Reevaluating US AI Strategy Against China
A key priority for the U.S. government is to restrict the Chinese military from accessing cutting-edge artificial intelligence (AI) technology to maintain technological superiority and protect national security. The Biden administration sought to achieve this goal by imposing export restrictions on advanced semiconductors and AI models. However, recent developments, including DeepSeek’s notable successes, have cast doubt on the effectiveness of the U.S. export control policy and show how U.S. firms may ultimately pay the price unless the Trump administration takes a new approach.
The U.S. policy on Chinese AI rests on several speculative and likely flawed premises, including the assumption that superintelligent AI is achievable in the near future, that the first country to develop it will obtain an insurmountable military lead, and that the U.S. government can prevent such advanced AI from reaching the Chinese government. None of these assumptions may prove to be true.
Consider the challenges of developing human-level, let alone superintelligent AI. Today’s most advanced AI models still lack fundamental capabilities necessary for human-like intelligence such as an understanding of the physical world, persistent memory, reasoning skills, and complex planning.
Moreover, restricting China’s ability to access America’s most advanced chips and models would only meaningfully impact China’s future AI capabilities if tomorrow’s breakthroughs emerge from the same technological paradigm. Even if the United States successfully prevents non-U.S. entities from developing a game-changing AI system, China’s demonstrated capabilities in cyber espionage raise the risk that it could simply steal the technology.
Why are proponents of export controls so worried about superintelligent AI? Dario Amodei, the CEO of Anthropic who has publicly lobbied for U.S. export controls on AI, has argued that such controls are necessary “because AI systems can eventually help make even smarter AI systems, a temporary lead could be parlayed into a durable advantage.” There is no doubt that AI developers can use AI to code faster and optimize processes. However, the idea that the first country to reach a certain level of advanced AI has not only won the AI race but has also gained a persistent long-term military advantage is a much more far-fetched scenario.
Indeed, the idea that AI will reach a point that not only surpasses human cognitive capabilities but then begins a self-perpetuating cycle of improvements is commonly referred to as “the singularity.” While Amodei has thus far refrained from making that claim directly, Sam Altman, the CEO of OpenAI, made waves on social media when part of his first post of 2025 stated “near the singularity.” The singularity is a popular plot device in science fiction, but it is an untested concept that should not be the cornerstone of U.S. national security policy.
The current U.S. approach also carries significant economic and strategic risks for the United States. Export controls are inadvertently driving demand for Chinese chip manufacturers. Each time an AI firm chooses a Chinese supplier over an American one, the U.S. semiconductor industry loses ground. As Chinese AI firms scale, they further incentivize the development of a domestic chip ecosystem, reducing reliance on U.S. hardware. China is investing in open-source chip architectures such as RISC-V, and DeepSeek has proposed optimizations that differ from current GPU architectures.
If Chinese AI firms gain dominance and their models run on alternative chip designs, U.S. semiconductor firms risk being sidelined in the next wave of AI innovation. In essence, U.S. policy sacrifices long-term competitiveness based on the fragile assumption that doing so will prevent China from developing or stealing superintelligent AI in the near future.
The reality is that the United States cannot stop China’s AI progress. As one of the world’s leading AI adopters, China represents an enormous market for chips. The U.S. faces a stark choice: Either abandon this entire market in hopes of temporarily keeping this technology from the Chinese military or accept that the Chinese military will likely find ways to circumvent any restrictions.
Today’s export restrictions force Chinese firms to use less advanced chips, and while this policy increases their costs, well-funded Chinese firms can absorb this burden. Unfortunately, the dual-use nature of AI technology makes preventing military access nearly impossible, unlike conventional arms embargoes where restricting specific materials can be effective.
The U.S. government should maintain targeted export restrictions of advanced AI technologies to countries of concern, even if these restrictions act more as hurdles than roadblocks. However, the U.S. government’s priority should be to expand the global market share of American AI firms. Current policies are undermining this objective by constraining American chipmakers and AI developers instead of supporting their growth.
A glaring imbalance is China’s restrictions on U.S. AI firms. China blocks many U.S.-based AI services, such as OpenAI’s ChatGPT, Google’s Gemini, and Meta AI. The United States should demand reciprocal market access rather than unilaterally restricting its own firms with export controls while allowing Chinese AI companies to compete freely in the U.S. market. Security concerns over Chinese AI services are valid but manageable. As a general best practice, users should avoid sharing sensitive or privileged data with untrusted providers, AI-related or otherwise.
However, DeepSeek and similar open-source AI models offer flexibility, allowing companies to deploy them on private infrastructure to mitigate data security risks. While censorship remains a legitimate concern, AI developers can fine-tune models to reduce politically motivated restrictions. Ultimately, the key issue is not security or censorship but fairness in market access.
If China refuses to allow U.S. AI firms to operate in its market, the United States should not allow Chinese AI services. Allowing DeepSeek to compete in the United States while China blocks ChatGPT creates an uneven playing field. Instead of doubling down on ineffective export restrictions, the U.S. should pursue policies that ensure American AI firms remain globally competitive.
DeepSeek’s success should serve as a wake-up call: The current U.S. strategy on China’s AI is failing and based on faulty assumptions. Policies that unnecessarily limit U.S. chip exports and restrict American AI firms from reaching legitimate foreign customers are counterproductive, helping China to develop alternatives while hurting U.S. global competitiveness. A smarter approach would be to prioritize reciprocal market access, support American AI innovation, and ensure the United States remains a leader in the global AI economy.
