Anthropic filed for its IPO on Monday, and the S-1 tells the story of an AI company wrestling with contradictions. Revenue is up, losses are mounting, and the whole operation runs on methane.

The company reported enterprise revenue growth that would make most SaaS founders jealous. More than half of Anthropic’s customers paying over $100,000 annually now use both the Claude API and a first-party app like Claude.ai or Claude for Work. That’s the kind of cross-selling that makes customers sticky and churn rates low. It’s also straight out of the Amazon playbook: get them using multiple products, make it harder to leave.

But the S-1 also makes clear what it takes to deliver that experience. Anthropic is burning cash and natural gas in roughly equal measure.

The methane problem

“Competition for access to electrical supply and data center locations with adequate power availability is intense,” the filing states. That’s corporate-speak for: we can’t get enough electricity, so we’re using whatever we can find.

What they’re finding is natural gas. The filing acknowledges that Anthropic’s data centers are “increasingly reliant” on the fossil fuel. The company lists energy access as a material risk factor. If they can’t secure adequate power, they can’t train models. If they can’t train models, the business doesn’t work.

This isn’t unique to Anthropic. Every AI company is hitting the same wall. But it’s particularly uncomfortable for a company that built its brand on responsible AI development. The S-1 devotes pages to AI safety commitments and governance structures designed to preserve the founders’ mission. Then it admits the mission requires industrial quantities of methane.

The filing doesn’t provide exact figures on energy consumption or carbon emissions, which is part of a broader pattern of omissions.

What’s missing

For a document that runs hundreds of pages, Anthropic’s S-1 leaves out quite a bit. Key financial metrics are missing or redacted. Specific customer counts, retention rates, and unit economics are absent. One analyst told WIRED that “Anthropic owes investors more information.”

The company did reveal its unusual governance structure. Like many founder-controlled companies going public, Anthropic is keeping decision-making power with insiders. But the rationale here isn’t about vision or control in the typical sense. It’s about preserving the AI safety mission even after public market pressures kick in.

Whether that works depends on whether investors believe the mission is compatible with the business model. Right now, the business model requires massive capital expenditure, uncertain energy access, and tolerance for significant losses in pursuit of model improvements.

The SpaceX comparison

On the same day Anthropic filed its S-1, TechCrunch reported that SpaceX locked in a $40 billion GPU financing deal with Apollo. The structure is telling. SpaceX is funding its AI ambitions without relying solely on traditional debt markets. It’s using specialized financing arrangements that treat GPUs as assets worth securing against.

That’s the kind of capital structure you build when you expect AI infrastructure to require sustained, heavy investment over many years. It’s also what you build when you think traditional financing won’t cut it.

Anthropic is going public. SpaceX is taking on $40 billion in GPU-specific debt. Both approaches reflect the same reality: AI at this scale requires new financial models.

What enterprise adoption actually means

Anthropic’s cross-selling success matters because it points to how enterprise AI buying is evolving. Companies aren’t just picking an API and building on top of it. They’re adopting multiple products from the same vendor, which means they’re making bigger bets on specific providers.

That’s good for Anthropic’s retention metrics. It also means switching costs are rising across the industry. If half your team is using Claude.ai and your developers are building on the Claude API, moving to a competitor gets complicated fast.

The S-1 positions this as a strategic advantage, and it is. But it also raises questions about lock-in and what happens when these companies need to raise prices to cover their energy and infrastructure costs.

The filing doesn’t answer those questions. It doesn’t provide enough detail to model out how unit economics improve as the company scales, or whether they improve at all. What it does show is a company growing revenue, losing money, and betting that enterprise customers will stick around long enough for the math to work.

Right now, that bet depends on access to electricity. And access to electricity depends on natural gas. The AI boom runs on methane, and Anthropic’s S-1 makes that dependency impossible to ignore.