The AI CEOs want deregulation, except the ones who want their competitors regulated.
The Summary
- Trump hosted Nvidia's Huang, Meta's Zuckerberg, Google's Pichai, and Anthropic's Amodei at the White House, where they pushed overlapping but strategically divergent policy asks
- Surface consensus: more data center capacity, faster permitting, federal AI spending, light regulation
- The split: Anthropic wants frontier AI regulation while fighting Pentagon restrictions; others want open season on chips and models
The Signal
The White House meeting exposed the fundamental tension in AI policy: everyone wants the government's help building infrastructure, but nobody agrees on who should face constraints. Anthropic is pushing a "regulatory ladder" that would tighten oversight as AI systems grow more capable, including external testing and incident reporting for frontier models. Translation: regulate the big guys, which conveniently includes their competitors at OpenAI, Google, and Meta.
This isn't altruism. Anthropic's "Advanced AI Framework" gives governments authority to block high-risk deployments. If you're behind in the race, slowing down the leaders is strategy. If you're ahead, any regulation is friction.
"Anthropic wants oversight on frontier AI, which means oversight on everyone racing past them."
Meanwhile, the infrastructure asks unite everyone. More data center capacity means more compute, which means faster training, which means staying competitive. Faster permitting means building in months instead of years. Federal spending on AI means subsidized infrastructure they'll all use. These aren't policy positions, they're shopping lists.
But the cracks show in the details:
- Chip export restrictions matter differently if you make chips (Nvidia) versus consume them (everyone else)
- Frontier model regulation hits differently if you're releasing new models weekly versus quarterly
- Pentagon contracts matter more to some balance sheets than others
Anthropic's fight with the Pentagon is the wildcard here. After refusing to drop restrictions on mass domestic surveillance and fully autonomous weapons, they got designated a supply-chain risk in March. A federal appeals court upheld it in September. Amodei walked into that meeting carrying baggage his competitors don't have.
This matters because it reveals the real game. Anthropic is trying to reframe safety concerns as regulatory moats. OpenAI and Google want to move fast and explain later. Meta wants open models and no liability. Nvidia just wants everyone buying more H100s. They all showed up asking for the same things on paper, but they're playing different games.
The Implication
Watch who gets what. If Anthropic's regulatory ladder gains traction, it will slow frontier model development for everyone, which helps whoever is currently behind. If chip export restrictions loosen, Nvidia wins and China gets faster access to cutting-edge compute. If data center permitting accelerates without environmental review, we'll see a construction boom that makes the last cloud infrastructure build look quaint.
The real tell will be whether Anthropic's Pentagon problem gets quietly resolved. If it does, that's the price of admission to this conversation. If it doesn't, Amodei just spent political capital he can't afford to waste.