Asian AI Startups Push Local Frontier Models as U.S. Export Curbs Open a Gap

Two Asian AI companies have moved quickly to seize a geopolitical opening left by Washington’s restrictions on advanced Anthropic models, unveiling new systems that they say can compete with some of the most capable tools in the market. One launch came from Tokyo-based Sakana AI, which introduced Fugu, a model built for agentic workflows and local enterprise use. The other came from Chinese cybersecurity firm 360, which reportedly released a pair of security-focused AI products designed to surface vulnerabilities and automate incident response.

The timing is no accident in market terms, even if one of the companies insists the release schedule was not tied to the U.S. government’s decision. The Trump administration’s export ban, which limits access to Anthropic’s Mythos and a more restricted variant called Fable 5 for non-Americans, has created uncertainty for Asian companies that rely on frontier U.S. AI systems. In that environment, local alternatives are suddenly more than just a national pride project. They are becoming a practical business continuity strategy.

For startups and enterprise buyers across Asia, the message is increasingly clear: if access to top-tier American AI can be withdrawn overnight, companies need options closer to home. The result is a fast-evolving regional market where model quality, language fit, regulatory comfort and sovereignty concerns are all starting to matter as much as raw benchmark performance.

Why these launches matter now

The most important backdrop to the recent product announcements is not technical, but political. Two weeks ago, the U.S. government imposed an export restriction that bars global access to Anthropic’s Mythos and its more limited companion model, Fable 5. The policy has not only limited availability; it has also underscored how quickly AI access can become entangled with national security and diplomatic considerations.

That has immediate consequences for companies across Asia that have built products, services and internal workflows around American AI infrastructure. If a model becomes unavailable, or if access is throttled, local companies can face business interruption, compliance headaches and spiraling costs. In sectors such as cybersecurity, finance, government services and manufacturing, the stakes are especially high.

Sakana AI and 360 are effectively making the same commercial argument from different directions: local AI systems can reduce dependence on foreign suppliers while offering better control over data, language, infrastructure and policy exposure.

Sakana AI’s Fugu targets agents, enterprises and policy risk

Sakana AI, founded in 2023 by former Google researchers Ren Ito, Llion Jones and David Ha, has been working on cost-efficient generative AI systems that are designed to perform well with limited data and to better reflect Japanese language and cultural context. The company says Fugu, named after the Japanese word for blowfish, is a frontier model that can stand alongside Anthropic’s latest systems.

More than a simple chatbot or general-purpose model, Fugu is pitched as an orchestration layer for agents. According to the company, it can coordinate access to other models through APIs, making it useful for multi-model environments where different systems may be better suited to different tasks. That design choice places Fugu in a growing category of AI infrastructure focused less on a single all-powerful model and more on managing a fleet of models effectively.

Sakana’s team said the product was developed over the course of last year and that its underlying research was presented at ICLR this spring, adding that the launch timing happened to overlap with the export-ban debate rather than being planned around it.

Still, the company is not hiding the fact that the policy shift gives its product a timely commercial angle. Its website now highlights the promise of frontier-level capability without the risk of export-control exposure, a line that speaks directly to business and government buyers uneasy about dependence on overseas vendors.

A model built for the Japanese market

Fugu is part of a broader strategy at Sakana to build systems that fit Japan’s language and business environment more naturally than imported alternatives. That matters because even powerful general models can struggle with local nuances, formal business tone, policy vocabulary, and the subtle differences in how organizations communicate across regions.

For Japanese firms, a model that is trained and tuned with those constraints in mind may be easier to deploy, easier to govern and easier to justify internally. That does not necessarily make it better in a universal sense, but it can make it more useful in practical settings such as customer support, document analysis, internal assistants and workflow automation.

The company’s positioning also reflects a deeper truth about AI adoption in Asia: many organizations do not want to choose between frontier performance and local control. They want both. The rise of models like Fugu suggests that the market is maturing beyond a simple import-versus-build debate into a more nuanced question of how to combine global capability with regional resilience.

The company says U.S. models still matter

Despite the clear opportunity created by U.S. restrictions, Sakana has been careful not to frame the moment as a permanent break with American AI. A company spokesperson told TechCrunch that U.S. models remain important to Asia and that the current situation should be viewed as a temporary phase, not a lasting realignment.

