after the Chinese government ruled it illegal in April.

Now that the wheels are firmly in motion, data collected during Manus’ time under Meta ownership will be removed, although users have an 11-day window to back it up. Interestingly, Manus said user data is stored in Singapore and the US, with no mention of China.

There’s no word on how Manus will be structured as an independent business. Multiple news outlets reported last month that Tencent was leading the deal to prise it from Meta, but the reports suggested it would not become a majority shareholder. Previously, the founders were said to be in the market to raise money that they’d put towards the deal.

There’s also been speculation that Manus might be headed to an IPO fairly soon. That seems likely given the appetite that retail investors have for all things AI, particularly on the Hong Kong Stock Exchange. Manus going public in the next year or two might be the most certain thing we have on the company’s future right now.

Deals

Accel closed an oversubscribed $550M India fund less than two years after its previous vehicle, part of a $3.5B global fundraising push as the firm leans into AI, consumer internet, fintech and advanced manufacturing in the country [TechCrunch]

The former architect of Alibaba’s Qwen AI model, Junyang Lin, just launched his new startup called Pragmatik Labs to build AI agents for digital and physical environments. The company is backed by Tencent, HSG and Gaorong Ventures. [Bloomberg]

Singapore-based Jungle Ventures is pushing nearly all new deals into AI and advanced manufacturing, with India now close to 70% of deployment and $200M-$300M earmarked for the country every two years [Economic Times]

Webtoon Entertainment will pay about $100M for 60% of Korean gaming studio RI Games Holdings, giving the Naver-backed comics platform more room to turn its webtoon franchises into games [Bloomberg]

Amkor Technology is exploring a sale of part of its China business at a possible $1B-$1.5B valuation, with Asian investment firms and industry buyers expected to take a look [Bloomberg]

Samsung SDI is taking full control of its $3.5B US battery venture after buying General Motors’ 49.99% stake [WSJ]

Indian electric mobility startup Yulu raised $93M, including $63M of equity led by GEF Capital Partners and $30M of debt, to expand its fleet and move deeper into e-commerce logistics, bike taxis and express parcel delivery [Economic Times]

Markets

What AI gloom? Singapore lifted its 2026 growth forecast to 4.5%-5.5% after stronger AI investment and a softer economic hit from the Middle East war helped second-quarter GDP rise 5.9% [Reuters]

Shein plans to launch its Hong Kong IPO as soon as next Wednesday at a $30B-$40B valuation, a steep reset from the $98.2B it fetched in 2022 [Reuters]

Rakuten reported its first quarterly profit in six years, swinging to 7.7B JPY ($49M) of net income as e-commerce and fintech growth pulled the Japanese group out of the red [Bloomberg]

Singapore data centre operator DayOne has confidentially filed for a US IPO that could come as soon as next quarter, with the company weighing a roughly $5B raise that would rank among the largest data centre listings [Bloomberg]

Earnings

Tencent’s second-quarter revenue rose 11% to 204.8B CNY ($30.4B), but net profit of 56B CNY ($8.3B) missed analysts’ 61.8B CNY ($9.2B) estimate as the Chinese tech giant lifted AI spending [Reuters]

Foxconn’s second-quarter net profit rose 35% to 60B TWD ($1.86B), beating forecasts as AI server demand kept lifting the Taiwanese manufacturer’s earnings [Reuters]

Tencent Music’s second-quarter revenue rose 5.8% to 8.93B CNY ($1.32B), helped by 11% growth in music services and $60M from newly acquired Ximalaya as membership revenue climbed 8.1% to $706M [Billboard]

AI and Chips

North Korea-linked Kimsuky is said to have built local AI systems to automate cyberattacks, analyse stolen data and generate more convincing phishing, raising the ceiling for state-backed hacking campaigns [Reuters]

Sony and TSMC plan to start mass-producing next-generation image sensor chips in Kumamoto from 2029 through a 60-40 joint venture expected to cost about 1T JPY ($6.3B) [Nikkei Asia]

Chinese humanoid robot makers supplied more than 97% of global shipments in the first half of 2026, with Agibot overtaking Unitree after shipping 8,400 units and taking 44% of the market [Bloomberg]

Taiwan’s Powertech Technology is investing $400M in a Singapore AI chip packaging venture with Broadcom, with AMD lined up as the first customer before mass production starts in mid-2027 [TechNode]

Singapore’s GIC and Macquarie partnered with Anthropic to build US data centres for Claude through a new entity called Theseus Infrastructure [Bloomberg]

Also in Singapore, NUS and OpenAI will give students, faculty and staff access to ChatGPT Edu and Codex, expanding a campus partnership that began with the university’s School of Computing in March [NUS]

India’s $250B IT services industry is under growing pressure from AI automation, threatening routine software and business-process work across a sector that employs about 6M people [FT]

Alibaba is testing paid demand for Qwen-powered office tools, with annual plans from 200 CNY ($29.60) to 1,499 CNY ($222) and extra video-generation credits costing up to 968 CNY ($143) [SCMP]

AI coding startup Cursor plans to open its first India office by year-end after its local user base tripled to more than 3M developers, making India its third-largest market [Moneycontrol]

Policy

Britain’s special forces used Royal Navy drones with Chinese-made camera components that secretly transmitted data to a device in China, prompting the Ministry of Defence to cut connectivity on the fleet [The Telegraph]

South Korea will launch a 5T KRW ($3.52B) semiconductor fund and provide another 5T KRW in trade finance as Seoul tries to speed up chip clusters and supplier support [Reuters]

South Korea also plans to put more than 1T KRW ($707M) of fresh capital into a new sovereign wealth fund for AI and other strategic industries in 2027, joining the global push to channel public money into high-tech sectors [Bloomberg]

India is preparing a 1.27T INR ($15B) second-phase semiconductor incentive programme for fabs, packaging, testing, supply-chain firms and local chip designers [Business Standard]

The FBI is investigating how a federal agency was swept into a years-long North Korean remote IT worker scheme that placed operatives inside major companies and organisations [Federal News Network]

In other news:

Shein is scaling back its attempt to move operations out of China after a Vietnam warehouse push failed to match the cost and speed of its Chinese supply chain [Reuters]

India’s UPI operator wants the country’s diaspora to carry the payments network overseas, linking it with systems in 15-20 markets over the next decade and testing agentic AI payments through tools like ChatGPT and Gemini [Bloomberg]

Microsoft says a financially motivated China-linked group is exploiting a critical bug in widely used security software, with its new StormEncryptor ransomware campaign starting on Aug. 2 [The Record]

Apple Pay is set to launch in India by October with Visa and Mastercard credit-card support but no UPI transactions, giving Apple a narrow first step into the country’s premium cards market [Business Standard]

VinSpace, from Vietnam’s Vingroup, signed a SpaceX launch deal for its first satellites in 2027, part of the conglomerate’s push into satellite manufacturing, operations and data services [Bloomberg]

China-linked hackers are using AI agents to conduct reconnaissance and break into targets in Taiwan, researchers say, showing how cyber operations can run with much less human input [FT]

After nearly a month in Europe, I’m back to normality in Asia so expect usual service to resume. Since we are in August, which is typically a quiet month, we plan to publish at least three times a week until things pick up. Today, we’re looking at DeepSeek’s evolution and why the numbers on investment often lie.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Subscribe now

DeepSeek, now a venture-backed company, is planning its first price hike and much more

Fresh from raising an eye-watering 50 ​billion yuan ($7.40 billion) funding round in June, DeepSeek is moving into its next phase as a company which includes a contrarian decision: raising prices.

The Chinese AI firm has made its name offering its high-quality AI model at prices that are among the lowest on the market. Simply put, DeepSeek has been the best option for getting the most bang for buck from AI whilst still getting quality output. Data released last week suggested it is the most economical model thanks to its efficiency even though others are lower priced.

Now, however, DeepSeek is planning its first major price raise.

“We plan to raise the overall pricing for DeepSeek API services in the near future, with a significant increase expected. Please plan your usage accordingly. The specific pricing plan will be subject to official notice,” it wrote to users on its website.

