Reliance on US technology is not advisable
When Anthropic withdrew its Mythos and Fable models from foreign markets under US government directive, it did more than disrupt a product launch — it exposed the quiet fragility of technological dependence across an entire continent. Asia's nations, each shaped by different histories of self-reliance and foreign entanglement, now face a shared reckoning: in a world where artificial intelligence is becoming foundational infrastructure, sovereignty over that infrastructure may matter as much as sovereignty over land or currency. The distance between those who have prepared and those who have not is already vast, and the events of last week made it impossible to ignore.
- Anthropic's abrupt model withdrawal — triggered by a US government compliance order — stripped away the comfortable assumption that foreign AI platforms are reliable long-term partners.
- The incident exposed a continent-wide fault line: China stands largely insulated, India is mobilizing urgently, South Korea is leveraging industrial strengths, while Southeast Asia remains dangerously fragmented and underfunded.
- India's response has been the most kinetic — startups like Sarvam chasing billion-dollar valuations, government GPU stockpiles crossing 38,000 units, and a national AI mission pivoting from access to infrastructure.
- Southeast Asia's predicament is sharpest: Thailand's $50 million program to subsidize foreign AI access for one year drew criticism as a stopgap that mistakes consumption for capability.
- The widening gap between nations that can build sovereign AI and those that cannot is no longer a future risk — it is a present and accelerating reality.
Anthropic's withdrawal of its Mythos and Fable models last week — pulled days after launch under a US government order restricting foreign access — sent an unmistakable signal across Asia: dependence on American technology companies carries geopolitical risk that can materialize without warning. The incident forced a continent-wide confrontation with a question many governments had been deferring.
China, shaped by years of adversarial pressure from Washington, had already answered that question. Its domestic stack — cloud services, homegrown chips, memory manufacturers, AI models, and venture capital — means US restrictions create opportunity rather than crisis. Forced self-sufficiency, long a burden, now looks like foresight.
India is moving with the most visible urgency. Startups like Sarvam, raising up to $350 million at a $1.5 billion valuation, and Krutrim, pivoting toward cloud infrastructure, are building indigenous AI capacity at speed. Government-backed efforts including BharatGen and IIT Madras's open-source AI4Bharat initiative add institutional weight. The IndiaAI Mission, originally designed to help Indian companies access AI services, is now retooling to build local infrastructure — and has already secured more than 38,000 GPUs.
South Korea is taking a more deliberate industrial approach, selecting five national AI model teams — Naver Cloud, LG AI Research, SK Telecom, NC AI, and Upstage — and drawing on world-class chip and memory manufacturers like Samsung and SK Hynix. Japan, by contrast, is pursuing corporate-led integration rather than explicit sovereignty, with NTT, Rakuten, NEC, and Fujitsu developing Japanese-language models oriented toward manufacturing and services.
Southeast Asia faces the steepest challenge. AI Singapore's Sea-Lion model supports eight regional languages but lags significantly behind global benchmarks. Indonesia, Malaysia, and others have launched small projects, but the resources and coordination required to compete remain out of reach. Thailand's $50 million program to fund citizen access to foreign AI tools for a single year drew criticism for its short-term logic — a telling illustration of the gap between aspiration and capacity.
The Anthropic incident has made the stakes legible. But clarity about a problem and the means to solve it are different things entirely, and the distance between Asia's most and least prepared nations is already widening.
Anthropic's sudden withdrawal of its Mythos and Fable models last week sent a clear message across Asia: relying on American technology companies comes with hidden costs. The company pulled the models just days after launch, citing a US government order to restrict foreign access. Rather than comply selectively, Anthropic removed them entirely while it worked through compliance questions. The incident crystallized something governments and tech leaders across the continent had been circling around for months—the vulnerability of depending on overseas AI providers.
The problem is that Asia's countries are nowhere near equally prepared to handle this vulnerability. China, unsurprisingly, barely flinches. It has spent years building its own stack: domestic cloud services, homegrown chipsets, memory manufacturers, venture capital networks, and AI models. The US restrictions that triggered Anthropic's withdrawal actually create opportunity for Chinese companies to fill the void left behind. Beijing's long adversarial relationship with Washington forced the issue years ago, and now that forced self-sufficiency looks prescient.
India is moving fastest to catch up. A cluster of startups is racing to build indigenous AI capacity. Sarvam, the leader, was raising up to $350 million at a $1.5 billion valuation as recently as April and operates its own AI chat application. Krutrim, an earlier entrant, pivoted toward cloud infrastructure after hitting obstacles. The government has its own players too: BharatGen and AI4Bharat, an open-source initiative from IIT Madras. When Sarvam's CEO Pratyush Kumar saw what happened to Anthropic's models, he posted on X that the moment proved India would have to build its own AI future. The government's IndiaAI Mission, launched two years ago to help Indian companies access AI services, is now retooling itself to create local infrastructure instead. It has already secured more than 38,000 GPUs.
