The safety-first rebels who left OpenAI over ethics concerns just filed to go public on a business model that depends on burning methane at scale.
The Summary
- Anthropic filed S-1 documents with the SEC revealing that its data centers are "increasingly reliant" on natural gas, with intense competition for power supply threatening the buildout
- The company was founded in late 2020 by senior OpenAI employees who left over concerns about inadequate prioritization of AI safety and ethics
- The IPO filing exposes a fundamental tension: the company built on safety principles is now admitting vulnerability to fossil fuel costs and supply constraints
The Signal
Anthropic's S-1 filing is the first time a major AI company has laid bare the infrastructure reality behind the agent economy in regulatory documents. The filing acknowledges that "competition for access to electrical supply and data center locations with adequate power availability is intense," while admitting data centers lean heavily on what the fossil fuel industry calls natural gas but is actually methane.
This matters because Anthropic isn't some crypto mining operation or web hosting provider. This is the company founded by OpenAI defectors specifically because they thought their former employer wasn't taking safety seriously enough. They structured as a public benefit corporation. They talked about constitutional AI and responsible scaling policies. And now their SEC filing reveals they're locked into the same energy trap as everyone else racing to train frontier models.
"The irony is sharp: you can't build safe AI if you can't build AI at all, and right now building AI means burning gas."
The S-1 also hints at a catch-22 that The Verge notes: rising global temperatures driven partly by fossil fuel use could further strain the energy infrastructure AI companies depend on. Climate change makes power grids less reliable. Less reliable grids make data center buildouts riskier. Riskier buildouts slow AI development or push costs higher. The snake eats its tail.
What the filing doesn't say is as important as what it does. There's no mention of breakthrough deals with renewable providers. No partnerships with next-gen nuclear startups. No hedging strategy beyond acknowledging the risk exists. Either Anthropic doesn't have better options lined up, or they're not material enough to disclose yet.
Key risks disclosed:
- Intense competition for power supply and data center locations
- Vulnerability to rising natural gas costs
- Supply chain constraints for fossil fuel delivery
- Potential climate impacts on grid reliability
Bloomberg is hosting a live Q&A about what the IPO reveals, which suggests the financial press sees this filing as a rare window into frontier AI economics. Most of these companies stay private as long as possible. Anthropic going public now means investors get to see the full cost structure of competing with OpenAI and Google, including the energy bill.
The Implication
If you're building agents or training models, this S-1 is your wake-up call about infrastructure costs. Power isn't getting cheaper or easier to source. The companies that win Web4 will be the ones that solve energy access, not just model architecture. Watch for AI companies striking deals with utilities, co-locating with power plants, or investing directly in generation capacity.
For investors, Anthropic's IPO will set the valuation benchmark for how markets price energy risk into AI companies. If the stock trades at a discount to OpenAI's private valuation or Google's AI multiples, energy dependency is the reason. That spread tells you what the market thinks this constraint is worth.