A Sneak Peek Into Anthropic’s Business

There's a lot of excitement and anxiety about how AI companies are doing. Are they a goldmine for investors or approaching a bubble? One of the biggest AI companies, Anthropic, the company behind Claude, is preparing to go public as soon as November. The Anthropic IPO is probably the most highly anticipated IPO of the year. That's our colleague Anisa Gardese. Investors are extremely excited about Anthropic going public as the first AI lab to be a public company. And like it's going to be one of the biggest IPOs in history when it happens. That's exactly right. Right now, the numbers that they're talking about are that Anthropic could raise around a hundred billion at an up to two trillion dollar valuation. So it's quite unprecedented. Yeah, it's a whopper. Anisa got a chance to look inside Anthropic's business by pouring over a massive document that details the company's finances. It's known as an S1. An S1 is this long document that private companies have to file with the SEC before they go public. Ideally it contains all of the answers to questions that investors might have. And it brings up all of the risks and concerns that the company has. So it's kind of the company for the first time being an open book about its finances and its vision for the future. Normally, this document is publicly available to investors a few weeks before an IPO. But Anisa was able to take a look at an early version before its release. Anthropic declined a comment on the contents of the draft S1. So this document is obviously a draft and things could change. But the reason that these numbers matter and the reason that right now is a pretty opportune time to look at them is that there is a lot of question on the AI trade in general. That's always sort of been the case that that investors are wondering how long things can keep going. And if there's ever going to be a slowdown and who's going to be impacted is the bubble going to burst. These are questions that are constantly coming up as we talk about AI companies. And so understanding these numbers really helps us understand how Anthropic fits into that broader moment. Welcome to the journal, our show about money, business and power. I'm Ryan Kunusin, it's Wednesday, October 21st. Coming up on the show. Let's take a peek under the hood of Anthropic. So let's talk about the financial picture that the S1 painted and let's start with revenue. What can we see about where this company is making money and how much is it making? So Anthropic makes money primarily from selling access to its sophisticated AI models. And what we saw in the draft S1 was that in the first half of 2026. They brought in around $16 billion in revenue. And that's compared with about a billion dollars in the same period last year. And the company separately outside of this draft S1 has indicated that revenue for the third quarter of 2026 is around 16 billion. So that's pretty significant revenue growth right there. Yeah, they're on pace for crazy revenue growth this year. And that is the main reason that people are excited about the company. What were you able to learn about who Anthropic's customers are? Who's actually paying them this money? Anthropic said 20% of its revenue in the first half of the year came from two customers. Wow. And they said that for the foreseeable future, they expect just a handful of customers to make up the most of their revenue. That to me was quite chalking, and investors are going to focus on that disclosure. Which are the two companies that are the customers, you know? Anthropic didn't disclose the names of the companies, so it's up to us to guess or to figure out. All right, I'm going to put you on that. Why is so much of Anthropic's revenue coming from only a few customers? Well, if you think about just the revenue that they're reporting, how big these numbers are, there aren't that many companies that can help them get to $16 billion in revenue. The big question in AI circles right now is, well, big companies pay, will consumers pay. And so there's a chance, you know, we don't know that there are only a few types of large companies that feel like spending billions of dollars on Anthropic tools make sense for their business. This is one of the many risks that Anita saw on the S1. Not only is Anthropic relying on a handful of massive customers, those customers can switch to a rival AI company at any time. And so that will be something that maybe gives people pause, because as we see in the AI market in general, when a new model comes out, everyone gets really excited about it and switches up which model they use. So I'll just say if OpenAI comes out with a new model, would one of those top two customers switch to OpenAI overnight or in a couple of weeks, and Anthropic is saying it actually doesn't know and can't prevent that from happening. All right, now let's talk about costs. What did this document say about how much Anthropic is spending and what are they spending it on? We saw that compute and infrastructure costs are by far the largest cost that the business faces. That means money it spends to rent chips or money it spends to buy chips which they use to train new models or to serve their product to customers. And we saw that in just the first half of 26, they spent around $9 billion on compute. And that is more than they spend in all of 2025. Can you remind us why computing power is so important? It is important because it is the only way that Anthropic can sustain its business. Every business needs compute so we compute to do this call right now and people need compute to access their email and websites and store photos. But for the frontier AI labs, compute kind of takes on this new importance. It's the thing they need to train their models. And so Anthropic has gone to the point where it's one of the biggest spenders on compute in the world actually. And they've indicated that unless they can continue accessing greater and greater amounts of compute, it's going to impact their business growth and their ability to sustain their lead. In order to sustain its growth, Anthropic needs tons of computing power. But according to the S1, it's having trouble getting enough. Currently, Anthropic is mainly paying cloud computing providers like Google and Amazon, but in the document. Anthropic said it tried to access more computing power than, quote, "those partners were willing to provide." The problem is that they've hit their limit on how much they can get from those companies, which, by the way, is quite crazy. And they need to pursue a different strategy for accessing their future compute. So we're seeing them work with smaller cloud providers who have a little bit less experience. And we're seeing them try to develop their own infrastructure themselves. And Anthropic is starting to have to put down massive amounts of money to secure future capacity. And that's taking a risk that they're betting they're going to need this and