In a rare briefing, the OpenAI CEO warned that the United States may not fully grasp the complexity and danger of China’s progress in artificial intelligence, noting that export controls alone are not a sufficient solution and that Chinese competition was a key factor in OpenAI’s recent decision to adopt a more open approach.
Article contents:
- Introduction: A concerning perspective from the AI summit
- A complex, multi-dimensional race
- Skepticism over the effectiveness of export controls
- The shifting US policy maze
- The influence of Chinese open source on OpenAI’s strategy
- OpenAI’s strategic shift: Open-weight models
- Mixed reviews for the new models
- Conclusion: An intertwined competition that will shape the future
- Frequently asked questions
Introduction: A concerning perspective from the AI summit
Sam Altman, CEO of OpenAI, has warned that the United States may be underestimating the complexity and danger of the progress China is making in the field of artificial intelligence, stressing that relying on export controls alone is unlikely to be a reliable solution. Altman said explicitly in a rare briefing to a small group of reporters: “I am worried about China.”
These comments come at a time when Washington is adjusting its policies designed to curb China’s AI ambitions, adding an influential voice to the debate on how best to manage this critical technological competition.
A complex, multi-dimensional race
Altman explained that the race in artificial intelligence between the United States and China is deeply intertwined and more important than simply scoring points to see who is ahead. He said: “There is reasoning capability, where China can perhaps build faster. And there is research, products; there are many layers to the whole thing.” He added: “I don’t think it’s going to be as simple as: Is the US or China ahead?”
This view indicates that superiority is not measured solely by the quality of the best language model, but also includes manufacturing capability, application speed, deployment in products, and research innovation.
Skepticism over the effectiveness of export controls
Despite escalating US export controls on semiconductors, Altman expressed doubt that this policy is keeping pace with technical reality. When asked if fewer GPUs reaching China was reassuring, Altman was skeptical. He said: “My intuition is that it doesn’t work.”
He elaborated: “You can impose export controls on one thing, but maybe not the right thing… People might build factories or find other solutions.” This suggests that China could overcome restrictions by building its own manufacturing capabilities (just as it is doing with Huawei) or finding alternative ways to obtain the necessary computing power.
The shifting US policy maze
US policy toward AI chips heading to China has seen significant fluctuations. The Biden administration initially tightened export controls, but President Donald Trump went further in April, completely halting the supply of advanced chips. However, last week, the US introduced an exemption for certain “China-safe” chips, allowing sales to resume under a controversial agreement requiring Nvidia and AMD to give the federal government 15% of their revenue from China chips. This complex system could be difficult to enforce effectively.
The influence of Chinese open source on OpenAI’s strategy
Altman revealed that China’s AI progress has also influenced how OpenAI thinks about releasing its own models. He acknowledged that competition from Chinese models—especially open-source systems like DeepSeek—was a factor in OpenAI’s recent decision to release its own “open-weight” models.
Altman said: “It was clear that if we didn’t do this, the world would trend toward being mostly built on Chinese open-source models. That was a factor in our decision, for sure. It wasn’t the only one, but it was a big one.”
OpenAI’s strategic shift: Open-weight models
Earlier this month, OpenAI released two open-weight language models, the first since GPT-2 in 2019, marking a major shift in the company’s strategy of keeping its technology closed. A model is considered open-weight if its parameters (the values learned during training) are publicly available, allowing developers to download, run, and customize them. However, this is not the same as fully open source, as OpenAI still does not release training data or complete source code.
With this release, OpenAI joins the wave of openness, betting that broader access will help grow its developer ecosystem and strengthen its position against Chinese competitors.
Mixed reviews for the new models
The debut of the open-weight models drew mixed reactions. Some developers described the models as disappointing, noting that many of the capabilities that make OpenAI’s commercial offerings so powerful have been stripped away. Altman did not dispute this, saying the team deliberately optimized them for a single core use case: locally run coding agents.
Conclusion: An intertwined competition that will shape the future
Sam Altman’s statements present a realistic and concerning view of the global AI race. It is no longer just a competition between companies, but a complex strategic struggle between nations with technical, economic, and geopolitical dimensions. His warning against underestimating China’s progress and his call for a more nuanced approach than mere export controls serve as a call for policymakers and the tech community to reassess their strategies in this competition that will undoubtedly shape the future of the 21st century.
Frequently asked questions
Q1: What is Sam Altman’s main warning regarding China?
A: He warns that the United States is underestimating the extent of China’s progress in artificial intelligence, and that the race is much more complex than simply knowing who has the best model.
Q2: Why does Altman doubt the effectiveness of chip export controls?
A: Because he believes China can find alternative ways, such as building its own factories or developing other technologies, to bypass these restrictions.
Q3: How has China influenced OpenAI’s strategy?
A: Altman said that competition from Chinese open-source models was a key factor that pushed OpenAI to release its own open-weight models to maintain its competitiveness in the global developer community.
