A software engineer working for a small AI startup in Silicon Valley this week had to weigh what seemed like a simple decision: continue renting expensive Nvidia servers, or experiment with a cheaper open-source model developed by China. This is precisely the choice thousands of AI developers worldwide are facing, and it reflects the technological power struggle that President Xi Jinping just made public in Shanghai.
Chip export controls may slow China down, but history shows they rarely stop it entirely.
Two Opposing Strategies
In his opening speech, Xi called for international cooperation on AI and warned countries against using national security as an excuse to put their interests above others — an indirect criticism of US export control policies. His speech came with concrete commitments: Beijing will provide 5,000 AI training slots for developing nations over the next five years, expand cooperation with ASEAN, the Arab League, the African Union, and BRICS, and share an AI weather warning system with 30 countries. A day before the conference, 29 countries — including Russia, Pakistan, and Kazakhstan — signed an agreement to establish the World Artificial Intelligence Cooperation Organization, headquartered in Shanghai.
The US is pursuing the opposite direction: tightening, not expanding. According to analysis by the American Action Forum, Washington has broadened export controls beyond physical chips to include cloud computing services that could allow foreign entities to access advanced computing power without requiring hardware transfer. In 2026, the Trump administration restricted foreign access to Anthropic's most advanced AI models.
Who's Winning, Who's Falling Behind
Data shows the US still maintains an overwhelming infrastructure advantage: in the final quarter of 2025, US companies held 96% of global computing capacity, with Nvidia alone accounting for 67%. But the gap in AI models is narrowing rapidly. China's Moonshot AI startup just launched its Kimi K3 model with approximately 2.8 trillion parameters, and this model topped the rankings for coding on the Arena platform run by researchers at UC Berkeley, just days after its release. A professor at the University of Pennsylvania assessed that this model had come very close to the frontier of leading-edge technology. According to Hürriyet Daily News, the volume of AI tokens used daily in China has increased more than 1,000 times over the past two years.
Why Export Controls Are Hard to Make Airtight
History shows that Cold War-era technology control measures slowed but did not prevent the other side from accessing technology — a precedent that the American Action Forum cites when analyzing pending bills before the US Congress, including the Chip Security Act and MATCH Act aimed at closing loopholes such as shell companies and intermediaries. Meanwhile, according to CNA, Chinese AI models are attracting global users thanks to significantly lower costs compared to American products — a market advantage that chip export controls cannot eliminate.
What to Watch Next
The real question is not who is more technically skilled, but which ecosystem — open or closed — the rest of the world, especially developing nations, will choose as their long-term platform. With small developers and businesses in the US calculating computing costs, Congress's direction on the aforementioned export control bills will determine whether the US infrastructure advantage can be preserved or will gradually be eroded by China's cheaper models.