Plenty will read the tea leaves as to what this means, but it feels fairly simple. The company was famously self-funded when it was spun out of hedge fund High-Flyer. Having taken investment from the likes of Tencent, CATL and others, it now has to focus on growth, revenue and unit economics so undercharging for its service makes less sense. There’s still VC pressure, despite Liang Wenfeng inserting unique terms to maintain his control, and reportedly putting in $3 billion of his own money. Indeed, the company is said to be back raising yet more cash, potentially as much as $8 billion at a valuation of over $74 billion.

We don’t yet know, of course, what the new higher prices will look like. It’s not likely to be a huge hike, but perhaps one that gives it more solid financials for an IPO, the path that every Chinese AI company is racing down right now.

DeepSeek’s plans are also focused squarely on expansion. We already know it is researching its own chips, it is developing a massive data centre in Inner Mongolia and now it is also working with one of China’s top robotics companies. DeepSeek invested 140.8M CNY ($20.8M) in robot maker Unitree as part of its impending IPO. The two companies will jointly develop AI models for humanoid robots, Reuters reported. That’s another tick in the box for China’s AI ecosystem players teaming up to work together.

This coming of age isn’t just happening for DeepSeek, other Chinese models are also chasing revenue growth and sustainability.

We wrote at the end of last year that China’s AI firms don’t make much money at all, based on the pre-IPO documents for MiniMax and Zhipu which went public in Hong Kong. Fast forward more than six months, and Kimi, Alibaba’s Qwen and others are pushing hard to generate revenue growth. China’s AI models are established and used by tens of millions worldwide, so the next phase is to turn the technology into sustainable businesses.

That’s essential if these are to become public companies, and that is clearly the goal.

Subscribe now

Southeast Asia AI report numbers are misleading but there’s cause for optimism

I’ve been concerned that Southeast Asia has been left behind in the AI race, but some figures seem to suggest that might not be the case. Data from Tracxn shows the region has raised nearly $10 billion ($9.3 billion to be exact) to date.

These types of reports tend to be attention grabbing but ultimately not particularly useful as there’s often a few deals that skew the numbers. That’s the case here as the $2.8 billion Series D for Kling AI, the Kuaishou-owned video and image generator, accounts for 68% of this year’s $4.1 billion funding and a large chunk of that total funding number. That’s despite the fact it is essentially a Chinese business, albeit one that’s registered in Singapore.

That said, there were two rounds for Singapore-based companies that do show promise:

• Acrab, a Singapore-based provider of an agentic AI compute infrastructure platform, raised $130M in a Series B led by existing investors Vertex Ventures SEA & India and Vertex Growth [FinSMEs]

• Singapore’s Multiplier raised $35M in a Series B led by The General Partnership, with Ribbit Capital and Lightspeed joining, and will use the money to fund acquisitions of specialist firms [CFOtech Asia]

At the risk of restating previous issues and making obvious comments, Southeast Asia isn’t going to be a market for AI models but there’s huge potential around AI implementation. I hope to get into this topic in the coming month or two.

Deals

Moonshot AI is expected to close its latest financing round by the end of this month at a valuation of up to $50B, as it dismantles its offshore structure to clear the way for a Hong Kong IPO as early as the end of the year [SCMP]

Monolith Management, the Chinese investment firm that backed Moonshot AI, is seeking to raise $500M in a new fund [Bloomberg]

Indian electric two-wheeler startup River raised $120M in a Series C led by Elev8 Venture Partners and Claypond Capital, with existing backers Yamaha Motor, Al-Futtaim Group and Mitsui also taking part [TechCrunch]

Vast, the Alibaba and Baidu-backed Chinese 3D-model generation startup, is in talks with investors for fresh capital in a round that could value it near $2B, and has begun preliminary talks with investment banks about a Hong Kong IPO as soon as next year [WSJ]

Southeast Asia’s native AI ecosystem has raised roughly $9.3B across 261 disclosed equity rounds as of July, with Singapore dominating as the region’s largest hub. Singapore alone accounts for $9.3B across 227 rounds, while Vietnam, Malaysia, Indonesia and Thailand together total less than $40M [TechNode Global]

SK Hynix is weighing options for its Chongqing, China facility, including bringing in an investor to accelerate growth, with a potential stake sale valuing the plant at roughly $3B [Bloomberg]

Shenzhen Adtek Technology is close to raising up to 2B CNY ($296M) in a private round ahead of its planned Hong Kong IPO, with Temasek and Morgan Stanley’s private equity arm among the likely participants. The optical connectivity products maker for data centres plans to use the proceeds to fund growth [Bloomberg]

Markets

China’s Unitree Robotics is seeking to raise about 6.1B CNY ($904M) in an IPO that would make it the mainland’s first publicly traded humanoid robot maker [Bloomberg]

Moonshot AI has revamped its corporate structure and brought in state-backed investors as it seeks Beijing’s approval for a Hong Kong listing to fund its next phase of model development, though a listing is unlikely before next year [FT]

Shein plans to start gauging investor demand for its Hong Kong IPO as soon as this week, and may kick off premarketing investor meetings on Thursday [Bloomberg]

Coupang swung to a $570M net loss in the second quarter, dragged down by roughly $410M in administrative fines tied to its South Korea data breach, even as revenue climbed 4% to $8.9B [Coupang]

China’s AI hardware stocks slumped after reports the Trump administration was drafting a ban on US imports of new Chinese data centre components, with optical module makers Zhongji Innolight, Eoptolink and Suzhou TFC leading the decline and the CSI300 Telecommunication Services Index tumbling as much as 9% [Reuters]

Evoken, the Chinese startup behind AI design agent Lovart, is considering a Hong Kong IPO, the latest AI firm seeking to tap booming investor demand after the breakout success of Manus [Bloomberg]

India’s OFB Tech, which operates supply chain and financing platform OfBusiness, is considering reviving plans for an IPO that could raise as much as $800M [Bloomberg]

Samsung and SK Hynix face growing investor calls for bigger dividends or buybacks after the pair gave scant detail on capital returns while reporting AI-driven record profit [Reuters]

SoftBank Group’s quarterly profit fell 18% as higher costs and derivatives losses outweighed gains from investments held outside its Vision Funds. The Japanese tech investor posted net profit of 347.3B JPY ($2.2B) for the April-June quarter, beating forecasts, driven largely by a 1.33T JPY gain on its Intel stake after the chipmaker’s shares more than tripled on AI-driven demand [Nikkei Asia]

South Korea’s Naver posted solid second-quarter earnings despite rising AI costs, helped by steady growth in search and e-commerce, as it pushes deeper into AI through multibillion-dollar partnerships with Nvidia and Brookfield to build gigawatt-scale AI cloud infrastructure [WSJ]

Chinese AI chipmaker Cambricon reported first-half revenue up 108% to 6B CNY ($890M), with profit up 122.6% to 2.3B CNY, as Beijing’s push to replace foreign hardware fuels demand [SCMP]

Moore Threads plans to list in Hong Kong at an ‘appropriate time’ after its shares surged more than 420% since their Shanghai debut, with first-half revenue up 147% to 1.74B CNY and the net loss narrowing to 11.6M CNY ($1.72M) [Bloomberg]

AI and Chips

Samsung Electronics and SK Hynix are testing chipmaking equipment from China’s AMEC for possible use at their Chinese factories, hedging against tighter US export controls. The South Korean firms began evaluating AMEC’s etching tools about two years ago, and while no deployment decisions have been made, the trials give AMEC a rare shot at validation from top global chipmakers [Reuters]

Tencent is ramping up the overseas push for its flagship AI model Hy3, rolling it out to global developers and enterprises through its workplace agent tool WorkBuddy, design studio Tencent Design Miora and model platform TokenHub [SCMP]

ByteDance founder Zhang Yiming told employees the company will not use distillation as a shortcut to improve its AI models, even if it temporarily falls behind Chinese rivals, a stance reportedly shaped by US government scrutiny over TikTok. The remarks came at an all-hands meeting with its Seed AI team shortly after Moonshot AI’s Kimi K3 model raised US concerns over Chinese AI progress [The Information]

China’s cheap, freely available AI models are gaining traction across Africa, with developers increasingly choosing them over more powerful US alternatives in a shift that could erode Silicon Valley’s influence in one of the world’s fastest-growing tech markets [NYT]

Envision Group has commissioned the first phase of a gigawatt-scale data centre in Inner Mongolia, an unusual move by the renewable energy company into AI infrastructure. The Galaxy Campus project in Ulanqab is running an initial 120-megawatt phase with plans to scale to 2 gigawatts [Bloomberg]

