Anthropic's draft IPO filing: $16.3 billion revenue, $80.7 billion loss and $450 billion in cloud bills

Anthropic is getting ready for what could become one of the largest stock market debuts in history, and the first detailed look at its books is now out. A draft of its S-1, the prospectus a company files with US regulators before an initial public offering, was viewed by The Wall Street Journal and reported on 20 October 2026. The headline numbers are extreme in both directions. The maker of Claude took in $16.3 billion in revenue in the first six months of 2026, up from $1.1 billion a year earlier. Over the same six months it recorded a net loss of $80.7 billion.

Those two figures sound like they cannot belong to the same company. They do, and the explanation matters for anyone trying to judge how healthy the AI business really is. Below is what the draft filing reportedly says, what is cash and what is accounting, where the compute money goes, and which risks Anthropic itself chose to put in writing.

This article is based on the Wall Street Journal report by Anissa Gardizy and on follow up coverage by Investing.com, TechCrunch, Al Jazeera with Reuters and Quartz. The document is a draft and has not been made public. Anthropic did not comment, and Reuters noted that it could not independently verify the figures. Numbers in a draft can still change before a final filing.

Revenue grew about fifteen times in a year

The growth figure is the easiest part to understand. In the first half of 2025, Anthropic's revenue was about $1.1 billion. In the first half of 2026 it was $16.3 billion. That is roughly fifteen times as much in twelve months, at a scale where most software companies would celebrate doubling.

The half year number also fits what Anthropic told investors at its investor day last week. As we reported in our article on Anthropic's $16.8 billion third quarter, the company reportedly showed revenue of about $4.7 billion for the first quarter and about $11.6 billion for the second. Added together, that is the $16.3 billion in the draft prospectus. The third quarter alone then reached $16.8 billion, more than the whole first half, and Anthropic said it was profitable on an adjusted basis in that quarter.

Much of this growth comes from businesses and developers, not from consumers chatting with a bot. Claude Code, the coding assistant that writes and edits software, has been one of the main drivers. Companies pay for Claude through the API, through Claude subscriptions for their teams and through the big cloud marketplaces of Amazon, Google and Microsoft.

Two customers bring in a fifth of the money

The filing also shows how concentrated that revenue is. According to the Journal, about 20 percent of the $16.3 billion came from just two customers. That is roughly $3.3 billion from two accounts in six months. The draft reportedly says that Anthropic expects a limited number of customers to keep generating a significant share of its revenue for the foreseeable future.

The filing, as reported, does not name the two customers. Concentration like this is common for companies that sell infrastructure to other tech firms, and it is a standard risk factor in any prospectus. But it means that the loss of a single large contract, or a single customer deciding to build its own models, would show up immediately in the numbers. For investors who are asked to value Anthropic at well over a trillion dollars, that is a real question.

Why the loss is $80.7 billion

Now to the number that made the headlines. Anthropic's net loss for the first half of 2026 was $80.7 billion, compared with about $13 billion in the first half of 2025. For context, Reuters reported in September that the company lost about $42 billion in all of 2025.

The key detail is that about $72 billion of the H1 loss came from non cash accounting charges. Anthropic raised part of its funding through convertible notes and similar instruments. These are loans or contracts that investors can later turn into shares. When the company's value rises, the value of those instruments rises with it, and accounting rules require the company to book that increase as a loss. In plain words: the more valuable Anthropic became, the more it had to record as a loss on paper. No cash left the company for this part.

If you take the $72 billion out, about $8.7 billion of loss remains for the six months. That is our own subtraction, not a figure from the filing, and the actual operating loss may differ once taxes, interest and other items are separated out properly. But it gives a far more useful picture. A company with $16.3 billion in revenue that loses several billion dollars while it builds out compute is a very different story from a company that burns $80 billion.

This matters because the headline number will be quoted a lot, often without the explanation. It is a real number in the accounts, and it does say something about the financing structure. It does not mean that Claude costs five times more to run than it brings in.

Compute spending is growing even faster than revenue

The real cash story is compute. According to the Journal, Anthropic spent close to $9 billion on computing capacity and infrastructure in the first half of 2026. That is more than the $7.3 billion it spent in all of 2025. Quartz, citing Reuters, puts the figure for the first half of 2025 at about $2.3 billion, so infrastructure spending nearly quadrupled year over year.

Training frontier models needs huge clusters of chips, and serving millions of developers and businesses needs even more. Every new version of Claude has been larger and more capable, and every new customer adds inference load. The investor day figures suggested Anthropic now earns about $1.70 in revenue per dollar of compute, compared with roughly $0.63 in 2025. If that holds, the model is getting more efficient even as the absolute spending climbs.

$450 billion in cloud contracts, and maybe $750 billion

The long term commitments are where the scale becomes hard to imagine. The draft reportedly lists about $450 billion in cloud compute agreements over the next decade. The three biggest are $111 billion with Google Cloud, $110 billion with Amazon Web Services and $31 billion with Microsoft. Many of these obligations are noncancelable. Anthropic has to pay them whether demand grows as planned or not.