The spokesperson said the company sees the present moment as evidence of why access matters, not as proof that Asia will abandon American providers entirely. In that framing, the goal is diversification, not decoupling.

That distinction is important. For many enterprise buyers, the ideal future is not a world in which all U.S. models are displaced. It is a world in which no single foreign supplier can create a strategic bottleneck. The more dependencies a business has, the more vulnerable it becomes to policy shocks, pricing changes and product shutdowns.

Ren Ito, one of Sakana’s co-founders, has been making that case publicly. In remarks around the G7 summit in Evian last week, where AI access and export controls were central topics, he argued that the U.S. should prioritize preserving access for allies. In a recent op-ed, he also warned against turning AI into a technology that is hoarded rather than jointly developed.

360 enters the field with security tools and a strategic message

While Sakana’s launch is aimed at enterprise AI orchestration and local-language capability, Chinese cybersecurity firm 360 appears to be leaning into a different but equally strategic use case: AI for cyber defense. According to reporting cited by TechCrunch, the company unveiled Tulongfeng, which it says can automatically discover software vulnerabilities, and Yitianzhen, which is intended to automate incident response and defensive operations.

In other words, 360 is not just offering another general-purpose model. It is positioning AI as a national security asset for cybersecurity operations, where faster vulnerability discovery and quicker response times can have real-world consequences for government systems, enterprise networks and critical infrastructure.

The company’s founder, Zhou Hongyi, reportedly framed AI-assisted vulnerability research as a strategic capability and warned about the danger of “one-way transparency” — a situation in which some actors can access advanced offensive or defensive capabilities while others cannot. That message mirrors broader concerns in China and elsewhere that the AI race is not only about commercial products, but about who controls the tools that protect digital systems.

A more assertive posture than Sakana’s

Unlike Sakana, which is carefully avoiding the impression that it is replacing American AI outright, 360 appears to be more openly using the export-ban moment as a signal of strategic independence. The company’s product pitch suggests that cybersecurity AI should be treated as core infrastructure, and that dependence on restricted foreign capabilities is a liability.

That posture fits China’s wider technology policy environment, where local substitutes are often encouraged when foreign tools become unreliable, politically sensitive or difficult to source. The cybersecurity angle also makes sense commercially because AI tools that can scan for vulnerabilities and speed up response workflows are directly relevant to businesses under constant threat from attackers.

360 did not respond to a request for comment, according to the source reporting.

The business case for local frontier models

The appearance of new models in Tokyo and Beijing highlights a broader trend: geopolitical friction is pushing Asia’s AI market toward regionalization. If the last few years were defined by the scramble to access the best American models, the next phase may be defined by redundancy, sovereignty and local optimization.

That shift could benefit startups that understand the language, regulations and procurement processes of their home markets. It could also encourage governments and regulated industries to prefer local vendors, especially when sensitive data or national infrastructure is involved.

There are several practical reasons these products may resonate beyond the current headlines:

• Access stability: local providers are less exposed to foreign export controls or sudden policy changes.
• Language quality: models tuned to Japanese or Chinese can perform better on culturally specific tasks.
• Data governance: some organizations are more comfortable keeping workloads within domestic or regional systems.
• Procurement certainty: governments and regulated businesses often prefer vendors with fewer geopolitical constraints.
• Multi-model flexibility: orchestration systems can reduce dependence on any one model provider.

That said, local models still face a major challenge: matching the performance, scale and product ecosystem of the leading U.S. labs. Frontier AI remains expensive to train and operate, and benchmarks are only one part of the equation. Distribution, reliability, developer tooling and trust also matter.

Anthropic’s scale shows what is at stake

The stakes are especially clear in Anthropic’s own growth trajectory. The company said in May 2026 that its annualized revenue run rate had crossed $47 billion, a figure that suggests massive enterprise demand and rapid commercialization. But the company’s exact exposure to Asia is not public, so it is difficult to know how much business could be affected if the restrictions remain in place.

Still, even a partial shift away from U.S. providers could have meaningful consequences. Enterprise AI contracts are sticky, and once a company standardizes on a local alternative for compliance or continuity reasons, it may be slow to return even if policy changes later.