Microsoft has brought its fourth India data centre region online in Hyderabad, part of its $20.5B investment to expand the country’s cloud and AI infrastructure [Reuters]

Indonesian telco Indosat Ooredoo Group, Nokia and Nvidia launched Zankore by Indosat, a full-stack AI infrastructure platform built to serve the region [Nokia]

Silicon Valley startups that supply data to OpenAI, Anthropic and the US government are also reportedly selling AI training datasets to Chinese labs, helping them build cheaper models that now rival their American clients [Forbes]

Alibaba will charge major users of its upcoming Qwen model a share of the revenue they earn, following Moonshot AI’s approach with Kimi K3 and planning a similar threshold for Qwen3.8-Max next week. It marks a shift from charging only for cloud-hosted models while most open-source versions ran free [Reuters]

Security researchers say Moonshot’s Kimi K3 tried to cheat on a test by reaching out to the internet on its own, breaking containment during evaluation [Wired]

ByteDance is in early pre-training on an AI model with up to 10 trillion parameters, three times the size of Moonshot’s Kimi K3 and comparable to Anthropic’s most advanced Mythos system [FT]

SK Hynix plans a 54T KRW ($38B) expansion of its South Korean chipmaking facilities, building new fabs in Yongin and Cheongju to meet what it called continuously growing demand for memory in the AI era [Bloomberg]

China’s AI race faces a new bottleneck as researchers warn that a looming shortage of high-quality Chinese-language training data cannot easily be solved through hardware workarounds, unlike chip constraints [SCMP]

Apple says Mac users in China can connect to Alibaba’s Qwen AI service, making the Chinese tech giant’s AI available to Apple customers in the country [Reuters]

Apple is said to have been testing memory chips from China’s CXMT in iPhones and MacBooks amid an AI-driven memory crunch, and held early talks about supplying components for China-sold devices, while seeking the White House’s blessing for the relationship [WSJ]

Crypto

Binance affiliates are suing the founders of Hong Kong-based crypto payments firm RedotPay, alleging they diverted hundreds of thousands of customers to a competing product in a scheme that caused nearly half a billion dollars in losses [Bloomberg]

Japanese stablecoin issuer JPYC raised 6B JPY ($38M) in an extended Series B to expand its financial and Web3 ecosystem and accelerate adoption of its yen-pegged token. The extension includes 1B JPY ($6.3M) from logistics firm AZ-COM Maruwa, following a 400M JPY ($2.53M) investment from Metaplanet Ventures in March [The Block]

Policy

The Trump administration is preparing to set a price floor and tariffs on polysilicon and related products, putting the material at the centre of US efforts to compete with China on AI and energy. The decision, expected later this month, aims to protect US polysilicon factories owned by Hemlock Semiconductor and Wacker Chemie from growing Chinese ambitions in the chip supply chain [Reuters]

The US is also banning exports of lithium-ion battery and tungsten waste from later this month, requiring certain battery scrap and tungsten waste to be sold domestically for a year as Washington seeks to curb reliance on China and secure critical materials for its defence industry [WSJ]

US telecom companies connected their systems to data centres in ways that exposed cybersecurity vulnerabilities, potentially enabling the Salt Typhoon hacking campaign that breached several mobile carriers two years ago, a House committee found. The companies’ networks had routine pathways to equipment and data centres possibly linked to three Chinese telecom firms barred from operating in the US, creating connections to infrastructure breached by the Chinese hacking group [Bloomberg]

People’s Daily, the official newspaper of China’s ruling Communist Party, warned of countermeasures against widening US technology curbs as Washington considers banning imports of some Chinese AI components, adding fresh strain to the two countries’ fragile trade truce [Bloomberg]

China tightened controls on drone exports to the US and imposed new sanctions on several American companies, with the commerce ministry setting out countermeasures to recent US restrictions on tech imports and an expanded ban on goods tied to alleged forced labour in Xinjiang [FT]

China launched a formal security review of products sold by US technology firm Palo Alto Networks, ramping up pressure on the company months after accusing it of harbouring links to intelligence services. The Cyberspace Administration of China said the review was necessary to protect critical infrastructure, citing national security laws [Bloomberg]

Nepal faces pressure as Washington pushes for special treatment of Starlink, with Elon Musk positioning SpaceX’s satellite internet service as an alternative to China-run networks in the country [Nikkei Asia]

India’s lower house passed legislation allowing banks and payment providers to charge fees on UPI transactions, scrapping the rule that barred merchant discount rates on the instant payment network built by the National Payments Corporation of India [Bloomberg]

Argentine grid operator CALF accused US officials of pressuring it to cancel a Huawei infrastructure project, saying an embassy official asked in July whether it could be dropped, part of Washington’s push to curb Beijing’s influence in Latin America [FT]

A key US agency is reviewing how Chinese AI firms legally access Nvidia chips overseas, including by renting computing power abroad, after breakthroughs exposed gaps in Washington’s export controls. The probe follows Moonshot AI’s Kimi K3 nearly matching top US models last month [Bloomberg]

A US federal judge temporarily blocked the Pentagon from designating Chinese biotech group WuXi AppTec as a Chinese military company, finding defence officials repeatedly misread the evidence. The injunction is an early legal victory while WuXi’s lawsuit proceeds [SCMP]

Trip.com, China’s biggest online travel platform, faces regulatory scrutiny after a watchdog concluded it abused its dominant position in hotel bookings, a market where it handles more than half the country’s reservations [FT]

A former SK Hynix employee was jailed for 18 months for leaking CMOS image sensor technology to a Chinese firm in 2022, with the Seoul High Court upholding the lower court ruling on appeal [Reuters]

In other news:

OpenAI banned a network of ChatGPT accounts tied to a Cambodia-based scam operation that used the chatbot to run fake dating profiles, forge documents and impersonate law enforcement, with signs of forced labour among the people running the scams [OpenAI]

Chinese-built spyware platform LightSpy has evolved into a more comprehensive surveillance tool now used in over 13 countries, letting paying users steal location data, audio, chat logs, camera footage and screen recordings, or wipe a target’s device entirely [Bloomberg]

Security researcher Vangelis Stykas hacked into North Korean hackers’ servers and has maintained access for nearly two years, uncovering evidence they breached hundreds of networks worldwide, a far larger scale of intrusion than previously known [Wired]

Flipkart-owned payments app Super.money is launching online shopping on its platform, letting users buy in interest-free instalments as it expands its fintech offerings. The new shop, splitStore, sells smartphones, electronics, fashion and home goods, with purchases made through peer-to-peer payments or UPI, India’s state-backed instant payments system [Bloomberg]

Grab’s newest earnings are in and while the company continues to run a tight and steady ship despite the recent fuel crisis, it is lacking in sparkle to really excite investors. That might begin to arrive in November when we see the effects of its first major push into fintech, as we explore today. Elsewhere, Alibaba and DeepSeek duke it out to be the best value AI model.

We resume normal service next week as I will be back from Europe, so look out for more regular emails from us.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Subscribe now

Grab is ramping up its fintech play to convince investors it’s really a tech company

The market is getting excited about Grab’s lean into fintech after it released its Q2 2026 financials.

The company’s share price rose 3% after it raised its full-year 2026 guidance, increasing revenue expectations by 1.4% and adjusted EBITDA by 2.8%. Grab said the “strength” of its core business is one reason, but its move to deepen its financial services business seems key.

Grab is posting decent financial results, as we noted in its Q1 results in May, but it isn’t growing like a business that gives investors fomo. Overall revenue rose by 22%, a little below the 23% growth in Q2 2025, and it recorded a $19 million operating profit, but there’s not a lot to jump around about.

Sure, Grab has steered its business through concerns around rising oil prices without taking a revenue hit. It hit growth despite incentives to partners (rides, shops, etc) and customers rising to $706 million from $546 million a year ago. Put simply, people in the region are watching their spending, CFO Peter Oey said. (His latest earnings explainer video involved a hot wings challenge, by the way.)

Grab needs to find a strong growth engine, and it is betting financial services can be it.

This isn’t a new bet. I wrote much the same in early 2021, and five years later there’s been no boom but there are flickers of promise.

Revenue from Grab’s financial services business grew by 59% year-on-year in Q2, having grown 43% in Q1. But it accounts for just $134 million of its total $997 million revenue right now.

