The US-China AI Race: Collaboration or Confrontation?

The US-China AI Race: Collaboration or Confrontation?The United States is increasingly fixated on "winning" the AI race against China, a strategy that may overlook the potential benefits of collaboration and trigger unforeseen consequences. Recent export controls implemented by the Biden administration, restricting the global spread of AI technology to prevent cutting-edge advancements from reaching China, continue a policy of AI containment pursued under both Trump and Biden

The US-China AI Race: Collaboration or Confrontation?

The United States is increasingly fixated on "winning" the AI race against China, a strategy that may overlook the potential benefits of collaboration and trigger unforeseen consequences. Recent export controls implemented by the Biden administration, restricting the global spread of AI technology to prevent cutting-edge advancements from reaching China, continue a policy of AI containment pursued under both Trump and Biden. Simultaneously, prominent figures in the AI field, such as OpenAI CEO Sam Altman and Anthropic co-founder Dario Amodei, have publicly called for defeating China, creating an environment ripe for further escalation by future US administrations.

Paul Triolo, a partner at the DGA Group, a member of the Council on Foreign Relations, and senior advisor to the University of Pennsylvanias Future of US-China Relations project, and Alvin Grewal, a seasoned entrepreneur with extensive experience in the Chinese market for a Taiwanese electronics company, have closely observed the development of China's AI industry and the impact of US sanctions. They engaged in an email exchange to discuss the new US AI diffusion rules, Silicon Valley's stance on these policies, and the risks of viewing AI as a zero-sum game. Their discussion reveals the global dangers of overemphasizing competition and calls for a reassessment of the potential for cooperation in AI development. The following is a detailed summary of their conversation:

I. Interpretation and Impact of the New US AI Diffusion Rules

Q: What are your thoughts on the US governments new AI diffusion rules this week, aimed at restricting Chinas access to advanced AI technology?

Triolo: The rules primarily target high-performance computing clusters and restrict proprietary weights for the most advanced frontier models. However, the specific definition of model performance levels remains unclear. Furthermore, most open-weight (freely shared) AI models are typically adjusted and optimized by users (including major Chinese AI companies) to suit their specific needs. This complex and ambiguous regulation creates significant uncertainty for large and mid-sized hyperscale companies in the US and the West. For hyperscalers like Google, Microsoft, Amazon, and Oracle, this could lead to slowed or complicated international expansion, increased compliance and legal costs, disrupted global R&D programs, and unclear enforcement requirements.

II. Impact of Previous Sanctions on China's AI Industry

Q: What has been the impact of previous policy measures, including sanctions implemented under the first Trump administration, on China's AI industry?

Triolo: US export controls have undoubtedly slowed the pace of China's AI industry to some extent. However, on a broader level, these sanctions have spurred greater government focus on unity and self-reliance. In response, the Chinese government has invested tens of billions of dollars, supporting domestic companies in catching up or expanding capabilities in key technologies. This effort has driven significant changes in the semiconductor industry, improving China's ability to develop the advanced hardware necessary for cutting-edge AI models. Chinese AI developers have shown remarkable adaptability, leveraging traditional AI hardware from Western companies while progressively integrating domestic alternatives into their development processes. While perhaps not matching the pace of their Western counterparts, the continued innovation momentum in Chinese AI hardware and software remains significant.

III. Analysis of the "Defeat China" Narrative in Silicon Valley

Q: Why do you think so many people in Silicon Valley are now talking about the need to "defeat China" in AI?

Triolo: A significant factor is the increasing alignment between conservative venture capitalists (concentrated primarily in Silicon Valley) and technology companies that rely on exaggerating the China threat to build their business models. This is concerning because it conflates the China threat, personal interests, and opposition to advanced AI regulation. More critically, this narrative frames the US-China competition in AI as a zero-sum game, an extremely dangerous perspective.

The Trump administration was heavily influenced by this zero-sum thinking, with tech leaders and venture capitalists playing a key role. Leading AI figures, such as OpenAIs Altman and Anthropics Amodei, who have close ties to the US government, have reinforced narratives about the Chinese AI threat and promoted the zero-sum competition through articles, commentary, and company statements. However, not everyone supports this position; Elon Musk, for instance, has consistently advocated for international cooperation and responsible AI governance, highlighting that AI safety is a global challenge requiring collaborative efforts, including collaboration and exchange with China.

IV. The Reality and Risks of US AI Leadership

Q: Shouldn't the US want to beat China in AI?

Grewal: This mindset suggests that any collaboration between the US and Chinaeven in win-win areas like medicine and energy developmentshould be avoided to maintain US and allied dominance in AI development. However, no nation can ensure continued dominance over another, let alone win such a race.

The core logic driving US AI policy toward China rests on assumptions about future conflict. While AI may play a significant role in economic and military spheres, this view remains poorly defined. US officials have thus far struggled to articulate precisely what constitutes the alleged significant military uses of AI.

On the other hand, decades of experience demonstrate the critical role of collaborative research in fundamental AI advancements. Nearly half (47%) of top AI researchers globally were born or educated in China. Disrupting this mutually beneficial cycle carries significant risks and could be counterproductive. The current trend seems to be towards isolation. Another significant risk is Chinas potential withdrawal from multilateral efforts to establish AI governance frameworks. This would make creating a viable global AI safety system even more elusive.

Q: How long has the US actually been ahead of China in AI?

Grewal: There was a widespread belief in Silicon Valley that the US led China in AI development by one to two years. However, that lead appears to be shrinking. Several popular Chinese models for generating videos and images have reached performance levels comparable to leading US products. In particular, the recently released DeepSeekr1 and v3 models have surpassed OpenAIs o1 inference model and GPT-4o model, respectively, in performance, suggesting a rapid narrowing of the gap.

Furthermore, as AI computation shifts from training to inference (the computation required when running a model, not training it), the reliance on chips manufactured using the most advanced semiconductor nodes decreases. This somewhat mitigates the impact of US sanctions on China. Notably, DeepSeek was founded only in 2023, has fewer than 100 employees, and has significantly fewer computing resources than top Western labs. The US governments technology restrictions on China might inadvertently be fostering greater creativity and more efficient resource utilization among Chinese developers. China may also have an advantage in accessing the private data needed to train frontier AI models.

However, a worrisome trend is emerging as countries restrict cross-border data flows: the potential future of 193 highly intelligent but culturally and value-bound sovereign AI models. This might be good news for Nvidia's stock price, but for humanity as a whole, it could be a dystopian nightmare.

V. Recommendations for Future US Administrations

Q: What advice do you have for the incoming president and their administration?

Triolo: The increasingly intense AI competition between the US and China poses a serious threat, not only to both countries but to the entire world. This competition could endanger global peace, economic stability, and technological progress. Therefore, I recommend that both the US and China re-evaluate and adjust their strategies towards AI development, moving away from viewing AI solely as a tool for military competition.

Both sides should establish robust dialogue mechanisms, collaborate on setting international standards for AI governance, and advocate for the creation of a global AI safety alliance. They should also agree on the responsible use of AI and jointly develop tools to monitor and curb the global misuse of AI. Furthermore, governments should support and incentivize transnational academic and industrial collaborations. A global AI collaboration project similar to CERN would create far greater value for all of humanity and could lead to peaceful solutions. The AI Manhattan Project advocated by many today might lead down a drastically different path, bringing more unnecessary conflict and risk. The choice we face is seemingly simple yet critical: continue down the road of confrontation that almost certainly leads to a lose-lose outcome, or turn towards cooperation to create a prosperous and stable future for all.