With an annual revenue of 4.6 billion, how dare Anthropic spend 518 billion US dollars?

Just yesterday, Anthropic's full details were completely exposed.

While its listing is in preparation and the prospectus has not been officially released to the public, many key pieces of information were first leaked by Reuters.

There are several sets of figures in it that are quite staggering.

Its revenue is around 4.6 billion USD, its target valuation may exceed 2 trillion USD, and its net loss is close to 42 billion USD, all denominated in US dollars.

Even Shi Chao felt dizzy looking at the numbers. You know, Anthropic only completed its last financing round in May this year with a valuation of just 965 billion USD, and it's only been a few months...

But more meaningful than just looking at the numbers is that everyone finally got to see a more complete picture of Anthropic.

How profitable the company really is, where all the money goes, how much it plans to invest in computing power in the future, and what Anthropic is worried about — all these points are basically covered in reports by several foreign media outlets.

Shi Chao also did some research and found that the truly interesting part of this prospectus is not just how fast Anthropic is growing, but also some of its awkward, conflicting situations.

Without further ado, let's first take a look at how profitable Anthropic actually is.

In 2025, Anthropic's full-year revenue was close to 4.6 billion USD, roughly 12 times that of 2024.

As for where this money comes from? Obviously, relying only on individual users like Shi Chao paying for Claude's monthly subscriptions cannot support such a large business scale.

Anthropic's core market has always been enterprises and developers.

In February this year, Anthropic disclosed that more than 500 enterprise customers had an annualized spending of over 1 million USD on Claude. Less than two months later, that number exceeded 1000.

Not to mention cloud vendors: on AWS alone, more than 100,000 customers are already using Claude.

In other words, Anthropic has now cultivated a group of large customers who are willing to pay millions of dollars to it every year.

However, large customers also bring their own troubles.

The prospectus shows that in 2025, Anthropic's two largest customers contributed nearly a quarter of its revenue. But many large customers have not signed long-term contracts with them. If they reduce usage or stop using Anthropic's services one day, the impact on Anthropic will be very direct.

But at the moment, compared to the risks, people are more concerned about the rapid growth of its business.

Because, 4.6 billion USD is just the beginning.

You know, this is the data for 2025, and Claude Code was only officially launched in May last year.

At the end of last year, Anthropic's annualized revenue was only about 9 billion USD, but according to Reuters reports, by the end of July this year, that figure had exceeded 65 billion USD.

The annualized revenue here is converted to a full year based on the revenue level at that time, not the actual full-year revenue.

In addition, according to the latest report from the FT, Anthropic's revenue in Q2 this year has reached 11.5 billion USD, and its adjusted operating profit has turned positive. Following this trend, it is very likely to remain profitable in Q3.

So it's not hard to understand why Anthropic's valuation has been pushed up to 2 trillion USD — its future is full of promise.

But at this point, some readers may be curious: even with your explosive growth, where did that nearly 42 billion USD net loss come from?

Anthropic did post a net loss of nearly 42 billion USD last year, but about 34 billion USD of that was not money the company actually spent out of its own pocket.

According to Reuters reports, this 34 billion USD mainly comes from some financing instruments that may be converted into Anthropic's stock in the future.

Here's the catch: Anthropic's valuation has risen so fast that the more valuable the company becomes, the more valuable these instruments that can be converted into stock in the future are. On the books, this money "owed" to investors is recorded as a current-period loss.

So to a certain extent, the reason Anthropic's loss was so staggering last year is largely because its market value rose too fast...

However, even after excluding this part, Anthropic's operating loss still amounted to more than 8 billion USD.

In addition, in 2025, Anthropic spent 7.33 billion USD solely on computing power and infrastructure, more than triple the figure of the previous year, accounting for more than half of its total annual operating expenses.

A simple conversion shows that for every 1 USD of revenue it earns, it spends about 1.6 USD on these items.

And this is only the money it actually spent.

There is another 518 billion USD in the prospectus, which represents Anthropic's future committed expenditures related to cloud services, computing power and infrastructure.

Last year, Alibaba announced that it would invest at least 380 billion RMB in AI and cloud infrastructure over the next three years.

When that figure came out, everyone already thought Alibaba was going all-in on AI.

But if you convert 380 billion RMB into US dollars, it's only around 530 billion USD. In comparison, even Alibaba seems to be sitting at the kids' table...

Of course, the investment periods and calibers of the two sides are not exactly the same, so they cannot be directly compared.

But putting them side by side, for Anthropic, a company established only 5 years ago, to commit to spending over 500 billion USD, it seems that the total amount of money AI can absorb is truly bottomless.

And according to Reuters reports, these commitments span the next ten years, about 80% of which are either non-cancellable, or even if the full amount is not used up, the payment still has to be made.

Although Anthropic is well aware of the risks, it explicitly mentioned in the prospectus that models may exhibit behaviors such as deception, manipulation, and resistance to shutdown, and more advanced AI may bring catastrophic even existential risks. Not long ago, Dario also called for pressing the brakes on AI development.

But Shi Chao thinks that right now, Anthropic may find it very difficult to actually slow down, or the cost of doing so would be extremely high.

After all, the computing power has been booked in advance, and the money has been promised to be spent. To digest these investments, it has to train more powerful models and get more people to use Claude.

Its competitors will not stop either, so to maintain its leading position, Anthropic must ensure that its models always stay in the first tier.

The result is that the models continue to iterate forward, and more money continues to be poured into computing power.

This is very realistic: shortly after Dario published the article "We Must Pace the Frontier", Anthropic released its new model Opus 5.5.

Anyway, Shi Chao hasn't found any manufacturer in the market that has taken the initiative to call a halt. Even Google, which seemed to have been lying low for a while, has spread news that Gemini 4 is about to be released.

What's even more staggering than model iteration is the crazy bet that all parties are placing on AI infrastructure.

In addition to Anthropic, OpenAI, Google, Meta, and a host of cloud vendors are all grabbing chips and building data centers, with investments easily reaching hundreds of billions or even trillions of US dollars.

In the end, this money does not only flow to NVIDIA.

Servers need power supply, chips need heat dissipation, data needs to be transmitted, and a whole industrial chain behind it gets to benefit from this boom.

Not long ago, Shi Chao attended the Yunqi Conference, and a very obvious feeling was that now anything that is even slightly related to data centers can be driven up by this wave of AI infrastructure boom.

So after looking at the key data in Anthropic's prospectus, whether the company is worth 2 trillion USD may not be the most important point.

We may still have underestimated exactly how much money AI is going to absorb.

Sources of images and materials:

Anthropic, The Verge

Reuters, Anthropic's IPO prospectus shows sweeping AI vision, surging costs

Financial Times, Anthropic warns of ‘existential risks to humanity’ in IPO prospectus

This article is from the WeChat official account "Xiacha X.PIN", author: Xixi, editors: Jiang Jiang & Mian Xian, published with authorization from 36Kr.