DeepSeek's January 2025 release of an open-weight reasoning model trained for under $6 million triggered the largest single-day market-value loss in U.S. equity history, wiping roughly $589 billion from NVIDIA's market capitalization. The shock wasn't just about the price tag; it revealed a fundamental asymmetry in how the United States and China approach artificial intelligence competition, and why traditional arms-control frameworks may be ill-suited to address the real strategic challenge ahead.

What Makes DeepSeek's Achievement So Threatening to U.S. Dominance?

DeepSeek's R1 model demonstrated that cutting-edge AI reasoning capabilities don't necessarily require the massive infrastructure investments the U.S. tech industry has been making. The model was reportedly trained using Western model outputs, despite prohibitions on their use in China. This matters because it suggests that China may be circumventing export controls and leveraging publicly available AI research to accelerate its own development at a fraction of the cost Western companies incur.

The timing amplified the shock. On January 21, 2025, President Trump announced Stargate, a $500 billion AI infrastructure partnership between OpenAI, SoftBank, Oracle, and MGX. Just six days later, DeepSeek's release sent markets into a tailspin, raising uncomfortable questions about whether the U.S. approach to AI competition through massive capital expenditure was the right strategy.

How Does China's AI Strategy Differ From America's Approach?

The fundamental difference lies in how each country views the relationship between regulation and competitive advantage. The United States, under the Trump administration, has publicly committed to winning what it characterizes as an economic and technological race. In February 2025, Vice President J.D. Vance told the Paris AI Action Summit: "I'm not here this morning to talk about AI safety. I'm here to talk about AI opportunity." Three months later, the Department of Commerce rescinded the Biden-era AI Diffusion Rule two days before its effective date, arguing that the rule "would have stifled American innovation".

Vance

China, by contrast, has pursued a dual strategy of international cooperation rhetoric paired with domestic protectionism. In July 2025, Premier Li Qiang opened the World Artificial Intelligence Conference in Shanghai with a Global AI Governance Action Plan and a proposal to establish a "World AI Cooperation Organization" (WAICO). This institutional architecture parallels the British and American AI Safety Institutes that have organized Western frontier-model evaluation since 2023, but with a crucial difference: Beijing sets the rules.

Why Does China Prefer International AI Regulation?

China has strong incentives to push for international arms control on AI technologies for two strategic reasons. First, Beijing intends to cheat and use regulatory frameworks to create differentiated advantage. China understands that in the United States and allied countries, the rule of law generally stands behind the enforcement of treaties domestically, so regulations will be obeyed in America. But the Chinese Communist Party will ensure such regulations are not obeyed in China. This dynamic mirrors historical precedents, including German implementation of the Versailles Treaty and Soviet nuclear arms control during the Cold War.

Second, China correctly understands that it is behind in several critical areas, most especially in the development of foundation models and the hardware to run advanced AI models. Slowing everyone else down while China tries to speed up is its net advantage. As one analysis noted, "Slowing everyone else down, while China tries to speed up, is its net advantage".

What's the Fundamental Problem With Treating AI Like Traditional Weapons?

Both the United States and China released formal statements of intent on AI in late July 2025, but they share a critical category error: treating AI as a class of weapons amenable to arms-control's count-and-cap verification logic, when AI is actually a general-purpose enabling capability. The real strategic competition is over industrial-base inputs, not the models themselves.

The strategic logic of AI competition centers on control over inputs rather than outputs. This includes:

• Compute Power: Access to the processing capacity needed to train and run advanced AI models, which requires enormous electrical infrastructure and cooling systems.
• Semiconductor Manufacturing: Control over fabrication plants (fabs) that produce the specialized chips required for AI training and inference.
• Talent: Recruitment and retention of AI researchers, engineers, and domain experts who understand frontier model development.
• Energy: Reliable, affordable electricity to power data centers and training operations at scale.

Traditional arms-control machinery was never built to verify or constrain competition over these inputs. A treaty limiting AI model parameters or capabilities would be nearly impossible to verify, and even if verified, would not address the underlying industrial-base competition that drives strategic advantage.

How to Understand the Geopolitical Stakes in AI Competition

The U.S. and Chinese approaches to AI reflect fundamentally different preferences for the organization of the global world order. After the Cold War, U.S. policymakers largely assumed that economic liberalization would result in Beijing's political liberalization, but this optimism proved misplaced. China is investing in military and technological capabilities to pursue political goals that are fundamentally at odds with U.S. ones, including planning an invasion of Taiwan.

Key strategic considerations include:

• Revisionist Intent: China views the current international order as one it seeks to reshape, while the U.S. seeks to maintain and strengthen the existing liberal order.
• Technology as Power Multiplier: AI has the potential to significantly amplify a state's power for both economic and warfighting reasons, making it a central focus of great-power competition.
• Enforcement Asymmetry: China has a consistent history of ignoring or avoiding the regulatory downsides of bilateral and multilateral agreements, accepting benefits while leveraging national power to prevent enforcement actions.

The 2026 National Defense Strategy claims that "a decent peace, on terms favorable to Americans but that China can also accept and live under, is possible." However, the Trump administration cannot decide for China whether such a peace will emerge. The administration can pursue programs and advance policies that deter China and frustrate its revisionist designs, but whether China is deterred ultimately depends on the Chinese Communist Party's calculus.

DeepSeek's breakthrough serves as a wake-up call: the real competition is not about negotiating limits on AI capabilities, but about winning the race for the industrial inputs that make advanced AI possible. Until U.S. policymakers grapple with this reality, regulatory frameworks and arms-control proposals will likely miss the mark.