The company has taken steps to accelerate growth. It acquired US-based investment app Stash in a cut price deal in February and then in May it consolidated Indonesia-based digital bank Supabank, once just an investment, into its business. These two entities will show up on the balance sheet in the next quarter.

Grab’s initial focus on payments with a side of banking didn’t really play out. Its digital bank plays are maturing now, but the bet is that AI-based investments or savings can propel it from just being a transportation business. That won’t happen overnight.

Qwen’s newest model bursts on the scene, but DeepSeek remains the value king

The world is a little too obsessed with finding China’s next DeepSeek or Kimi K3 moment, but the launch of Alibaba’s newest model is another moment for Chinese tech to bask in.

Qwen3.8-Max, the e-commerce giant’s newest model, is built on 2.4 trillion parameters and the company cites benchmark results that rival Fable from Anthropic and surpass Moonshot’s Kimi K3. The new Qwen’s significance goes beyond performance, since its pricing blows many rivals out of the water.

The Qwen team claims the model costs just $2/m input and $6/m, which would make it 1/3 of the price of Kimi and some 80% cheaper than Western models, which it claims it can perform alongside. The overall cost of using a model will depend on its efficiency and capabilities, as explained below, but the price drop is notable.

Alibaba’s management was said to be disappointed that Moonshot, a company that it invested in and helped secure Nvidia chips for training, could ship a model in Kimi K3 that surpassed Alibaba’s own efforts. Now it has its own player and at a competitive cost.

Is this Alibaba flexing its financial weight and sacrificing revenue for attention, usage and narrative? Maybe.

But, as mentioned, cost isn’t just about token price. One analyst firm calculated that Qwen3.8-Max is less efficient overall than DeepSeek, which tops the list as the cheapest AI model to use:

San Francisco-based Artificial Analysis estimated V4-Flash’s average cost at 3 cents per test, compared with 86 cents for Kimi K3 from Chinese rival Moonshot AI, $1.86 for OpenAI’s GPT-5.6 Sol and $3.15 for Claude Fable 5.

The ​comparison provides a more realistic measure of value than pricing alone because it accounts for the amount of ​data a model must process and generate to complete a task. A model with low headline price can still prove expensive ‌if it ⁠requires significantly more steps to produce an answer.

DeepSeek will get a lot scarier when it puts the $7 billion-plus funding round that it closed in June to work.

Deals

AI cloud company CoreWeave is entering Asia with its first data centers in the region, planning three facilities in Indonesia totaling 360 megawatts of capacity. The US firm expects the centers online by 2028 and will spend billions on the project, though it gave no specific estimate. [CoreWeave]

Uzbekistan’s TAS-1 project, backed by Saudi Arabia’s DataVolt, will deliver 6 megawatts at Tashkent’s IT Park by year-end, the first step toward 500 MW as Uzbekistan targets $1.5 billion in AI revenue by 2030. Kazakhstan’s rival effort is far bigger: the Nvidia-backed Data Center Valley in Ekibastuz aims for a full gigawatt using 100,000 next-generation chips, part of a $10 billion deal led by US firm Firebird. [Nikkei Asia]

South Korean AI chipmaker DeepX has raised fresh funding at a valuation near 3.14 trillion won ($2.2 billion), roughly four times its previous round [Bloomberg]

NTT Data plans to spend at least $9.6 billion through 2033 to quadruple its computing capacity to 1 gigawatt, adding roughly 750 megawatts over the next seven years to meet surging AI demand in Japan [Bloomberg]

Robbyant, the embodied artificial intelligence division of Ant Group, has begun seeking external funding, becoming the fourth unit of the fintech giant to step up capital-raising efforts [SCMP]

Markets

Shein is said to be considering offering cash payouts and free extra Class B shares for late-stage backers who invested at valuations up to $64 billion. The move would effectively lower those investors’ cost basis to around $40 billion, closer to the valuation Shein is targeting for its planned Hong Kong listing. [Bloomberg]

Bithumb, South Korea’s second-largest crypto exchange, confirmed it is preparing for an IPO targeted for 2028 [The Block]

Vast, the Chinese 3D-modeling startup behind Tripo Studio that’s backed by Alibaba and Baidu, is reportedly exploring a Hong Kong IPO with Bank of America and CICC advising [Bloomberg]

Chinese robot maker Unitree Technology is expected to be valued at more than 50 billion yuan ($7.4 billion) after its planned Shanghai IPO [Reuters]

Africa-focused fintech PalmPay is preparing for a potential Hong Kong IPO and is also in talks to raise about $200 million in a funding round to scale its business. The company serves roughly 40 million users, mostly in Nigeria, making it one of the largest consumer-finance and payments platforms across Africa and parts of Asia, with backers including Transsion, MediaTek and Singapore’s GIC. [Bloomberg]

AI and Chips

Huawei’s top chip scientist Liao Heng warned that Western chipmakers are nearing physical limits in the race for more powerful processors, predicting “an avalanche” once that threshold is crossed [Bloomberg]

Anthropic will let Indian organizations process Claude requests entirely on domestic servers, rolling out in-country inference through Amazon Bedrock in the coming weeks [Moneycontrol]

China’s CXMT is preparing to produce advanced sixth-generation low-power LPDDR6 memory chips in small quantities by year-end, positioning it to compete with Micron, SK Hynix and Samsung [Bloomberg]

CXMT is also considering building a second memory-chip plant in Beijing ​and is in financing talks with a tech manufacturing hub backed by the local government [Reuters]

HP, Asus and Acer have begun using small amounts of DRAM chips from China’s CXMT in some notebooks as an AI-driven memory shortage squeezes supply [Nikkei Asia]

JD.com launched an AI-powered smart helmet for food couriers, joining rivals Alibaba and Meituan in rolling out similar tech to improve rider safety and delivery efficiency [SCMP]

China’s air force is using an AI-enabled “intelligent strike planning system” to draw up battle plans for operations involving more than 100 units, a state television documentary revealed [SCMP]

Policy

India has proposed extending tax exemptions for foreign firms supplying machinery to contract manufacturers until 2041, a major win for Apple [Reuters]

Apple’s annual sales in India topped $10 billion for the first time last fiscal year, growing at a double-digit rate from about $9 billion a year earlier, with iPhones driving the bulk of revenue [Bloomberg]

India will seek parliamentary approval for tax changes aimed at boosting foreign investment and local manufacturing, according to sources [Bloomberg]

TikTok is finalizing agreements to settle three upcoming trials related to allegations that its video product is addictive and harmful to minors [Bloomberg]

The Trump administration is drafting a ban on imports of new Chinese optical transceivers, key data center components for AI infrastructure [Reuters]

India is introducing legislation that could reshape its Unified Payments Interface, potentially allowing merchants to pay charges on some UPI transactions [TechCrunch]

In other news:

An unsecured police dashboard offered a rare glimpse into how Chinese authorities monitor foreigners in the country, revealing the scale at which officials collect and aggregate private data, from travel records to personal communications, to build detailed profiles. [NYT]

Hong Kong police have flagged a $3.3 million romance scam involving fake crypto app [The Block]

Chinese autonomous-driving company WeRide is entering Denmark through a partnership with electric-mobility operator GreenMobility, marking its first move into the Nordic region [The Next Web]

Several popular Samsung smart TV apps have been found to contain code that shares owners’ internet connections with strangers, putting millions of TVs at risk of hijacking [TechCrunch]

Southeast Asian films and dramas are winning bigger audiences on streaming platforms at home and abroad [Nikkei Asia]

Today’s issue is very short as I’m still in Europe. I return to Asia next week, so expect normal service to resume then.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Subscribe now

The summer is a great time to nail down your marketing and content strategy. At Fully Drafted, we are helping a handful of companies with a free audit and plan for the rest of the year. We initially helped two startups, but enjoyed it so much that we are open to helping more. Get in touch if we can help you. More details on LinkedIn.

Nobody knows what trained breakout AI model Kimi K3

China’s Moonshot AI burst onto the global scene last month with its Kimi K3 model which grabbed headlines and attention as a cheaper but comparably capable alternative to Claude or ChatGPT. Despite the movement around China’s self sufficiency push for AI and semiconductors, however, Kimi K3 was trained using Nvidia chips.

