Visits to Washington, D.C., this week by Nvidia CEO Jensen Huang and OpenAI CEO Sam Altman come at an absolutely pivotal moment for AI policy.

Governor Kathy Hochul’s moratorium on data centers in New York is merely the latest in a flurry of misguided regulatory ideas that could subvert America’s most dynamic economic sector.

Legislators are introducing thousands of anti-AI bills across the 50 states. The White House in June bypassed its own executive order to impose export controls on America’s leading AI model. Google DeepMind CEO Demis Hassabis just suggested a new international regulatory body. And last week, 200 economists and AI leaders, including 16 Nobel laureates, signed a letter warning of “large-scale job displacement” and urging policymakers to act “now” with undefined “guardrails.”

Two powerful new open-source AI models from China should jolt the U.S. out of its fashionable AI fears. The Kimi K3 and GLM 5.2 models from Moonshot and Z.ai nearly match the performance of America’s leading frontier models from Anthropic, OpenAI, and SpaceX—and at lower cost. China is no longer nine months behind. It has nearly caught up, and Xi Jinping just ordered full-speed ahead.

>>> Meta Abandons Its AI-Generating Tool, but Public-Safety Guardrails Are Needed

The American AI juggernaut rests on decades of light-touch, entrepreneurial policy toward information technologies. Between 1995 and 2025, the U.S. invested $3 trillion to build the Internet—fiber, broadband, cell towers, and the first two generations of data centers. Those investments transformed the economy and made America the leader in software, cloud, mobile, and Internet services.

The new AI data-center boom dwarfs that buildout. In the next five years U.S. firms are slated to invest more than $6 trillion in chips, networking, power, and structures—ten times the capital intensity of the Internet. We are short of electricity and electricians.

Contrary to the dominant narrative, AI creates both software and blue-collar jobs. By some estimates, AI data centers are driving nearly three-quarters of current economic growth.

Yet the immediate construction effects pale beside the productivity waves that could improve countless industries. We are already seeing an uptick in entrepreneurship as AI delivers superpowers to individuals and small teams.

If the U.S. abandons its successful technology policy for a bureaucratic AI “safety” mindset, it could imperil trillions in near-term investment and tens of trillions in long-term gains.

The irony is that several American AI firms are the chief advocates of heavier regulation.

Anthropic was founded to warn against the dangers of its own product. In late 2020, CEO Dario Amodei and colleagues left OpenAI because they thought AI was moving too fast and urged industry and governments to slow down. Anthropic then sped things up and recently seized the lead. Last spring it warned that its newest model, Mythos, was a dangerous cyber-weapon; the White House believed them and blocked the supposedly safer version, Fable. A month later, inexpensive Chinese open-source models may nearly match Fable’s cyber capabilities. This pattern of rapid global advance will not stop.

Anthropic protested the White House move, yet its own advertisement asks, “Who’s going to hit the brakes, if we need to?” Anthropic and the AI safety community may be getting exactly what they want: regulatory capture that cements one or two current leaders and blocks upstarts. They argue open-source AI is dangerous and ungovernable—not just Chinese models, but also U.S. innovators including Nvidia and Thinking Machines.

A monopoly or duopoly would amplify the dangers of the social-media era, when dominant Big Tech firms collaborated with government to censor and propagandize. A centralized partnership between government and Big AI, under the guise of safety, is the ultimate threat.

>>> Slowing Data Center Construction Signals the Need for Pro-Growth Reforms

AI presents real challenges for children, families, education, and labor markets; we should address them. The safety-ists, however, both overestimate AI’s harmful capabilities and underestimate its benefits. Just as true information is the best antidote to misinformation, good AI can counter bad AI. The advanced models cited as cybersecurity threats have already found and patched thousands of previously unknown vulnerabilities. When OpenAI’s internal closed-source model recently “broke containment,” Hugging Face used the open-source GLM 5.2 to repel the attack.

The best regulation is often robust competition—among open and closed models and multiple providers of apps, services, and infrastructure. Enterprises and consumers will choose providers that honor privacy, security, and economic interests.

Most data-center concerns can be solved. Hyperscalers are building their own power plants behind the meter, have agreed with the White House to hold ratepayers harmless, recycle water in closed-loop systems, and work with local communities on siting. Many of these worries may become moot as SpaceX deploys thousands of AI data centers into orbit.

In the meantime, China’s AI surge should disabuse the U.S. of its fears and unleash the next great American entrepreneurial expansion.