Anthropic isn’t at SuperAI in Singapore but it stole the show all the same
Tencent, Google, Stripe and others came together for top AI show, but the newest Claude model is the news everyone is talking about
Welcome back to Asia Tech Review, your curated digest to keep up to date with tech news across Asia.
I really wanted to focus on day one of SuperAI, given there’s 10,000-plus attendees, great chats and a host of major tech firms, but Anthropic’s news is the biggest talking point. Ironically, it may create a huge opportunity for AI companies in Asia, especially China.
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.
Anthropic’s price and performance opens opportunities for everyone else
The first day of SuperAI in Singapore was hectic, with an East-meets-West that put Google, Stripe, Mistral and others in the same room as Alibaba, Tencent, StepFun, MiniMax, Z.ai and others. Notable examples were OpenAI and Anthropic, who were not just absent from the stage but absent from the entire show itself.
It’s tough to build a commercial show about AI without the two biggest companies that are building in AI, but there’s a broad mix of companies that range from those providing the building blocks like compute, storage and hardware, to more ambitious companies developing services and those on stage giving us the one-liners. AGI is apparently a decade away, according to MIT physicist Max Tegmark, but a high-level debate between futurist and former VC Balaji Srinivasan and tech analyst Benedict Evans showed that we really can’t tell what the future will bring.
There’s more certainty that Singapore will continue to be a major part of the development of AI in Asia. It continues to be the logical base for companies from the West to hit when they expand to this region, for example the aforementioned Mistral and Stripe, and a key hub of internationalisation for Chinese firms.
The biggest news from SuperAI wasn’t anything said on stage, it was he release of Anthropic’s newest model Claude Fable 5, the first time one of its new Mythos models has been available publicly.
I’m assuming most readers are already familiar with the details, but Fable 5 ships with built-in safeguards in response to the claimed power that Mythos brings. Mythos 5, the version without the guardrails, ships only to selected partners, which include cybersecurity organisations, infrastructure providers and others who had access to the preview model in order to “secure critically important software.”
Instead of SuperAI, Anthropic took nearly 500 software engineers to Tokyo for a Code with Claude event in Japan, its first this year in Asia
Also, and maybe most importantly, Anthropic is giving users of its subscription plans access to Fable for two weeks. From then, it will be available on a usage credit basis until “sufficient capacity” allows it to be restored to subscription plans.
That feels like a major moment in how AI is being sold.
For the first time, there’s a real tiering system. Not just in terms of how Mythos was distributed during its preview mode but how it will be used going forward.
Anthropic is the market leader when it comes to business and enterprise adoption of AI, but Mythos and Fable are considerably more expensive than ever before.
It’s already challenging for big-spending companies to track the ROI of their increased AI spend, Uber’s COO is one executive to note that in public. Some are spending as much on tokens as salaries, if not more.
Despite Mythos/Fable’s increased capabilities, that justification debate gets harder for Anthropic’s target customer. For others, these solutions likely become prohibitively expensive or just simply unnecessary for the scope of what they need. That could leave a window for a pack of other AI companies.
Notably, Chinese AI firms have had to build more with less due to limited access to top-of-the-range hardware and more constrained finances. There are also companies like Weka, who I met at SuperAI, that help optimise the tech stack, whether that is hardware usage, token spend or other costs, which could have just become even more important.
The days of everyone moving to the newest model the day it comes out may be gone. What happens next could be hugely important for AI companies in Asia.
AI, chips and data centres
Meta Platforms is getting its first data centre in India after it agreed to lease a 168-megawatt data centre in Jamnagar, which will run on renewable energy and use desalinated seawater for cooling [Reuters]
China plans to spend around 2 trillion yuan ($295 billion) over five years building a nationwide network of interconnected data centres. The National Development and Reform Commission is drafting the blueprint, with state-owned carriers China Mobile and China Telecom set to operate the bulk of the infrastructure. [Bloomberg]
Singapore launched a national research supercomputer capable of 115 petaFLOPS, nearly four times the combined capacity of its predecessors. ASPIRE 2B, as it is called, is backed by a $270 million investment from the National Research Foundation. [The Edge Singapore]
Foxconn will invest in 1 gigawatt of solar and wind capacity in Vietnam to power its local factories and suppliers [Nikkei]
India’s Zoho unveiled a homegrown server platform called Nathu La which marks its first foray into hardware and it took five years to develop. It claims the servers have already reduced its data centre power consumption by as much as 18% and lowered costs by 20-30%. [Business Wire]
Deals
A coalition of Chinese tech firms including memory chipmaker CXMT and Alibaba has launched a 3.91 billion yuan ($577 million) private equity fund for deep-tech sectors with a focus on long-term financing over shorter-term VC-type deals. [South China Morning Post]
Ant Group’s international arm is in talks to raise about $1 billion at a valuation of $10 billion or more. Existing backers General Atlantic and Silver Lake are said to be among those being approached as the firm weighs a Hong Kong IPO. [Bloomberg]
Japan’s Mujin, which makes operating systems for industrial robots used by Toyota and Fast Retailing, is said to be raising an extension to its $233 million Series D which would value it over $1 billion ahead of an IPO before the end of the year. [Bloomberg]
NTT, Japan’s largest telecom operator, is launching a $500 million venture fund this month with offices in Silicon Valley and Tokyo to back optical networking startups competing in the AI data centre market. [Nikkei]
Markets
SoftBank’s efforts to raise at least $6 billion through a margin loan backed by its OpenAI stake are said to have stalled, just weeks after it cut the target from $10 billion [Bloomberg]
SK Hynix is reportedly looking for a US listing as soon as August to maximise its incredible stock market run [Reuters]
Xi’an UniIC, a Tsinghua UniGroup-backed DRAM maker that’s a smaller rival to CXMT, is also heading to public markets after completing mandatory phases for a Beijing listing [South China Morning Post]
Restructuring
Moonshot, DeepRoute.ai and Kling, the AI business from Kuaishou, are unwinding their offshore holding structures (‘red chip’ structures) to enable domestic listings in China and avoid Manus-like scrutiny [Financial Times]
A Singapore court is seizing the majority stake that Capital A, AirAsia’s holding company, holds in fintech firm BigPay and its 11.45% share in logistics unit Teleport after it lost an arbitration case brought by two shareholders [Bloomberg]
Sea is cutting around 8% of its developer workforce for Shopee, removing hundreds of jobs, as the company doubles down on AI [The Straits Times]
Microsoft is cutting between 200 and 400 jobs at its Azure cloud operations in China across Beijing and Shanghai [South China Morning Post]
Other news
India has reportedly frozen final commercial clearances for Elon Musk’s Starlink due to security concerns [Bloomberg]
China’s new drone regulations have hammered the country’s market, formerly worth $26 billion. Second-hand device retailer Leyishou saw a 30% price crash as owners offloaded unwanted machines [Financial Times]
Taiwan is weighing tighter export controls on AI chip sales to China, seeking to align more closely with US restrictions and crack down on semiconductor smuggling [Bloomberg]
How robotics firm Unitree’s growth strategy mimics those used by BYD and DJI to leverage its quadruped robot dominance to create and lead the humanoid market [SemiAnalysis]
OpenAI says it banned China-linked accounts that used ChatGPT to seed anti-data-centre content in US social media debates [OpenAI]