Anthropic Opens Its Books: $16 Billion in Sales, an $80 Billion Loss, and a Computing Bill That Could Hit $750 Billion

For years, anyone trying to judge Anthropic had to rely on leaks, funding rounds and educated guesses. That’s changing. A draft of the company’s IPO prospectus, known as an S-1, has been reviewed by The Wall Street Journal, and it offers the clearest picture yet of how the Claude maker makes money, how much it spends, and what keeps its lawyers up at night.

The short version: the growth is real, the losses look scary, and the spending is on a scale few companies have ever attempted.

Growth that’s hard to believe

Start with the headline number. Anthropic brought in $16.3 billion in revenue in the first half of 2026. In the same six months last year, it made $1.1 billion.

That’s roughly fifteen times more in a single year. And it didn’t slow down after June. Anthropic separately told potential investors that it earned $16.8 billion in the third quarter alone, about 45% more than the quarter before. The company also said that quarter showed adjusted profitability and positive free cash flow, though “adjusted” profit isn’t the same as profit under standard accounting rules.

The $80.7 billion loss, explained

Then comes the number that will grab the headlines: a net loss of $80.7 billion in the first half of the year, up from about $13 billion a year earlier.

It sounds alarming, but most of it isn’t money that actually left the bank. About $72 billion came from an accounting charge linked to changes in the value of convertible notes and other financing tools the company used to raise early capital. As Anthropic’s valuation shot up, the value of those instruments did too, and accounting rules require that to show up as a loss.

Take that charge out, and the remaining loss is closer to $8.7 billion. Still big, but a very different story.

The real headline: the computing bill

If there’s one number that explains Anthropic’s business, it’s this one.

In the first six months of 2026, the company spent nearly $9 billion on computing power and infrastructure. That’s more than the $7.3 billion it spent in all of 2025.

And that’s just what’s already been spent. Looking ahead, Anthropic has signed roughly $450 billion in cloud agreements running over the next seven to ten years, and most of them can’t be cancelled. The biggest pieces:

• About $111 billion with Google Cloud
• About $110 billion with Amazon Web Services
• About $31 billion with Microsoft

On top of that, it has deals worth up to $86 billion with smaller providers like Nscale and Lambda to rent Nvidia chips, and more than $160 billion tied to Broadcom arrangements for Google’s custom chips. Altogether, locked-in commitments are above $600 billion. Add the roughly $150 billion in deals still under negotiation, and the potential total reaches about $750 billion.

There’s an awkward twist in the filing too. Anthropic says Amazon, Google and Microsoft increasingly see it as a competitor and haven’t been willing to give it all the computing capacity it wants. That’s pushed the company toward third-party data centres and even more spending.

Two customers, one-fifth of the money

Another detail investors will look at closely: about 20% of first-half revenue, around $3.26 billion, came from just two customers. The filing doesn’t name them.

Anthropic expects a small group of large customers to keep making up a big share of its income. Some major enterprise clients also haven’t signed long-term contracts. If one of them cuts back or moves to a rival, it would show up in the numbers quickly.

A risk section unlike any other

Nearly a third of the draft prospectus is about risks, and some of them read more like science fiction than a stock market filing.

The company warns that its AI models could act in ways people didn’t intend, including showing self-preserving behaviour or manipulative responses such as blackmail to avoid being shut down. It says it could face legal and financial liability for damage caused by its models, from data security incidents to serious real-world disruptions, and that losses from a severe AI-related security event could go beyond what its insurance covers.

There are more ordinary risks too: falling AI prices, competition from open-source models, and data centre projects delayed by power shortages, land issues or local opposition. And because so much of the computing spending can’t be cancelled, the bills keep coming even if demand ever slows.

What it’s worth

Earlier market talk put Anthropic’s possible IPO valuation as high as $2 trillion. After seeing the latest financials, some investors are reportedly thinking closer to $1.5 trillion. Neither figure is final, and the company hasn’t announced a listing date, though reports have pointed to as early as November.

The filing also offers a small peek at the people running the company. Dario Amodei and his sister Daniela, the CEO and president, each earned a base salary of about $655,000 in 2025. Both have pledged to give away 80% of their personal shares, and the company recently set aside 5% of its stock for a new public benefit fund.

Why this matters beyond Wall Street

Anthropic isn’t just one more tech IPO. It’s the first time the public gets to see, in detail, what it actually costs to build frontier AI. And the answer is clear: unbelievably fast growth, paid for with an unbelievably large bet on computing power.

For Indian readers, there’s a direct link too. Indian IT companies are increasingly building on top of models like Claude. Infosys, for example, set up a dedicated centre with Anthropic earlier this year to build and deploy AI agents for clients. How stable Anthropic’s finances are matters to the companies building on it.

The big question investors will now have to answer is simple to ask and hard to settle: can revenue keep growing fast enough to cover a computing bill measured in hundreds of billions? The S-1 shows Anthropic is betting yes. The market will soon decide whether it agrees.