The Canadian AI executive says China's models are genuinely innovative, not just copies of American technology
One of the AI industry’s most prominent founders just went against the grain of Washington’s preferred narrative on China. Cohere CEO Aidan Gomez, a co-author of the original transformer paper that underpins virtually every modern AI system, told CNBC that characterizing China’s AI progress as mere theft fundamentally misreads what’s happening on the other side of the Pacific.
His comments arrived two days after the NSA, CISA, and FBI issued a joint advisory on September 8 naming six Chinese firms they accused of conducting industrial-scale distillation campaigns. Distillation is the process of using outputs from a more powerful model to train a cheaper, smaller one.
The case against the copycat theory #
Speaking on CNBC’s “Tech Download” on September 10, Gomez didn’t deny that some distillation has occurred. What he pushed back on was the idea that it explains the full picture.
His argument is simple and hard to dismiss: Chinese AI models are now outperforming their US counterparts on various benchmarks. If these systems were merely distilled copies, they’d be bounded by the capabilities of whatever they were trained on. They wouldn’t be surpassing the originals.
Gomez pointed to the latest GLM models coming out of China, which he said are processing roughly 10 trillion tokens per day, entirely on Chinese-made silicon. That last detail matters enormously. It means China isn’t just building competitive AI software. It’s running that software on domestically produced chips, sidestepping the very export controls the US has spent years tightening.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Washington’s counterargument #
The US government’s position isn’t without evidence. The joint advisory from three major intelligence and security agencies detailed what they described as systematic efforts by Chinese companies to extract knowledge from leading American AI models. In one striking data point from the broader distillation landscape, approximately 28.8 million exchanges through around 25,000 fraudulent Claude accounts were reported within a six-week window in June 2026.
China’s Ministry of Foreign Affairs responded the following day, on September 9, calling the accusations unfounded. Beijing framed its AI achievements as the product of sustained domestic investment, not intellectual property theft.
Where Gomez parts ways with Silicon Valley #
What makes Gomez’s stance notable is how sharply it contrasts with other tech leaders. Nvidia CEO Jensen Huang has generally aligned more closely with the narrative that US technology remains essential to China’s AI progress.
Gomez has different incentives. Cohere, which he co-founded, has built its business around enterprise AI and what’s increasingly called “sovereign AI,” the idea that governments and companies should be able to run AI systems without depending on any single nation’s technology stack.
The distillation debate also raises uncomfortable questions about the durability of AI moats more broadly. If a model’s outputs can be used to bootstrap a competitor, even partially, then the value of being first diminishes rapidly. By that measure, China’s ability to run trillions of daily token operations on domestic chips suggests a level of self-sufficiency that Washington’s policy architects did not anticipate arriving this soon.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our