It’s well known that China is building systems, processes and companies to avoid any reliance on the US for technology. Memory firm CXMT’s massive $8.6 billion IPO is proof it isn’t playing around. But the narrative needs some time to become reality because the country’s top AI firms are still reliant on Western kit. How they get it, however, is worth digging into.

Bloomberg reported that Kimi K3 was trained on a 20,000 Nvidia cluster that was supplied by Alibaba. That seems surprising given the export restrictions on tech from Nvidia and others, but Moonshot is one of many that skirt that restriction by using Southeast Asia. In this case, apparently, Thailand.

Bloomberg was able to confirm that Alibaba supplied Nvidia chips to Moonshot but a lot is unclear:

• “Multiple people close to Moonshot” claimed they were H200 chips, once illegal in China but now available in limited quantities

• Alibaba denied that they were H200 chips, and would only confirm that they rented some Nvidia capacity to Moonshot

• US government official Michael Kratsios claimed they were from Nvidia’s newest Blackwell family, which is banned from sale to Chinese companies

• “A person familiar with Moonshot’s procurement strategy” said the company has a channel to access Blackwell chips, but they didn’t specify if it was a rental agreement (legal) or via purchase (illegal)

It’s a messy situation, and one that the US government is clearly keen to clear up through enforcement. That’s not so easy with shell companies, multiple countries, etc.

The most interesting part might be Moonshot and Alibaba’s relationship.

Alibaba is a major investor in the startup, and it clearly helps on tech access. But its own AI models rival Kimi, and its role in enabling a competitor to outpace its own solution is said to be frustrating.

Moonshot is in the market for more Blackwell chips to train its newest Kimi model, will Alibaba still be its partner? You’d think not based on these revelations.

Deals

Tesla is reportedly considering separating or selling its China business to smooth a potential merger with SpaceX, as US-China tensions and SpaceX’s defence work raise regulatory and national-security concerns [WSJ]

Tesla denied the report, with Elon Musk calling it “absurdly fake news” [Nikkei Asia]

But SpaceX is also said to be moving to keep Chinese nationals and China-made components out of parts of its supply chain [Nikkei Asia]

Thailand’s True Internet Data Center is seeking about a $2 billion-equivalent loan to build a new facility, split across a roughly $1.8 billion dollar-denominated tranche and two Thai baht portions [Bloomberg]

Ant Group’s three spin-offs, each operating independently since 2024, are moving on separate fundraising tracks. Ant Digital Technologies is preparing a financing round ahead of a planned IPO, while Ant International just closed a $1.2 billion Series A backed by Ant Group, Alibaba and international investors. OceanBase, the database unit, is reportedly eyeing its first outside equity raise [SCMP]

Indonesia’s TelkomMetra is weighing a sale of MDI Ventures, one of the country’s biggest corporate VCs with $830 million in committed AUM. We recently wrote that the jailing of former MDI and BRI Ventures executives over a $25 million TaniHub investment brought the very idea of state-backed investment into question in Indonesia. [CNBC]

Singapore’s ADA has acquired India’s Algonomy in an undisclosed deal to expand its AI-powered retail decisioning business across 34 markets [TNGlobal]

Groww’s four cofounders are said to be setting up a INR4-5 billion ($45-57 million) second seed and early-stage fund to back consumer internet and deeptech startups, expanding the investment programme they have run for the past few years [Economic Times]

Markets

After a rout that hit many retail investors hard, SK Hynix and Samsung Electronics surged more than 20% in Seoul, helping revive South Korea’s AI-linked chip rally after this week’s brutal sell-off. The move tracked a rebound in US technology stocks and came days after heavy losses in Korean chip shares and leveraged retail bets exposed how volatile the AI trade has become [CNBC]

Samsung’s chip division posted KRW89.2 trillion ($64 billion) in June-quarter operating profit, more than 250 times last year’s level and ahead of forecasts, as AI memory shortages lifted earnings. Group net income reached KRW71.3 trillion ($51 billion), showing how far Samsung has narrowed SK Hynix’s lead in the AI memory boom [CNBC]

Zhongji Innolight fell 5.9% on its Hong Kong trading debut after raising HK$53.41 billion ($6.81 billion), giving the Chinese optical transceiver maker a weak start to the city’s largest listing since Alibaba in 2019 [WSJ]

Pakistan’s Abhi plans to list Abhi Microfinance Bank in an IPO this month, putting a fintech-owned lender on the public market after its acquisition of FINCA Microfinance Bank. The YC-backed company, founded in 2021, has expanded from earned-wage access into embedded finance across Pakistan and the Gulf [Bloomberg]

Zepto announced a pre-IPO private placement from existing investors as it prepares to list within SEBI’s approved timeline, after earlier pausing IPO plans to raise more than INR10 billion ($105 million). The quick-commerce startup is expected to seek a $4-4.2 billion valuation, below its $7 billion private valuation six months ago [Moneycontrol]

South Korea plans to inject KRW20 trillion ($14 billion) into its sovereign wealth fund for strategic investments in AI, data centres and infrastructure, expanding its mandate to domestic assets for the first time. The move follows a rout in technology stocks and gives Seoul another state-backed vehicle for AI-era capital allocation [Bloomberg]

China’s Unitree has launched a STAR Market IPO that could raise about $620 million, giving one of the country’s best-known robotics startups a domestic listing path as humanoid and quadruped robots draw heavier investor interest. The Hangzhou company sells consumer and industrial robots and has become a flagbearer for China’s embodied-AI push [Caixin]

Japan’s Kioxia approved a 3-for-1 stock split effective Oct. 1 after its April-June results missed analyst estimates despite a sharp AI-driven memory rebound [Bloomberg]

Kioxia has started shipping samples of its ninth-generation NAND flash memory chips for smartphones and PCs, with the Japanese chipmaker saying performance is more than 30% higher than the prior generation [Nikkei Asia]

Southeast Asia PE investment fell to $935.5 million across 10 deals in Q2 2026, down from $9.2 billion in the previous quarter, as capital concentrated in mid-market transactions. Exits were the stronger signal, reaching a four-year high of $4.2 billion, led by Singapore real estate and infrastructure deals [TNGlobal]

Chinese venture capital firms are seeking about $35 billion for at least 60 new dollar-denominated funds after a three-year fundraising slump, as AI breakthroughs, robotics advances and recent tech IPOs revive investor interest. HSG, IDG Capital, Matrix Partners China and Future Capital are among firms marketing or planning new funds [FT]

AI and Chips (China)

Xi Jinping called for wider defence use of unmanned and AI technologies as China pushes to build a more advanced PLA before its 100th anniversary in 2027. The directive came at a Politburo study session and paired military-civilian integration with tighter oversight after a widening anti-graft crackdown [Bloomberg]

New CTGT research found censorship in Chinese open-source AI models did not necessarily carry into models distilled from them, challenging one argument for banning DeepSeek- and Moonshot-based systems. The study tested 152 prompt pairs on China-sensitive topics, including Uyghur detention camps and Tiananmen Square, as US officials weigh the risks of cheaper Chinese models [Semafor]

China’s AI boomtowns are pulling further ahead of the wider economy, with growth concentrated in a handful of cities that have attracted model developers, data-centre projects and state support [Bloomberg]

Chinese military researchers have used outputs from OpenAI and Anthropic models to train domestic AI systems for defence uses, based on a review of more than 80 Chinese academic papers and patents [Reuters]

ByteDance is pouring TikTok cash into a full-stack AI push, with Doubao now China’s most popular AI app on 324 million monthly users and founder Zhang Yiming backing investment across models, cloud and chips [FT]

ByteDance folded Lark’s product team into the group behind Doubao and shifted its sales and service staff under Volcano Engine, as its annualised AI revenue reached $4 billion [SCMP]

MiniMax has drifted under the radar compared to its peers but it just released H3, a video-generation model aimed at commercial uses including advertising, e-commerce, gaming, product design and content creation. Its shares jumped 17.4% after the launch [Reuters]

DeepSeek opened a public beta API for its flagship AI model, giving developers a direct way to build on the Chinese startup’s latest systems [Bloomberg]

DeepSeek is planning a 1-gigawatt AI data centre in Ulanqab, Inner Mongolia, and aims to bring at least part of the capacity online by late 2027 or early 2028 [Bloomberg]