South Korea has taken a more deliberate industrial strategy approach. The government selected five teams for a national AI model program: Naver Cloud, LG AI Research, SK Telecom, NC AI, and Upstage, an AI startup rumored to be heading for an IPO at a $3.3 billion valuation. Korea already possesses much of the underlying infrastructure—Samsung and SK Hynix are globally competitive in chips and memory, and the country has cloud providers, telecom operators, and a government committed to ambitious technology goals. Naver, LG, and Upstage have all developed models with distribution potential. SK Telecom recently partnered with Nvidia on infrastructure.
Japan's approach is more scattered. Sakana AI, founded by former Google researchers and backed by investors like Khosla Ventures and Lux Capital, has become a startup success story valued at $2.65 billion, but it isn't specifically focused on sovereign AI. The real push has come from established corporations—NTT, Rakuten, NEC, and Fujitsu—developing Japanese-language models. The emphasis is less on sovereignty and more on integrating AI into corporate services, robotics, and manufacturing.
Southeast Asia faces the steepest climb. The region is fragmented across multiple countries, each with different resources and priorities. AI Singapore developed the Sea-Lion model, an open-source tool supporting Indonesian, Malay, Thai, Vietnamese, Filipino, Burmese, Lao, and Khmer. But it significantly lags behind top-tier international models in capability. Indonesia's GoTo built Sahabat-AI, Malaysia has ILMU, and smaller projects exist elsewhere. The real problem is that developing sovereign AI models competitive with global options requires enormous time and investment. Thailand's approach—a $50 million program called TH-AI Pass to fund access to premium foreign AI tools for five million citizens for one year—reveals the mindset of governments that have fallen behind. The initiative drew controversy for its short-term thinking and tender scrutiny, but it illustrates the gap between aspiration and capacity.
The Anthropic incident has made the stakes visible. Countries across Asia now understand that reliance on US technology carries geopolitical risk. But understanding the problem and solving it are different things. China has already solved it. India is mobilizing rapidly. South Korea is leveraging existing strengths. Japan is integrating AI into existing industries. Southeast Asia is still trying to figure out what sovereign AI even means for a region of diverse nations with limited resources. The gap between these positions will only widen.
Notable Quotes
From our vantage point, it is super clear that India will build, leverage, and create massive business value and societal impact with sovereign AI— Pratyush Kumar, CEO and co-founder of Sarvam
The Hearth Conversation Another angle on the story
Why does it matter that Anthropic pulled those models? Governments can just use other American AI services, right?
The point isn't that Anthropic specifically matters. It's that the US government can order any American company to restrict access to technology, and that company has to comply. If you're a country or a company depending on that technology, you have no control over what happens next. Anthropic chose to pull the models entirely rather than comply selectively, but the underlying vulnerability is the same.
So China saw this coming years ago?
China didn't see it coming so much as it was forced into it. The US has been restricting Chinese access to American technology for years—chips, software, cloud services. That pressure forced Chinese companies to build their own alternatives. Now when something like the Anthropic withdrawal happens, China barely notices because it already has domestic options.
India seems to be moving fast. Can they actually catch up?
They're moving fast, but they're starting from much further back. Sarvam is well-funded and serious, but building a world-class AI model takes time and enormous compute resources. India's government is helping by securing GPUs and funding, but they're still years behind where China is now. The advantage is that India has the talent, the capital, and now the political will.
What about Southeast Asia? Why is it so far behind?
It's partly fragmentation—you have eight different countries with different languages and resources. But it's also that the region has historically been a market for technology from elsewhere, not a builder of it. When you're a smaller country without a massive tech industry already in place, building sovereign AI feels like a luxury you can't afford. Thailand's approach—just buying access to foreign tools—is pragmatic in the short term but leaves you vulnerable long-term.
Does Japan's approach of integrating AI into existing industries instead of building sovereign models make sense?
It makes sense given Japan's strengths. Japan has world-class robotics, manufacturing, and corporate infrastructure. Integrating AI into those areas creates real value. But it also means Japan isn't building the foundational AI models themselves—they're relying on others to do that work. It's a different bet on where the value actually lies.
What happens next? Do all these countries eventually build their own models?
Some will, some won't. China will keep deepening its self-sufficiency. India will probably build competitive models within five to ten years. South Korea will likely succeed because it has the infrastructure and government support. Japan will probably stay focused on applications. Southeast Asia will remain fragmented unless there's a regional push to collaborate, which seems unlikely. The gap between the leaders and the laggards will keep widening.