When venture capitalist David Sacks takes office as the Trump administration’s AI “czar,” he’ll face a thorny problem: despite U.S. controls on chips and capital, China is developing capable AI models — and sharing them openly.
Models are the reasoning engines that power tools like ChatGPT; if released openly, developers can freely modify them for different applications. Chinese models, such as Qwen from Alibaba, Yi from 01.AI, or DeepSeek, are now among the most capable, most preferred and largest open models globally, and fast setting a global standard in AI.
These may be signs of China’s emerging AI strategy: wielding open technology to create global dependencies on Chinese talent, capital and industry.
That should be a source of concern, and a call to action. However, U.S. policymakers are focusing on the wrong threat. They are reacting with alarm to the prospect of China achieving AI parity with frontier AI labs, urging a “Manhattan Project” to accelerate the development of cutting-edge AI capabilities and introducing new controls on the most powerful chips and largest closed models to keep those capabilities out of China’s hands.
Yet these efforts miss the mark. The national security risk isn’t just that China might rival frontier labs like OpenAI or Anthropic. The most pressing threat is that China is openly sharing capable AI models that could eventually underpin AI infrastructure across the globe.
Open models are popular because they can help overcome significant barriers to AI adoption at scale, including reliability, privacy, security and cost. Developers can modify open models for better performance in specific tasks. They can test, refine and operate their AI systems without sharing confidential data — about users, clients, patients or citizens — with the widely mistrusted Big Tech firms that develop closed-source models. They can deploy AI systems independently and securely through their own on-premise hardware or compute providers. And they can build AI systems using a range of competitive models without incurring ongoing fees to a handful of frontier AI labs. These are some of the reasons why open models like Meta’s Llama have topped 350 million downloads since 2023.
If the U.S. retreats from open innovation in AI, other nations may “decouple” from U.S. technology, investing in their own infrastructure to build satisfactory AI capabilities or turning to alternatives. That would entrench Chinese AI models like Qwen, Yi or DeepSeek, establishing a global reliance on Chinese labs and eroding the influence of U.S. firms.
For example, today, Alibaba’s models are downloaded several million times each month by developers. These models will embed different values (for example, Chinese models are required by law to adhere to Communist Party doctrine, and models from Alibaba exhibit signs of political censorship) and those values could shape the technology stack of countless businesses and governments globally. Imagine a future where Baidu powers the AI search engines used by billions of people globally instead of Perplexity or Google.
Unfortunately, while some policymakers support open innovation in AI, others favor restrictions on open technology. Although the federal telecommunications agency refused to endorse restrictions, senators have tabled bipartisan proposals for a licensing regime that would limit access to capable AI models. The House has advanced legislation to subject AI models to export controls, and the current administration is preparing to impose further restrictions on the computing hardware necessary to train, modify or run capable models. And while the vice president-elect is believed to support open models, frequently criticizing “woke” values embedded in AI systems from Big Tech firms, the president-elect may expand controls for critical technology as part of his trade policy on China.
By chilling or impeding the release of open models, these measures risk stifling AI innovation at home while accelerating the uptake of Chinese technology abroad. U.S. policymakers must recognize that AI leadership isn’t just about building advanced technology behind a paywall —it’s about driving the domestic adoption of AI and establishing global dependencies on U.S. industry.
Here’s how the incoming Trump-Vance administration can act to preserve America’s open-source advantage.
First, the U.S. must invest in open development. Training advanced AI models is costly, and we can’t take open models for granted. The president-elect’s proposal for a U.S. sovereign wealth fund could spur open innovation by making investments in infrastructure, talent and resources that support open technology in the national interest. Other countries are doing so already. Saudi Arabia, the UAE, Canada and France have each committed amounts up to $100 billion and other incentives for AI, including significant investments in open model development.
Second, the U.S. must refocus policy around AI diffusion and adoption — not just AI safety. The Trump-Vance administration should create and execute a unified national strategy for promoting open innovation in AI. This may require a dedicated entity to coordinate efforts across the government and apply regulatory tools to encourage private sector investment. For example, the U.S. Departments of Energy and Commerce could streamline permit and energy access for investments in open model development. In addition, federal procurement policy could prioritize open models for sensitive government applications, like defense and public services.
Finally, the administration should avoid restricting the open release of useful models, such as through export controls or licensing regimes. Restricting open models would limit global access to AI technology, leaving a vacuum for Chinese AI labs to fill. Open models pose unique challenges for AI oversight, including the possibility of misuse, and policymakers should remain alert to emerging risks. However, evidence of catastrophic threats from open models is limited, and developers can apply layers of mitigation. Policymakers must balance these risks against the benefit of open models for transparency and competition in AI, and the opportunity cost of ceding global AI leadership to China.
Rather than treating open models as a threat, the new administration should embrace them as an opportunity. Open models can accelerate domestic AI adoption, shape global AI deployment, and establish lasting dependencies on U.S. talent, capital and industry. To win in AI, the U.S. must champion open-source diplomacy.
Ben Brooks is a fellow at Harvard’s Berkman Klein Center, and previously led public policy for Stability AI, developer of Stable Diffusion.
Michelle Fang leads strategic projects at the AI computing firm Cerebras, and previously organized Sam Altman’s congressional hearing while a Senate AI staffer.
