Beyond Chips: Why AI Diffusion is the Next Great Geopolitical Battlefield

The closing of COMPUTEX 2026 barely gave Taipei time to catch its breath. The day after the global tech showcase wrapped up — drawing professional buyers from over 150 countries and keynote addresses from industry leaders including NVIDIA founder Jensen Huang — a different kind of gathering took shape across the city. The Supply Chain Resilience Strategy International Summit, organized by Taiwan's national think tank DSET, turned the conversation from product launches to strategic risk.

At the heart of the forum's AI resilience session was an argument that cut against the usual framing of US-China tech competition. Martijn Rasser, vice president of the Special Competitive Studies Project's Technology Leadership, made the case plainly: while chips, computing power, and data infrastructure dominate headlines, it is AI diffusion — the deployment and global spread of AI systems — that constitutes the real battlefield.

Rasser identified several vulnerabilities on the American side: US companies move fast on development but are slow to integrate new technology into government operations; export control rules shift unpredictably, undermining business confidence; and Chinese AI models have gained considerable traction internationally. He called for a more stable and predictable regulatory framework that would allow Washington to coordinate effectively with allies in Europe and East Asia. Taiwan, he emphasized, is an indispensable partner in this contest — one whose cooperation is essential if Western AI is to become the global standard.

Matt Cronin, a senior national security adviser at Andreessen Horowitz (a16z), echoed the diffusion-first argument, framing AI development across three parallel tracks: advancing frontier capabilities, deploying systems at scale, and ensuring broad accessibility.

What makes China's approach particularly potent, Cronin explained, is the scale of state-backed support — subsidies covering energy, talent, and infrastructure — that allows Chinese open-source AI models to spread rapidly around the world at low cost. Their price-to-performance ratio has made them attractive to governments and businesses globally.

But Cronin urged caution. Chinese AI models, he noted, operate under safety guidelines that prohibit outputs critical of the Chinese Communist Party — a constraint baked into the architecture, not bolted on as an afterthought. Combined with the outsized influence AI-generated content has on how people receive and process information, this creates what Cronin described as a cognitive warfare risk. His conclusion was unambiguous: the Western world should not be relying on Chinese AI models.

He also made a point that went beyond national security calculus. AI adoption, he argued, should not be limited to elites and institutions. Broader public access — making AI genuinely useful to ordinary people rather than a source of anxiety — would itself contribute to social stability.

Michael Robbins, president and CEO of the Association for Uncrewed Vehicle Systems International, brought a different concern to the table. China's speed, he said, is not simply a product of raw investment — it is accelerated by the systematic theft of intellectual property, which feeds a self-reinforcing development cycle. The real risk is not how quickly China deploys existing technology, but how quickly it can absorb stolen innovations and build on them.

Robbins also pointed to an asymmetry in where American leadership is strongest and where it falls short. The US remains clearly ahead on software and AI development broadly, but it is moving too slowly on hardware — specifically on embodied AI, the integration of artificial intelligence into physical systems such as bipedal and quadrupedal robots.

That lag has strategic implications. China has developed its own humanoid robots, but Robbins warned against treating them as purely commercial products. Chinese tech firms, he argued, operate within a framework oriented toward national development goals, which means the line between commercial and military use is deliberately blurred. Robots built for civilian applications could, in a crisis, be weaponized by the People's Liberation Army.

On the policy side, Dean Ball, a senior fellow at the Foundation for American Innovation, outlined the logic behind the Trump administration's AI Action Plan. The framework, he explained, is designed to support innovation across both open-weight and closed-weight AI models — rejecting the idea that openness and security are necessarily in tension. Ball's prescription for the US is to focus on exporting its full AI technology stack globally and to work with allies toward a compatible, interoperable AI ecosystem.

（Related： When Candles Become Taboo: How Beijing Is Scrubbing the 1989 Massacre from Games and History ｜ Latest ）

Boston Dynamics Chief Legal Officer Jason Fiorillo closed out the session by situating the robotics conversation in demographic terms. Declining birth rates are producing labor shortages across major economies, and intelligent robots will increasingly be called upon to fill those gaps. Fiorillo described Boston Dynamics' mission as threefold: deeply embedding AI into robotic systems, maintaining strong customer demand and satisfaction, and securing the trust and safety of the robot supply chain itself.