Musk is betting $2.8 billion that AI training is worth the public relations nightmare of becoming the infrastructure everyone hates.
The Summary
- SpaceX is spending $2.8 billion on gas turbines to power AI data centers for Grok, positioning xAI as a cloud computing player while facing carbon emission complaints.
- 71% of Americans oppose AI data centers in their area, with 48% strongly opposed. That's worse than nuclear power plants (53% opposition), and bipartisan.
- The industry's only card: money. Ben Thompson suggests the crass but obvious solution is paying residents directly.
The Signal
SpaceX isn't building rockets with that $2.8 billion. It's buying gas turbines to power data centers for xAI's Grok model, a bet that training frontier AI is worth vertical integration into energy infrastructure. This is the same playbook Meta and Microsoft are running, but with Musk's characteristic velocity. The carbon emissions will be real, the complaints already are, and the company is pushing ahead anyway.
The timing is brutal. Gallup polling shows 71% of Americans oppose AI data centers in their communities, with nearly half strongly opposed. To put that in context: that's higher opposition than nuclear power plants have ever scored since Gallup started asking in 2001.
"The anti-AI-data-center sentiment is bipartisan: There are partisan differences, but only in slight degree."
This isn't a left-right split. Republicans and Democrats both hate these facilities, which means any politician looking for an easy win can grab this issue and run with it. In our polarized era, bipartisan resentment is rare and dangerous. The hyperscalers have gotten used to regulatory capture and lobbying their way through opposition. This is different. You can't lobbyist your way out of 71% public opposition.
The industry has three problems stacking up:
- Energy demand that's stretching grids and requiring new power plants
- Carbon emissions at a moment when climate commitments were supposed to be dropping
- Local communities that see massive infrastructure, noise, water usage, and zero upside
Ben Thompson's suggestion, reported by Daring Fireball, is to just pay people. Direct payments to residents. It's crass, it's naked, and it might be the only move that works. Because right now, the value proposition to locals is: we bring electricity demand, carbon emissions, and construction disruption. In return, you get... nothing. Maybe some tax revenue that goes to the county. Maybe a few dozen jobs that aren't for you.
The SpaceX spend shows how seriously xAI is taking the infrastructure game. $2.8 billion is bet-the-company money for most AI startups. For Musk, it's table stakes to compete with hyperscalers who own their power generation. The move positions xAI as a cloud computing player, not just a model shop. That's the real signal: the AI companies that win Web4 will be the ones that controlled the full stack, from silicon to electrons.
But if public opposition hardens into regulation, all that vertical integration hits a wall. And unlike crypto, where the industry could route around hostile jurisdictions, you can't train models in international waters.
The Implication
Watch for politicians to discover this issue in the next election cycle. 71% bipartisan opposition is a gift to anyone running against Big Tech. The smart AI companies will start cutting deals with local communities now, before they're forced to. Direct payments, revenue sharing, community solar, something that puts money in residents' pockets.
For everyone building in the agent economy: compute availability is about to get political in ways it hasn't been since net neutrality. If you're betting on cheap, abundant cloud training, factor in the regulatory risk. The turbines SpaceX is buying might get built. The permits for the next thousand data centers are less certain.