Asia's AI labs ship homegrown models as US export curbs bite
As Washington walls off Anthropic's top models, Tokyo's Sakana and China's 360 are shipping their own. For India's sovereign-AI push, the warning could not be louder.
The News
The race to build sovereign artificial intelligence has a new urgency in Asia, and the trigger is American policy. Roughly two weeks ago the Trump administration barred non-US organisations from Anthropic's flagship Mythos system and its cut-down sibling Fable 5. The response from the region's labs has been swift: rather than lobby for reinstated access, several are simply shipping their own frontier-grade alternatives.
Tokyo's Sakana AI, founded in 2023 by Ren Ito, Llion Jones and David Ha, has rolled out a model it calls Fugu, named after the Japanese blowfish. The startup is no minnow. It raised a $135M Series B in November 2025 at a $2.65B valuation, giving it the balance sheet to chase capability that no longer depends on Washington's goodwill. In China, the cybersecurity group 360 has gone further into the security niche, launching Tulongfeng for hunting software vulnerabilities and Yitianzhen for cyber defence and incident response.
The pitch is blunt. A Sakana spokesperson framed Fugu as a way of "delivering frontier capability without the risk of export controls", while chief executive David Ha warned that "access to top models can disappear overnight". Co-founder Ren Ito argued the United States should "preserve access" for allies, adding that "AI should not become a technology that is hoarded".
Why It Matters
The ban lands at an awkward moment for Anthropic, whose run-rate revenue had reached $47 billion in May 2026. Cutting off paying customers across Asia is not a costless act of policy, and the long-term danger for US labs is structural rather than quarterly. When a supplier becomes politically unreliable, buyers do not wait; they build.
The personal-computer industry learned this in the 1980s, when export limits on advanced chips accelerated rival ecosystems abroad rather than smothering them. A model embargo risks the same outcome, seeding the very competitors it was meant to contain.
Indian Angle
India sits squarely in the blast radius, and not as a bystander. The country's developers and enterprises are among the heaviest consumers of US frontier models, and any rule that severs non-American access applies to Bengaluru just as surely as to Tokyo. That makes the IndiaAI Mission's sovereign-model push look less like national pride and more like supply-chain insurance. Sarvam AI, anointed under the mission to build an indigenous foundation model, and Krutrim, the Ola-backed lab that became India's first AI unicorn, are suddenly working on the same problem Sakana just answered: how to guarantee that the smartest model in the stack cannot be switched off by a foreign cabinet.
The cost calculus matters too. Indian startups already pay in dollars for inference, so rupee depreciation quietly inflates every API bill. A homegrown model hosted on domestic infrastructure converts that volatile foreign-currency expense into a fixed local one, a meaningful hedge for cash-tight founders.
For MeitY and the regulators around it, the episode is a live case study in why GPU procurement, data-centre capacity and open-weight strategy belong on the same desk. The lesson from Tokyo is simple: sovereignty is cheaper to build before you need it than after the door closes.
FAQ
What did the US actually ban?
Roughly two weeks ago the Trump administration barred non-US organisations from accessing Anthropic's flagship Mythos system and its restricted version, Fable 5. The curb covers paying customers across Asia, which is what prompted regional labs to launch their own frontier-grade alternatives rather than wait for reinstated access.
Which Asian models have launched in response?
Tokyo's Sakana AI released Fugu, named after the Japanese blowfish. In China, the cybersecurity firm 360 launched Tulongfeng, aimed at finding software vulnerabilities, and Yitianzhen, built for cyber defence and incident response. More are expected as labs treat US access as politically uncertain.
What does this mean for Indian developers?
Any rule that severs non-American access applies to India too. Heavy users of US models in Bengaluru and beyond face the same switch-off risk, strengthening the case for the IndiaAI Mission's sovereign models from Sarvam AI and Krutrim, and for hosting inference on domestic infrastructure priced in rupees.
Where can I read the original report?
TechCrunch broke the story, reported by Kate Park on 27 June 2026. The full coverage details Sakana's funding, the 360 launches and quotes from the founders, and is linked in the attribution below.
This story was reported by TechCrunch. Read the full original coverage at TechCrunch.