RedNote is planning a 600MW data centre in Inner Mongolia with a budget of about 15 billion CNY ($2.2 billion), excluding semiconductor costs, as it builds capacity for AI infrastructure. The Xiaohongshu operator is in talks with VNET and is targeting first-quarter 2027 delivery, while Beijing expects 4 trillion CNY of national computing-network investment over five years [SCMP]

US senators urged Apple to drop any plan to buy memory chips from blacklisted Chinese suppliers CXMT and YMTC, warning it could become reliant on a US adversary for key components [Bloomberg]

AI and Chips (Beyond China)

SoftBank and NTT are leading around 30 Japanese companies, research groups and universities in a public-private consortium to build cross-industry AI data-sharing infrastructure. The xIPF Consortium spans telecom, electronics, energy, real estate and other sectors as Japan looks for common technology to let companies share data more easily [Nikkei Asia]

India’s Sarvam is building a trillion-plus-parameter foundation model from scratch in India, aimed at coding, cybersecurity, simulation and scientific research workloads. The startup also priced its 105B model at $0.80 per 1mn blended tokens, claiming it is 5.5 times cheaper than GPT-5.4 Mini and over 11 times cheaper than Gemini 3.5 Flash [Moneycontrol]

TSMC has developed new AI chip-packaging technology to counter Intel, which is pitching its EMIB approach as an alternative for customers squeezed by advanced-packaging bottlenecks. The move keeps packaging at the centre of the AI chip race, where TSMC’s CoWoS capacity has become a constraint for Nvidia and other accelerator designers [The Information]

MediaTek has approved a discretionary $5 billion financing budget to support long-term growth, including its push into AI data-centre chips. Taiwan’s biggest chip designer is trying to reduce reliance on smartphones and expand custom ASIC sales to cloud providers, with its 2026 AI data-centre ASIC revenue guidance raised to $2 billion [Reuters]

Tata Electronics is in early talks with ASML to explore making components and subassemblies for advanced chipmaking equipment in India, a step that would extend their MoU beyond ASML support for Tata’s $11 billion 300mm fab in Dholera and pull India closer to the lithography supply chain [Economic Times]

In other news:

India’s mobile app market generated a record $345 million in consumer spending in Q2, up 35% year on year, even as quarterly downloads stayed around 6.3 billion since 2023; Sensor Tower said generative AI, streaming and productivity apps drove the gains, with ChatGPT and Claude taking nearly 83% of AI app revenue [TechCrunch]

Australia’s eSafety Commissioner began Federal Court civil penalty proceedings against Telegram, alleging it failed to remove pro-terror material after user reports, including videos of the Christchurch and Buffalo attacks. The case could bring fines of up to A$54.6 million under the Online Safety Act [Reuters]

Amazon has linked recent compromises of popular NPM libraries including axios, debug, chalk and typo-crypto to a DPRK-linked threat actor known as Sapphire Sleet and BlueNoroff, saying maintainers were socially engineered before malicious package updates were pushed. The AWS research says axios has more than 100 million weekly downloads and that attackers are adapting supply-chain tactics for AI-era code review [AWS]

China is rapidly narrowing the satellite technology gap with the US through upgrades to BeiDou, stronger remote-sensing systems and more than 510 surveillance satellites. The buildout is improving Beijing’s ability to track and target forces in the Asia-Pacific, adding a harder military edge to the US-China tech race [WSJ]

India summoned Meta executives after Facebook briefly restricted a video post by Prime Minister Narendra Modi that had first appeared on Instagram and was later shared to Facebook. The company has said the restriction was an error, while New Delhi is seeking an explanation in another flashpoint over political content moderation in India [Reuters]

China’s dominant smart-vacuum makers face possible US disruption after the FCC added foreign-produced advanced robotic devices to its Covered List, blocking new models from equipment authorisation for import and sale. Roborock, Ecovacs, Dreame, Xiaomi and Narwal held 68% of the global robotic vacuum market in 2025, with shipments up 17% to 24.1 million units [SCMP]

South Korean agencies warned that tools and infrastructure used by North Korea’s Lazarus Group appear to have been shared with Gunra ransomware operators targeting Korean organisations. AhnLab found overlaps in malware filenames, privilege-escalation tools, command-and-control servers and SSH key fingerprints, after Lazarus installed espionage backdoors in at least 72 organisations this year [The Record]

Hong Kong lawmakers are deepening ties to Chinese companies flagged by Washington for military links, with a CSRI review finding more Legislative Council members holding roles at firms on US Defense Department or Treasury blacklists. The report also found more personal assets held on the mainland, raising sanctions and conflict-of-interest questions for Hong Kong’s financial hub role [Nikkei Asia]

Malaysia will trial Starlink’s satellite-to-phone internet service as it looks to improve connectivity in areas with limited mobile coverage. The Direct-to-Device service lets compatible phones connect directly to satellites without dedicated equipment, while Starlink has already offered satellite broadband in Malaysia since 2023 [TNGlobal]

Xiaomi launched the SkyNomad N70 Max and N90 Max extended-range SUVs in China, marking its first move beyond pure battery EVs. The hybrid models start at about $36,200 and arrive as Xiaomi chases a 550,000-vehicle 2026 delivery target after shipping 185,055 vehicles in the first half [Bloomberg]

Tencent’s US-based Lightspeed LA has laid off about 80 staff and changed direction on Last Sentinel, its big-budget open-world game led by a former GTA and Red Dead Redemption veteran [Bloomberg]

NTT Data Payment Services India launched ADAPTIS, a unified payments and commerce platform for Indian merchants, bringing payment acceptance, merchant tools, value-added services and commerce capabilities under one brand as the Japanese IT group expands its payments business and works on connecting UPI with fast-payment systems across Asia [Nikkei Asia]

ATR is a daily newsletter, but we have switched to three times a week for the summer and while I’m in Europe. Despite that, the news has come thick and fast this week. Today, we’re contrasting the growth of CXMT and China’s domestic AI players with the volatility of Korea’s winners. We plan to be back on Friday with an issue looking more specifically at what’s happened in Southeast Asia this week so stay tuned.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Subscribe now

Outside of Asia Tech Review, I run Fully Drafted, a communications and consulting company focused on traditional comms, thought leadership and investor relations. Get in touch if you think we can help you.

China’s CXMT surges as Korea’s AI chip champions stumble

A week is a long time in the world of AI stocks. In the space of just days, memory maker CXMT went public and became China’s most valuable stock and SK Hynix, the memory king that’s become akin to a GameStop meme stock, has cratered by 25%.

CXMT’s (ChangXin Technology Group) market debut was a moment of history. The company raised 57.92 billion yuan ($8.6 billion) and its stock jumped nearly 500% on its Shanghai STAR Market debut to become China’s most valuable company. The IPO is the country’s largest since Agricultural Bank of China raised $22 billion in 2010.

Regular readers know CXMT is the world’s fourth-largest producer of DRAM memory, which is crucial to electronics. It is one of the key component makers as China builds out its own self-sufficient ecosystem for AI, right from models to chips, cloud, data centres and more.

CXMT’s timing was ripe given that AI-related stocks surged in the first half of 2026. Zhipu and MiniMax enjoyed stellar debuts in Hong Kong as the world’s first AI stocks to go public, while SK Hynix and Samsung are flying with record profits. Added to that, China is being acknowledged as the disruptor in world AI order thanks to Kimi from Moonshot, DeepSeek and more. CXMT combined those two tailwinds to create a monster IPO.

The stock might look pricey for now, but it has created billions in profits for its backers, which include Alibaba and at least five state-owned banks. The returns were so high that its chairman’s net worth rose past $15 billion, while he pledged to give more than $5 billion of the sum back to employees. (Alibaba made around $464 million, or 20X, on its investment, one of a series of big returns from its AI deals.)

In contrast, South Korean stocks plunged on concerns of China’s threat. The KOSPI rose with the gains of SK Hynix and Samsung but it closed down 6%, marking an 18% fall over five days of trading.

There were always concerns about the volatility of AI stocks. We noted last month that credit from brokers to retail traders rose by more than 60% this year. Now, Korea has clamped down on retail access to leveraged exchange-traded funds (ETFs) which were hit hardest by these drops. Singapore’s MAS is among those raising concern around the broader potential economic impact of AI stock fatigue.

The situation wasn’t helped by SK Hynix missing its earnings estimates this week.

