The US-China AI Race Has a Paradox: One Side Wants to Win, the Other Wants to Slow Down

China is moving aggressively to dominate global AI development by the end of the decade, backing that ambition with coordinated state funding and resources that its one-party system makes easier to mobilize. While Chinese leadership is known for bold rhetoric, officials genuinely view AI as tied to the survival of their political system, and they're investing accordingly. The result: China is closing the gap with the United States faster than many expected. At the same time, the American AI industry is sending a very different signal — one of caution. US developers have voiced concern that the technology they're building could eventually slip beyond human control, with consequences serious enough to threaten humanity itself. Whether AI actually carries that level of destructive potential remains fiercely debated and unresolved. Still, the US and China find themselves in a race that echoes the nuclear standoff of the Cold War — a rivalry so dangerous it eventually pushed adversaries to negotiate shared rules, even while both sides retained the capacity for mutual destruction. Two Countries, Two Very Different Calculations Both nations are wrestling with how to balance regulation against the drive for technological superiority, and each holds distinct advantages in the competition. Meanwhile, the broader AI industry appears to be approaching a pivotal, though still loosely defined, moment that could force a reckoning over how much power these systems should be allowed to have. Washington has tried to preserve its lead by restricting China's access to top-tier AI chips, a move that created real obstacles for Chinese developers. But Chinese firms responded with an unexpected workaround. David Bellman, product lead at data intelligence firm SynMax, said that without access to the most advanced chips, Chinese companies focused instead on making their software more efficient and powerful. That strategy paid off. By February 2025, China's DeepSeek model trailed OpenAI's leading system by less than half a percentage point in benchmark testing — a dramatic improvement from earlier generations. Since then, Chinese and American models have remained closely matched. Stanford University's 2026 AI Index report concluded that the performance gap between US and Chinese models has effectively disappeared. Andrew Chin, an AI law professor at the University of North Carolina, noted that the US still holds significant advantages in frontier-model development, private investment, and data center infrastructure — meaning which country is truly "ahead" depends heavily on which metric is used. China's All-In Strategy — and Its Limits China isn't backing down. In 2025, it introduced its New Generation Artificial Intelligence Development Plan, a state-driven roadmap leveraging government agencies and party-run tech firms to build infrastructure and train large language models, with the broader goal of establishing dominance in fields like batteries, robotics, and automation. Researchers at the University of Virginia's National Security Data and Policy Institute wrote in a white paper that China's centralized planning model has allowed it to rapidly build out sovereign technology and extend its global influence. But that same model carries drawbacks. The UVA researchers noted that China's approach to AI invites political favoritism and embeds bias into its systems, largely because Beijing's core concern isn't human extinction — it's the survival of Communist Party control. As a result, China's AI regulations focus heavily on limiting what information AI systems share with users and reinforcing the party's political authority. That anxiety surfaced clearly in an article published Sunday by China's State Security Minister Chen Yixin in the state-run China Cyberspace magazine. Chen warned that China must retain dominance over AI or risk losing control over its own population, cautioning that the US could exploit the technology for cybersecurity or espionage advantages, or use it to wage a propaganda campaign against the party. Is the US Industry's Caution Genuine? American AI companies have at least publicly embraced the idea of self-regulation, with executives meeting with lawmakers to discuss oversight. President Trump, however, has dismissed those efforts as capitulating to unfounded fears — and warned that slowing down would hand China a strategic advantage. It remains uncertain how committed the industry actually is to pacing its own development, or how seriously it believes AI could cause catastrophic harm. Skeptics point out that American AI firms still face major obstacles in acquiring the computing power needed for truly dangerous capabilities. The country's data center expansion has fallen well behind the industry's ambitions, hampered by labor shortages, limited building materials, chip scarcity, permitting delays, rising interest rates, and local opposition to new facilities. Brookings Institution senior fellow Darrell West said these are fundamental obstacles that could undermine tech companies' business models entirely. Still, no one can say with certainty how much computing power would actually be required for AI to pose a catastrophic threat. Carnegie Mellon computer science professor Vincent Conitzer said anyone claiming a specific safety threshold is being dishonest, adding that today's existing computing capacity might already be sufficient if combined with the right techniques. That uncertainty is precisely why regulation advocates argue oversight efforts shouldn't wait for an agreed-upon danger threshold. Chin said infrastructure delays might buy time, but breathing room isn't the same thing as actual governance. Amodei's Proposal — and China's Rejection of It Anthropic CEO Dario Amodei argued in a weekend essay that the US must maintain its AI advantage over China, but that American developers shouldn't accelerate development faster than necessary to stay ahead. Doing so, he argued, would give the industry time to ensure safety while creating leverage to eventually bring China to the table for global AI regulation talks. Amodei warned that slowing down too much would let Chinese Communist Party-linked projects overtake the US, creating serious national security risk — but argued that maintaining a lead, rather than complicating relations with China, actually makes cooperation more likely down the road. China's state-run Global Times dismissed the proposal in an editorial, framing it as a "Cold War playbook" aimed at containing Chinese development. A Shared, Unspoken Fear Despite the public friction, China may share — or even exceed — America's underlying worry about AI systems acting beyond anyone's control. Philip Potter, executive director of the University of Virginia's National Security Data and Policy Institute, said Beijing likely harbors serious concerns about AI's potential to destabilize its own political and social order. That shared anxiety, Potter suggested, could open the door to cooperation — arrangements where both governments restrain private-sector AI development while treating advanced AI research as a tightly controlled state and national-security function. The alternative, according to Chin, is a self-reinforcing arms race in which neither country can afford to prioritize safety if each assumes the other won't. He acknowledged a legitimate national-security interest in preventing rivals from gaining a decisive technological edge — but said there's an equally legitimate interest in avoiding the creation of systems whose behavior and consequences no one can reliably predict or control.