Anthropic's Leaked S-1 Shows an $80.7 Billion Loss. $72 Billion of It Is Non-Cash. Here's the Number a Solo Builder Should Actually Watch.

The Wall Street Journal got a look at Anthropic's draft S-1 this week, and the reported numbers are loud: $16.3 billion in revenue for the first half of 2026, up from $1.1 billion a year earlier, against a net loss of $80.7 billion. Read that fast and you conclude the vendor behind your Claude bill is on fire. Read the second sentence of the coverage and the picture changes. About $72 billion of that loss is reportedly non-cash accounting adjustments tied to convertible notes and other financing instruments.

None of this is confirmed by Anthropic. Everything below comes from reporting on a preliminary draft, and drafts change before anything is filed publicly. I wrote about the IPO route on September 1, when the story was the valuation. This is a follow-up because the income statement is new.

The loss is mostly a rounding error for your purposes

Strip out the $72 billion and the reported remainder is roughly $8.7 billion, which sits close to the nearly $9 billion Anthropic reportedly spent on compute and infrastructure in the first half. For comparison, that is more than the $7.3 billion it reportedly spent in all of 2025. So the business looks like this: revenue growing about fifteenfold year over year, and compute spending growing almost as fast.

I would not lose sleep over the headline. Non-cash charges on financing instruments tend to swing with the company's own valuation, which is a strange thing to base a vendor-risk decision on. The operating picture is the part that touches you.

The line that actually matters is the commitments

The same coverage reports roughly $450 billion in cloud commitments over the next decade, with $111 billion tied to Google Cloud, $110 billion to AWS and $31 billion to Microsoft named explicitly. It also reports up to $750 billion of potential compute spending including deals still being negotiated. Much of the cloud money is payable whether or not customers show up to use it.

That is a fixed-cost business with a very large fixed-cost base. When a supplier has fixed costs that big, it has two levers: grow usage faster than the commitments come due, or raise the price of the usage it already has. Neither one asks your permission. If growth in demand slows, the pressure lands on list prices, rate limits and the discounts that currently make your margin work.

There is a flip side that I find more reassuring than worrying. A company that has signed for that much capacity is motivated to fill it, which usually means cheaper tokens and generous caching, not tighter ones. Both outcomes are plausible, and the S-1 numbers cannot tell you which one you get.

Two risk factors that read differently when you build on the API

The reporting flags that about 20% of first-half revenue came from two customers. A few large buyers anchoring the revenue means the product roadmap and capacity allocation will lean toward them. When capacity is tight, I would expect enterprise contracts to get served before a hobby-tier API key does.

It also reports that Anthropic sought more capacity than the hyperscalers would supply and that those hyperscalers increasingly treat it as a competitor. For a solo builder, the practical translation is simple: availability incidents and rate-limit tightening are more likely to come from a capacity squeeze than from a technical failure. That is a different failure mode to plan for.

What I'd actually do

Three things, none of them dramatic.

First, price your product as if the vendor's per-token price can move 30% in either direction within a year, and check whether your margin survives both ends. If it only works at today's price with no cushion, that is the real finding.

Second, put a second model behind a thin interface in your code now, while nothing is on fire. You do not need to run it in production. You need to have run your test prompts against it once, so a switch is an afternoon and not a project.

Third, keep your prompts cache-friendly and your contexts short. Whatever happens to list prices, the tokens you do not send are the only cost savings that no one can take back.

Where this reasoning could be wrong

I am reading a draft through a newspaper, and the draft is missing the number I would want most: gross margin, which the coverage does not report. If inference gross margins are healthy, the commitments look like an aggressive growth bet and your prices are safe. If they are thin, the pressure on pricing is real. Until a public filing shows that line, anyone claiming to know which one is true is guessing, me included.

Author

Lukas

@lukcombinator

Sources

Anthropic revenue hits $16.3 billion as preliminary S-1 outlines IPO risks, WSJ via Investing.com Anthropic's draft S-1 reveals $16.3B revenue, $80.7B loss and up to $750B compute bill, Dealroom Anthropic's S-1 leaks again, Fortune