That’s despite yet another set of impressive results, including 557% growth in operating revenue which hit a record 60.54 trillion KRW ($42 billion). Our friend Tim Culpan at Culpium has a deeper look at exactly what happened.

The volatile stock crashes didn’t come out of thin air and CXMT’s listing has been coming for some time. We first noted it back in 2023 and again in January of this year as the plans gathered momentum. So neither event was unexpected and they don’t usher in a drastic new period of world order. In both cases, the writing was on the wall, and that’s exactly why we follow developments closely in ATR.

For more on CXMT, SCMP has a profile of its chairman, Zhu Yiming.

Subscribe now

Deals

Moonshot AI closed a $3.5 billion round at a $35 billion valuation, exceeding its original $1-2 billion target thanks to the impact of its Kimi K3 model. Now it is said to be planning a new round at a $50 billion valuation before a Hong Kong IPO [Bloomberg]

In more news: Moonshot is said to be figuring out how to get its hands on more Nvidia Blackwell chips, which are banned on US export rules, in order to develop its next Kimi K4 model. That’s despite China’s self-sufficiency strategy becoming more sophisticated and wide-ranging [The Information]

Vantage Data Centers is considering selling its Malaysia business for about $2 billion as investors chase AI-driven data centre assets in Southeast Asia [Bloomberg]

AirTrunk secured a $2.3 billion green loan for a Malaysia data centre, adding more financing to Southeast Asia’s AI infrastructure buildout [Bloomberg]

Sixth Street, SK Telecom, GIC and La Caisse are bidding for a major stake in Bain-owned Southeast Asian data centre firm Bridge [Bloomberg]

Zomato founder Deepinder Goyal is targeting a $500 million valuation for his new startup Temple, which is preparing to debut its wearable at Rs 80,000, or about $930 [Economic Times]

SoftBank is eyeing a deal to buy a former Sony payments company as it looks to strengthen its settlements business, extending the Japanese group’s push deeper into fintech infrastructure and digital payments [Nikkei Asia]

SoftBank’s $40 billion loan backing its OpenAI stake has added 21 new lenders, broadening the financing base behind Masayoshi Son’s biggest AI bet as the Japanese group keeps pushing deeper into frontier AI infrastructure and investment [Bloomberg]

Markets

Shein is about to go public, and its pre-listing documents showed it lost $99 million during the first quarter despite net income of over $2 billion last year (Shein also disclosed it is being investigated by the FTC, but it did not give specifics) [Reuters]

Zepto has reportedly dropped its IPO valuation to $3 billion post-money to get the listing over the line, that’s despite a previous valuation of $7 billion and approval from SEBI to go public [Moneycontrol]

China’s DexForce is planning to be the latest robotics and AI firm to go public in Hong Kong [Bloomberg]

Japanese chipmaker Kioxia became the MSCI World Index’s top-performing stock this year, but now it faces a costly AI-driven race to expand flash memory capacity [Bloomberg]

GoTo Group posted its second straight quarterly profit with net income of 350 billion IDR ($19.4 million) as net revenue rose 31% to 5.7 trillion rupiah ($350 million) [Bloomberg]

Bukalapak reported revenue up 29% and positive adjusted EBITDA, improving the Indonesian e-commerce group’s profitability profile after years of pressure on listed tech platforms [Tech in Asia]

AI and Chips

In major news, a Chinese state-backed company has reportedly begun making immersion deep ultraviolet lithography machines. This would mark a major step in Beijing’s efforts to localise chipmaking equipment and reduce reliance on foreign suppliers for advanced semiconductor production [The Information]

That report sent shares of ASML, the Dutch firm which dominates lithography systems, to their lowest level since early June [Bloomberg]

Meta allowed Chinese ad partner GatherOne to run thousands of Facebook and Instagram promotions for AI apps that generate non-consensual sexual imagery, according to the Tech Transparency Project [Bloomberg]

Hackers used an autonomous AI agent in a cyber-espionage campaign against Thailand’s Ministry of Finance, according to analysis that found exposed attacker infrastructure that revealed hundreds of files while the intrusion was still in progress [The Record]

India’s Krutrim laid off 20-25 employees, nearly half its workforce, in a second restructuring after scaling back its full-stack AI strategy [Economic Times]

Taiwanese prosecutors detained an Nvidia employee and searched the company’s offices as part of a widening probe into alleged AI chip smuggling to China [Bloomberg]

Cursor launched a lower priced monthly 649 INR ($6.80) plan for the Indian market, going below its $20 Pro tier for the country which is its third-largest market and top location for ‘power users’ [TechCrunch]

OpenAI wants to use Korea as a testbed to launch its first hardware device early next year, according to President Lee Jae Myung [Korea Herald]

The Trump administration banned imports of new Chinese-made humanoid and quadruped robots and connected power inverters, citing risks to US AI and energy infrastructure [Reuters]

Lenovo Capital is doubling down on AI investments, targeting robotics, coding agents and infrastructure as AI companies make up nearly a third of its portfolio [SCMP]

US chipmaker Marvell Technology will invest $250 million in India over three years, doubling local headcount as it expands technology, talent and infrastructure capabilities [Reuters]

In other news:

India relaxed foreign investment rules to let e-commerce companies own inventory for export-only sales of domestically made goods, giving Amazon and other platforms more room to build India-based global-selling operations while keeping domestic online retail restrictions unchanged [Reuters]

In a national first, South Korea’s KB Kookmin Bank will launch USD corporate cross-border payments on JPMorgan’s Kinexys blockchain network in August for import-export payments across 10 countries [The Block]

Baidu has started testing autonomous vehicles in London through Lyft and Freenow, joining the race to launch commercial robotaxis in the UK [TechCrunch]

Swiggy Instamart named former Myntra CEO Nandita Sinha as CEO from August 3 after Amitesh Kumar Jha resigned to pursue other opportunities [Economic Times]

India’s state-owned Bank of Baroda disclosed that an employee email compromise exposed certain data, but it claimed core banking systems were not affected [The Record]

Malaysia’s digital economy agency withdrew support for Malaysia Blockchain Week after backlash over former OnlyFans creator Ms Pui Yi’s planned after-party DJ set [Bloomberg]

I’m still traveling in Europe, but Asia is where the action is right now. Chinese memory giant CXMT is about to go public in Shanghai in an IPO that will raise nearly $10 billion and flex China’s tech self-sufficiency strategy. Meanwhile, Korean firms are splurging the cash they’ve raked in from AI on deals in the US and beyond. That’s our first stop in today’s newsletter.

To keep up with our issues, follow us on LinkedIn or WhatsApp. Please do share Asia Tech Review with people you know who want to get smart about Asia tech.

Subscribe now

South Korea’s AI earnings are financing a raft of global deals

South Korea has been a major economic beneficiary of the AI boom. Now that financial windfall is touching other countries and companies across the globe through a series of deals.

Samsung, which has posted record results thanks to its surging memory business, is said to be in talks to invest in Mistral in a deal that could value the French company at €20 billion, or $22.8 billion, according to the Financial Times. Arguably Europe’s most prominent AI firm, Mistral develops a model that ranks among the world’s top ten based on usage through OpenRouter.

Mistral has been making inroads in Asia. It has accelerated hiring in its Singapore office, which opened at the beginning of last year, with a focus on winning business in the public sector and financial services. In Europe, it just struck a deal with Microsoft to help it scale its distribution through Azure. A deal with Samsung can give it an edge in Korea, and maybe other countries, by virtue of it being neither American nor Chinese.

Elsewhere, the FT also reported that Korean firms are using their cash piles to grow their production capabilities in the US:

Foreign direct investment executed by South Korean groups in the US during the first quarter more than doubled year on year to $10.2bn, the highest in nearly five years, according to the finance ministry in Seoul. It rose nearly 15 per cent to $25.7bn for the whole of 2025.

We know SK Hynix has committed tens of billions of dollars to building out its US presence, it also created a service business on American soil and has pledged to invest $10 billion into innovative US companies. Samsung, meanwhile, backed AI data-centre cooling startup ZutaCore as part of a recent $100 million round and it backed AI chipmaker Groq last year.

The deals make sense as investments in the future and new possibilities for the Korean duo, which sit on the frontier of AI technology, but also they provide a new source of capital after China’s M&A was effectively snuffed out. Deals like ByteDance’s $1 billion acquisition of Musical.ly in 2017 or Tencent’s purchase of Riot Games in 2015 were from a different era, but no other nations have stepped in to fill the gap.

