Inside Anthropic's S-1: How It Lost $80.7B While Turning an Operating Profit

Anthropic's long-awaited S-1 is finally public, and the numbers look very different from the ones we examined just a month ago.

‍

In September, we looked at Anthropic's confidential IPO prospectus. The company had generated $4.59 billion in revenue in 2025, but spent considerably more than that on compute alone. Its $42 billion net loss also raised questions about how much money Anthropic was actually losing from running its business.

‍

The new filing offers some answers. Anthropic generated $16.29 billion in revenue during the first half of 2026 and recorded $974 million in operating profit in Q2. Yet it also reported an $80.75 billion net loss over the six-month period.

The question now is not whether Anthropic can make frontier AI profitable, but whether it can sustain that profitability while committing hundreds of billions of dollars to infrastructure.

Anthropic's revenue is growing faster than its compute costs

For much of 2025, Anthropic had a fairly obvious problem. Running Claude was enormously expensive, and the company wasn't bringing in enough revenue to cover those costs. That's beginning to change.

‍

In 2025, Anthropic spent approximately $1.60 on compute and infrastructure for every dollar of revenue. During the first half of 2026, that figure fell to $0.55. Revenue also reached $11.56 billion in Q2 alone, more than double the previous quarter.

‍

‍

It's an impressive improvement, and there's a reasonable argument that Anthropic is beginning to benefit from the scale it has spent years building. However, the compute-to-revenue ratio doesn't tell us how much each Claude request costs, or whether those economics will hold up as more infrastructure comes online.

‍

There's also the question of distribution costs. Anthropic records substantial sales, marketing and partnership expenses separately from compute, and some of those costs may be directly related to bringing its products to customers.

‍

Still, covering its operating expenses in Q2 is encouraging, although whether it can sustain that performance as infrastructure spending grows remains to be seen.

So why did Anthropic lose $80.7 billion?

As we explained in September, Anthropic raised money through convertible notes and other instruments that can eventually turn into shares. Some are recorded as liabilities, and their values increase as Anthropic's valuation rises. Those increases can then appear as losses in its income statement.

‍

In H1 2026, these fair-value adjustments amounted to $72.39 billion, accounting for nearly 90% of its reported net loss. It's a peculiar situation where a company's rising valuation can make its reported losses look significantly worse.

‍

Those charges shouldn't be dismissed entirely, though. The financing instruments represent real economic claims on Anthropic, and their eventual conversion into shares could dilute existing shareholders. But they're very different from the expenses involved in developing models and serving Claude customers.

Anthropic is making a very expensive bet on future demand

The biggest risk in Anthropic's filing may have less to do with its current losses than with the amount of infrastructure it has committed to buying.

‍

Its previous prospectus disclosed approximately $518 billion in future compute and infrastructure commitments. The new filing details around $417 billion in minimum obligations from new and amended agreements, alongside another $125 billion to $150 billion in expected payments for additional capacity. Some amounts overlap with earlier disclosures, and some agreements are still being finalized.

‍

These are enormous sums, especially when you consider that Anthropic is committing to capacity years before it knows exactly how customers will use its models. But waiting until demand materializes could be just as expensive, because the company risks falling behind competitors with more computing capacity. Customers could face usage limits, new models could take longer to develop, and competitors could capture demand Anthropic cannot serve.

‍

Committing early creates its own problems. AI models keep getting cheaper to run, competitors keep releasing alternatives, and nobody knows what customers will be willing to pay three or five years from now. Anthropic could get its demand forecasts right and build a highly profitable business, or find itself paying for capacity at prices that no longer make sense.

‍

There's also a risk for infrastructure providers on the other side of these contracts. Companies building data centers and supplying chips are relying on AI labs to keep spending at extraordinary levels. If demand disappoints, Anthropic wouldn't be the only company affected.

What would have to go right for Anthropic?

Anthropic has demonstrated that it can generate an operating profit, and there's reason to be optimistic. But maintaining that performance will depend on customers continuing to spend more on Claude, the company retaining enough pricing power, and revenue growing fast enough to support its infrastructure commitments.

‍

However, sustaining those improvements could become more difficult as competition increases. Customers have more AI models to choose from, putting pressure on pricing, while developing frontier models continues to require enormous investment.

‍

Anthropic's Q2 profit is encouraging, but there's no guarantee its margins will hold up as more of its infrastructure commitments come into effect. The next few quarters should give us a better idea.