The Journal also reports that Anthropic could spend $750 billion in the coming years on additional capacity, including deals it is still negotiating. We covered the earlier leak about $518 billion in compute commitments a week ago. Quartz now describes that as a minimum of $518 billion over ten years for infrastructure, computing power and cloud services. The new $450 billion cloud figure and the older $518 billion total measure slightly different things, but they point in the same direction: Anthropic has signed up for spending that dwarfs its current revenue.

The bet behind this is simple to describe. If revenue keeps growing anywhere near the current pace, the contracts will look cheap in a few years, and early access to chips will have been a competitive advantage. If growth slows sharply, Anthropic carries decades of fixed costs. That is the same bet OpenAI, xAI and others are making, but a prospectus forces the company to spell it out in numbers.

The cloud giants are partners and rivals at the same time

TechCrunch highlighted one of the more uncomfortable passages. According to the leaked filing, Anthropic has had trouble getting as much capacity as it wanted from AWS, Google and Microsoft, which increasingly see the Claude maker as a competitor rather than just a customer. Anthropic says it has sought more capacity than the hyperscalers were willing to provide.

That tension is easy to explain. Amazon and Google are both major investors in Anthropic and sell Claude to their own cloud customers. But both also build their own models, Google with Gemini and Amazon with its Nova family, and Microsoft is deeply tied to OpenAI while also developing in house models. Every GPU or TPU that goes to Anthropic is one they cannot use for their own AI products.

To get around the bottleneck, Anthropic is leasing its own data centers and working with third party data center operators. The draft reportedly warns that some of those operators have a limited operating history and might be less reliable than the big cloud providers. In other words, the company is trading one risk, dependence on three giants, for another, dependence on younger firms that may not deliver on time.

The risk section: alignment, catastrophes and insurance

Every prospectus has a long risk section, and most of it is boilerplate. Anthropic's version reportedly includes a few items you do not see in a normal tech filing. According to the Journal and follow up coverage, the company describes uncertainty about how to prevent its products from causing harm, including through sophisticated cyberattacks. Seeking Alpha's summary mentions models that show self preservation behavior or could try to manipulate people, potential catastrophic incidents and political resistance to new data centers as listed risks.

One line stands out. Anthropic reportedly warns that damages from severe AI related incidents could exceed its available insurance coverage. That is a candid admission. Insurers are still working out how to price the risk of AI systems acting in harmful ways, and a company that sells frontier models to millions of users cannot fully insure itself against a worst case.

Anthropic has always talked about AI risk more openly than its rivals, so seeing it in a legal document is not a surprise. But there is a difference between a blog post and a prospectus. In a filing, risk factors are part of the legal disclosure to investors, and leaving out a known risk can lead to lawsuits later. Writing down alignment and catastrophic incidents in that context makes them part of the investment case.

Founders, salaries and the public benefit structure

The draft also offers a glimpse into governance. According to the Journal, CEO Dario Amodei and President Daniela Amodei each received a base salary of about $655,000 in 2025. Both co founders have pledged 80 percent of their personal equity to philanthropic causes, and the company plans a public benefit fund equal to 5 percent. Anthropic is organized as a public benefit corporation, which allows its board to weigh its stated mission alongside shareholder returns.

What this means for the IPO

Anthropic is reportedly aiming to list its shares in November. Before the investor day, a valuation of around $2 trillion was widely expected. Reuters reported afterwards that some current and prospective investors discussed a figure closer to $1.5 trillion, partly because of concerns that growth might be slowing from its extreme pace.

The draft S-1 gives both sides material. Bulls will point to fifteen fold revenue growth, an adjusted profit in the third quarter and a loss that is mostly an accounting effect. Bears will point to revenue concentration, hundreds of billions in fixed compute obligations, strained relations with the companies that supply its chips and a risk section that openly discusses catastrophic outcomes.

Our take

The $80.7 billion loss is the number that will spread on social media, and it is mostly a distraction. It mainly reflects how fast Anthropic's valuation rose, not how much money the company burned. The numbers to watch are the ones that involve real cash: about $9 billion of compute spending in six months, about $450 billion in long term cloud contracts and a fifth of revenue resting on two customers.

The most interesting signal is the conflict with the hyperscalers. For years the cloud giants were happy to fund AI labs and sell their models. Now that those labs are big enough to compete with them, compute access itself has become a strategic weapon. If Anthropic cannot get the capacity it needs from Amazon, Google and Microsoft, its plan depends on newer data center builders delivering on time. That will be worth watching long after the IPO.

Until the final prospectus is public, all of these numbers come from a draft seen by one newspaper. We will update this article when the official filing appears.

Sources: The Wall Street Journal, Investing.com, TechCrunch, Al Jazeera and Reuters, Quartz, Seeking Alpha.