The Politics Beneath the AI Race
The Politics Beneath the AI Race
The rivalry between the United States and China over artificial intelligence is often described as a contest of speed and ingenuity. Commentators count patents, compare benchmarks, and track investment flows. These measures are not trivial, but they miss the underlying structure. The decisive factor is not who invents faster but who sustains discipline. America has treated AI as a trade lever, subject to the contingencies of negotiation. China has treated it as an arm of state strategy. This divergence explains why the United States risks decline even as its firms still lead in several domains.
The United States began from a position of strength. In the early 2020s, American firms built the most advanced models, funded by venture capital, supported by universities, and encouraged by permissive regulation. By 2023, GPT and Gemini outperformed Chinese models by wide margins. U.S. cloud providers held more than half of the global market. Yet by 2024, Chinese firms such as DeepSeek and Qwen had closed the gap on accuracy and scale. Export controls on chips were meant to slow this advance but proved porous. Beijing relied on shell companies, stockpiles, and its own semiconductor programs. Engineers improved software to maximize weaker hardware. The episode revealed the limits of a policy built on restriction rather than coherence.
China’s progress followed a different course. The state linked national planning to technological goals. The Next Generation AI Development Plan directed resources to data centers, workforce training, and semiconductor manufacturing. Local governments experimented with AI in public services. Firms integrated AI into production: Xiaomi used more than 700 AI-guided robots to produce a new car every seventy-six seconds. These developments showed more than technical ambition. They expressed an ability to align state priorities with industrial practice, binding innovation to strategy.
American policy, by contrast, drifted toward short-term bargaining. In 2025, the Trump administration reversed an export ban and allowed Nvidia to sell its H20 inference chip to China. The H20 was faster than earlier models and expected to generate billions in sales. Officials presented the reversal as part of a trade truce over rare-earth minerals. Beijing described it as a concession given without reciprocity. The effect was to signal that national security was negotiable. Analysts warned that China could now use rare-earth leverage to obstruct future controls. The United States had traded a durable advantage for temporary relief.
The pattern extended further. Trump proposed tariffs of up to one hundred percent on semiconductor imports from allies, while imposing a fifteen percent export duty on U.S. chip sales to China. Industry groups argued this would raise costs, weaken competitiveness, and deter investment. South Korea warned it would slow American progress and reduce global competitiveness. More damaging was the signal to allies. Export controls had worked only when multilateral. By attaching fees and exceptions, the administration showed that security measures could be monetized. As one analyst concluded, if a fifteen percent fee could override security concerns, partners had little reason to trust U.S. leadership.
These choices revealed a deeper asymmetry. China advanced by treating AI as infrastructure, embedding it in a national strategy. The United States faltered by treating AI as one more asset to exchange in trade disputes. One path builds resilience; the other invites erosion. Yet alternatives exist. American policymakers have proposed standards for transparency, accountability, and cost in AI models. They have considered systems to reduce the difficulty of shifting between platforms, and adjudication frameworks to compare outputs across models. Such measures could make U.S. systems attractive even if Chinese firms set the performance frontier. They would not guarantee supremacy but would establish coherence and predictability.
The underlying lesson is that technological races are not won by invention alone. They are decided by the ability of states to hold policy steady across cycles of pressure. China has advanced by integrating its industrial and strategic aims. The United States has weakened itself by converting security into negotiation. To win, or even to lose well, requires coherence. Without it, technical advantages will continue to erode. In this race, superiority belongs not to the country that trains the largest model but to the one that sustains the most durable strategy.