That creates a powerful opening for local players. If a Japanese or Chinese model becomes deeply embedded in business workflows during a period of restricted U.S. access, it may not be easy for an American competitor to reclaim that customer relationship afterward.

Trust, not just performance, will decide the winner

In many enterprise settings, the decision is no longer just “Which model is smartest?” Instead, decision-makers are asking a longer list of questions:

1. Will the model still be available next quarter?
2. Can our data stay inside approved jurisdictions?
3. Does the model handle our language and terminology well enough?
4. Can we build processes around it without risking a future shutdown?
5. Does the provider align with our regulatory and political environment?

Those questions can matter more than benchmark scores, especially in markets where AI adoption is being driven by operational necessity rather than experimentation. That is why local companies are now pitching resilience as aggressively as intelligence.

What this means for Asia’s AI market

Asia is unlikely to form a unified AI bloc. Japan, China, South Korea, India and Southeast Asia each have different regulatory systems, procurement norms and market incentives. But the current wave of launches suggests that a more fragmented, regionally adapted AI landscape is emerging.

For Japan, Sakana’s Fugu represents a model of quiet strategic diversification. It is not claiming to dethrone U.S. AI. Instead, it is positioning itself as a dependable option that can sit alongside foreign tools and reduce the risk of sudden access loss.

For China, 360’s move reflects a more forceful argument that local capability is not just preferable but strategically necessary. Cybersecurity in particular is a field where domestic control may be viewed as essential.

Together, the launches suggest that export controls are not simply restricting access to advanced models. They are also accelerating the creation of rival ecosystems that may prove harder to reverse than policymakers expect.

How the current moment compares

The situation can be understood as a quick pivot from dependence to diversification. The following table summarizes the main players and what each is trying to achieve.

This comparison shows that the two launches are not identical, even if they share a common catalyst. Sakana is building flexibility into a multi-model future, while 360 is emphasizing security autonomy and strategic control.

Timeline of the recent developments

The sequence of events matters because it shows how quickly policy can reshape product strategy.

The broader strategic debate

At the center of this story is a question that will shape AI markets for years: should frontier AI be treated like a globally shared general-purpose technology, or like a strategic asset controlled by national governments?

The U.S. export-ban approach suggests the latter. Sakana’s response suggests that when access becomes uncertain, allied markets will respond by building substitutes rather than waiting indefinitely. 360’s move suggests that China is prepared to frame local AI not merely as a commercial opportunity, but as a pillar of digital sovereignty.

This is where the politics and the economics converge. Once buyers begin to assume that AI access is conditional, they start planning around that risk. And once enough buyers do that, local suppliers get a chance to grow, improve and lock in new customers.

Ren Ito has argued that the most productive path is not a closed race for control, but a model in which access is preserved and AI is developed collaboratively with allies.

That view may appeal to policymakers who want to limit security risks without fragmenting the technology ecosystem entirely. But as the launches from Tokyo and Beijing show, companies on the ground are already adapting to the reality they face now, not the one they would prefer.

What to watch next

Several questions will determine whether these launches become a lasting trend or a temporary spike in attention.

• Will Japanese enterprises adopt Fugu in meaningful numbers, or continue to rely on U.S. providers once access conditions stabilize?
• Can Sakana prove that orchestration models are a real product category, not just a response to market disruption?
• Will Chinese customers view 360’s tools as enterprise-ready, scalable and trustworthy for high-stakes security operations?
• Could additional Asian startups launch similarly positioned models if the export controls remain in force?
• Will the U.S. revise, narrow or expand the restrictions on Anthropic’s models?

For now, the most immediate effect is that a policy decision in Washington has created room for local AI builders in Asia to argue that independence, resilience and language-specific capability matter as much as raw scale. In an industry where access can change overnight, that argument may be one of the strongest commercial pitches available.

Whether this becomes a genuine market realignment or simply a short-lived scramble for alternatives will depend on what happens next in both Washington and the major Asian tech hubs. But the early signs are clear: when the frontier gets fenced off, others move in to build a gate of their own.

superintelligencenews@gmail.com