Anthropic's IPO prospectus by the numbers

After months of speculation, Anthropic finally released its IPO prospectus on Monday. The company could raise more than $100 billion at a valuation approaching $2 trillion, though neither figure is confirmed.

But buried in the filing is another number that tells us a lot about whether Anthropic can actually make the economics of AI inference work.

The key numbers behind Anthropic's IPO

Anthropic generated $2.08 in revenue for every dollar spent on compute and infrastructure in Q2 2026, up from just $0.40 in Q1 2025.

Over those five quarters, revenue surged nearly 29-fold to $11.6 billion, while compute and infrastructure expenses increased roughly sixfold to $5.6 billion.

That's different from the roughly 85% gross margin on customer inference I mentioned above. This broader compute number also includes spending on research, model development, and internal workloads.

The trend is promising, nonetheless. Revenue is growing much faster than Anthropic's overall compute bill, suggesting the company is beginning to benefit from economies of scale.

Products like Claude Code are helping Anthropic squeeze much more revenue out of the infrastructure it has already committed to.

What Anthropic's IPO means for investors

If SpaceX's blockbuster IPO is anything to go by, Anthropic's debut could be massive. But attracting investors and building a sustainably profitable business are two very different things.

Anthropic remains deep in the red on a net income basis. The company reported an $80.7 billion net loss in the first half of 2026, though much of that reflected accounting adjustments tied to its rising valuation.

Operationally, the picture looks much better. Anthropic lost $693 million at the operating level in the first half, but actually generated nearly $1 billion in operating profit in Q2 alone.

That means Anthropic is getting dramatically more efficient at turning compute into revenue, but it has also locked itself into hundreds of billions of dollars of future compute commitments.

📌 Bottom line: The trillion-dollar question is whether Anthropic can keep scaling fast enough to cover the enormous amount of compute it has committed to buying.