Asian tech firms build rival AI models as US curbs persist

Published: 2026-06-30T11:19:21.000Z

Restrictions designed to hold rivals back have a habit of spurring them to build their own alternatives, and the latest evidence is arriving fast from Asia. As Washington's export controls on the most advanced American AI continue to bite, technology companies across the region are racing to launch their own frontier models, pitched explicitly as capable substitutes that come without the political risk of an export ban. The clearest sign of the trend landed this week, when within days of one another a Tokyo startup and a Beijing cybersecurity firm both unveiled models they claim stand level with the very American systems that non Americans can no longer access. The new entrants are revealing in both their timing and their framing. Sakana AI, the Tokyo based startup co founded by the former Google researchers David Ha and Llion Jones alongside the ex Mercari and Stability AI executive Ren Ito, launched a frontier model called Fugu, named after the Japanese blowfish. The company says it stands shoulder to shoulder with leading American models and, tellingly, its website advertises frontier capability without the risk of export controls. Days earlier, the Chinese cybersecurity firm 360 reportedly unveiled a tool called Tulongfeng, which it pitches as a direct rival to the powerful, security focused American model currently barred from non American hands. Sakana insists the timing was coincidental, with the underlying research presented at a conference in the spring, but it has not shied away from the moment. What makes this more than opportunism is the strategic argument the founders are making. Ha has framed Fugu as something beyond a simple land grab during a vulnerable moment for American competitors, describing it as an orchestration model designed to coordinate the use of many different AI systems rather than simply being a bigger model in its own right. His broader point cuts to the heart of why export controls may backfire, since relying on a single provider for national infrastructure is a risk he argues the recent restrictions have made impossible to ignore. Access to top models, he warned, can disappear overnight, and a diversified, collective approach is the practical hedge against that concentration of power. Ito has gone further in public writing, urging Washington to prioritize preserving access for allies and arguing that AI should be developed together rather than hoarded. The China dimension shows the same dynamic in a harder form. Locked out of the best American chips, Chinese labs have learned to wring more performance from less capable hardware, with one prominent model achieving near frontier results by redesigning its architecture and optimizing training efficiency on lower grade chips. The lesson, as one industry figure put it, is that Chinese researchers are getting more out of less, and once those techniques mature they spread quickly across the ecosystem, eroding the strategic value of raw compute supremacy. The risk for American labs is stark, since local alternatives trained to better understand local language and nuance are already filling the gap, and trust, once lost, may never fully return. The regional thread runs straight to the Gulf. Saudi Arabia and the UAE have poured sovereign capital into building their own AI champions, from the PIF backed HUMAIN to Abu Dhabi's cluster of model and infrastructure bets, precisely because they do not want to depend wholly on systems that a foreign government could restrict. The proliferation of capable, export friendly Asian models gives the region more options to draw on as it builds out sovereign AI, reinforcing a world in which frontier capability is no longer the preserve of a single country.