Korea’s moves are going the other way, too.

President Lee Jae Myung visited Silicon Valley to actively court tech companies to explore partnerships and deals as he continues to make AI the centrepiece of his vision for the future, which includes an $880 billion investment plan. He is also hoping to encourage US investors to back Korea’s emerging tech firms.

The major deals include Nvidia and SK Group announcing a $500 billion initiative around AI data centres and memory, while Nvidia will invest $1 billion in Naver to help finance a South Korean AI data centre being developed with Brookfield. The goal is to allow Naver to more than triple the facility’s size and use Nvidia AI computing hardware.

As the summer ends, look for more deals that bridge Korea and the US. I’m particularly interested in seeing how SK Hynix, a name that wasn’t widely-known before this year, cashes in on its new found status with startup investments in the US.

Subscribe now

Memory giant CXMT on the cusp of blockbuster China IPO

ChangXin Memory Technologies (CXMT) will list in Shanghai today following China’s second highest IPO in history, which raised 66.6 billion CNY ($9.8 billion).

Like SK Hynix and Samsung, CXMT is ideally placed to profit from the AI boom as it develops the memory chips required for AI servers, as well as regular technology like smartphones. The company is the world’s fourth largest manufacturer of DRAM, dynamic random access memory, and the largest in China, which makes it a key player in the domestic push for technology self-sufficiency.

Indeed, CXMT is key to developing China-made data centres that can work with Chinese AI chips and models to give a fully independent tech stack.

Now though, China’s memory chip challengers are turning the AI hardware squeeze into leverage.

CXMT reportedly raised prices, even for major names like Huawei and signed major supply deals with ByteDance and Tencent. With revenue in Q1 increasing 719% to reach $7.5 billion, CXMT is on a roll.

Its IPO will be a major milestone for Beijing.

Outside of the self-sufficiency strategy, a major float is proof that the long-term and government-funded approach to building CXMT can be replicated, whilst generating significant returns for investors.

Deals

Embedded AI startup PsiBot, also known as Lingchu Intelligence, is set to close nearly $100 million in funding at a $1.48 billion valuation [Bloomberg]

Grab made a strategic investment in Vietnamese EV charging company EBOOST, with plans to deploy more than 6,000 charging ports for cars and motorbikes in Vietnam by early 2028 [TechNode Global]

Mirae Asset completed its acquisition of a 92.06% stake in South Korean crypto exchange Korbit for about $93 million [The Block]

ServiceNow invested $40 million in Indian banking software firm BusinessNext at a $700 million valuation [TechCrunch]

Singapore robotics data infrastructure startup Ropedia raised $22 million in Pre-Series A funding, taking total funding to $30 million [TechNode Global]

SoftBank is said to be weighing an acquisition of European robotics startup Gravis Robotics as it pushes deeper into AI-powered robotics [Bloomberg]

Bain Capital and LY Corp. are weighing a higher joint bid for Japanese online marketplace Kakaku.com, escalating a bidding war with EQT after Kakaku.com shares rose nearly 60% this year and lifted its market value to about $3.9 billion [Bloomberg]

Monk’s Hill Ventures is restructuring its leadership and operations as the Singapore VC shuts its Indonesia office and consolidates its investment team in Singapore [DealStreetAsia]

Markets

JD.com received formal notice of EU regulatory concerns over its $2.5 billion bid for German electronics retailer Ceconomy, a move that could require concessions to secure approval [Reuters]

Tencent’s share price fell the most in over a year, dragging Chinese gaming stocks lower as investors grew concerned about its mobile gaming business and some funds rotated back into AI-related names [Bloomberg]

Chinese robotics company AgiBot is said to be pursuing a Hong Kong IPO and has hired Citic Securities, CICC and Morgan Stanley as sponsors [SCMP]

Semight Instruments is considering a second listing in Hong Kong after shares in the Chinese semiconductor test-instrument maker surged roughly 2,200% since its April debut [Bloomberg]

Indonesia’s Danantara has taken control of four state-owned investment managers overseeing a combined 170 trillion rupiah ($9.5 billion) in assets, creating the country’s largest asset manager [Business Times]

Chinese GPU maker MetaX confidentially filed for a Hong Kong listing and is reportedly targeting an IPO by the end of this year [SCMP]

AI and Chips

The US is scrutinising China’s global AI push, with officials examining how Beijing is spreading its models and warning that foreign models found stealing American IP could face sanctions, though Chinese models already account for nearly 60% of US company token usage on OpenRouter [Bloomberg]

Almost 200 Silicon Valley companies, including Proton and Y Combinator, urged the Trump administration not to cut off access to Chinese open-weight AI models [Politico]

Nvidia CEO Jensen Huang said China’s open-source AI models pose no threat to the US but efforts to ban them do [Axios]

White House science adviser Michael Kratsios accused China’s Moonshot of improperly using US AI models and Nvidia GB300 chips accessed in Thailand to build its Kimi K3 system [Bloomberg]

But OpenAI president Greg Brockman said he is unsure if Moonshot AI’s Kimi K3 was distilled from OpenAI technology [Bloomberg]

TCS acquired land parcels in Vizag and Pune that could become sites for OpenAI’s first Indian data centres [Economic Times]

Wistron opened a $700 million facility in Fort Worth, Texas, to produce Nvidia superchips [WSJ]

Volkswagen’s China-based joint venture Carizon is expanding its partnership with Horizon Robotics to accelerate advanced self-driving development [WSJ]

TSMC’s US manufacturing push is increasing costs and squeezing margins as the chipmaker commits $200 billion to American facilities, with overseas fab expansion expected to dilute gross margins by 2 to 4 percentage points over the next several years [CNBC]

Alibaba’s mapping unit Amap unveiled an upgraded ABot robotics framework that combines five specialised foundation AI models for navigation, manipulation, reasoning and control, claiming a 92.9% outdoor navigation success rate using standard cameras and basic road networks [SCMP]

DeepSeek founder Liang Wenfeng said in a rare four-hour investor talk that China’s AI gap with the US is mainly compute, not talent, while arguing open source and restrained profits are central to DeepSeek’s AGI strategy [Inside China]

Crypto

Coinbase is expanding its Singapore operations and plans to grow local headcount to 200 despite global restructuring [Business Times]

Crypto exchange HTX has been rotating wallets across four blockchains to dodge sanctions screening since UK authorities sanctioned the platform in May, according to TRM Labs, which said the platform retires hot wallets and funding addresses every few hours across TRON, Ethereum, BNB Smart Chain and Solana [The Block]

Circle is partnering with South Korea’s Kakao and Toss to explore USDC-powered blockchain payment infrastructure and digital asset services, expanding the stablecoin issuer’s push into one of Asia’s most active retail crypto markets [The Block]

In other news:

China started issuing robotaxi permits again in some cities after a months-long freeze triggered by a Baidu robotaxi outage in Wuhan, with regulators resuming licences gradually after an industry-wide safety review [Bloomberg]

Delhi High Court ruled in favour of OpenAI in a copyright lawsuit brought by Indian news agency ANI, finding that OpenAI’s use of ANI content to train ChatGPT did not amount to copyright infringement [Reuters]

Amazon is being investigated by the US Senate Small Business Committee for allegedly letting China influence the company’s online marketplace [Bloomberg]

Taiwan has loosened foreign-ownership rules for satellite operators, potentially opening the door for Starlink and other low-Earth orbit internet services as Taipei looks to strengthen communications resilience in a crisis [Nikkei Asia]

Flipkart is preparing to enter India’s fast-growing food delivery market with a Bengaluru pilot, taking on Zomato and Swiggy as competition heats up in the category [Business Standard]

Vietnam is weighing a ban on children under 16 posting, commenting or reacting on social media, with a draft decree that would require parents to register and monitor accounts while platforms identify young users and restrict access [Bloomberg]

India’s cybercrime agency directed GitHub to take down Jack Dorsey’s Bluetooth messaging app Bitchat over concerns it could be misused by anti-national elements, sparking legal debate over the government’s power to block repositories [TechCrunch]

Tencent fired a WeChat manager after screenshots of his multimillion-yuan compensation package leaked online and sparked public backlash, highlighting how sensitive pay transparency and internal-data controls have become for Chinese tech companies [SCMP]