they're putting down the money now, but no one really knows, they even say they don't know how to predict future demand. Right. It is they have to build it and hope the customers will come. Right now, it looks like the customers are coming because they can't even keep up right now with the amount of demand that there is, but they need an enormous amount of computing and compute is very expensive. Yeah, I'm getting this wrong either way would be bad. If they have too much compute and no one's using it, then they're overpaying for something that isn't being used. And then they say if they don't secure enough compute, then their revenue will be limited and customers might flock to other models. So it's a tricky situation, no matter what happens with AI demand. And while Anthropic's customers can switch anytime, Anthropic doesn't have that same flexibility with the company's it's paying for compute power. So with their deals with cloud providers, where they run chips from other companies, they're on the hook to pay around $450 billion over the next seven to ten years. And most of those commitments are not cancelable. They cited this term that they're going to pay for this compute regardless of actual usage. In all, Anthropic is a signal that could spend around $750 billion in the coming years on computing power, including deals that still negotiating. And I understand that one of the most eye-popping figures you found in this S1 was that, Anthropic posted an $80.7 billion loss in just the first six months of this year. Can you explain that? That seems like an enormous amount of money. It's enormous. $80 billion is quite incredible. Now one thing that's really important to know about that number is that that number is their net loss and $72 billion of that $80 billion was actually an accounting adjustment that the company had to make. And that doesn't mean it's free money and that it doesn't matter, but it does mean that that wasn't money going out the door on on compute spend. The accounting charge comes from one of the ways Anthropic raised it. money. In the past, the company promised some of its shares in return for loans, but since then, the value of those shares is increased, and the difference shows up in the books. So, that $81 billion figure does not mean it's spent that money during those six months, but it did have to account for these liabilities that it has and the liabilities are directly tied to anthropic valuation. One of the unusual things about this S1 is that roughly a third of it is just about risk, risk to investors and risks to humanity. That's next. Risk factors are something that every company lists in its perspective, but anthropic is an unusual company. You can kind of get chills reading some of the risk factors because anthropic is talking about this future world of dangerous AI that it can't control necessarily, so if the company making the models and selling the models is basically saying, we don't know if we can make sure that these models don't cause harm. In the S1, anthropic said it sees damages caused by AI as a giant financial risk. The company said AI models can show what they called self-preserving qualities and that in test scenarios, they've been generated black male type responses to keep humans from shutting them down. In some cases, anthropic said it could bear liability if its AI does certain harmful behaviors. And this kind of harm they could cause could take many different forms. Some of them so larmy forms are sophisticated cyber attacks. They said that there could be some sort of a catastrophic AI related incident and they said it could be so large and it could cause so much damage that all of the insurance policies could uncover the potential damages. Not the kind of line that you normally see in an investor perspective. Did anthropic in the risk factors or anywhere else in the perspective talk about how it's going to try to prevent that from happening? The company says very infallibly that it takes model safety very seriously. It might delay the release of models in order to make sure that it's not releasing something that could be harmful, but it also doesn't know. So I think anthropic is taking the safety thing seriously and they're dedicating time to it, resources to it, to test their models before they come out. Another thing that could hold back on anthropic's growth is nationwide opposition to data centers. Back off to data centers has really picked up in the past six months. And I think that's why it made it into the risk factors. So we talked about why they need all this compute. Well, they might actually not be able to get that compute given growing resistance to data centers specifically in the US. So that's why they're getting most of their computing infrastructure in the United States. And anthropic said that they're looking to obtain data center capacity outside of the US just to have a more diverse base of compute. In case there was some sort of regulatory issue that specifically impacted the United States. But I suppose another risk of that also is that this is just fundamental like an increased costs for this company as it needs to try to build out all this compute. Yes. And the increased costs are coming from anthropic having to offer more incentives when it builds a data center. So they talked about having these voluntary payments that they're making to sort of make people feel better about their data centers. And that's all making things more expensive. And everything in the data center industry is getting more expensive. So, you know, we could see some of those large numbers get even bigger, but not for more compute just to alleviate concerns about them being in someone's neighborhood. So given the explosive growth in revenue that we're saying, but also like the enormous need to spend money on building out compute, how do you think investors are going to think about or digest the information that they're learning from this S1? I think that investors are going to really want anthropic to talk more and share more information about who they think is going to be buying their products in the future and why they're going to continue buying it from anthropic. Having a few large customers who are really excited and have billions of dollars sitting around to spend on your products is great. But unless anthropic turns that around into multi-year commitments or something a little bit more stable, the customer concentration is potentially the part that might give people pause. Because, you know, there's major questions on whether there's enough money in the world to sustain that pace, even if people really like AI, do they have the money to keep paying for it? I think it's a big open question. That's all for today. Wednesday, October 21st. The journal is a co-production of Spotify and the Wall Street Journal. If you like our show, follow us on Spotify or wherever you get your podcasts, we're out every weekday afternoon. Thanks for listening. See you tomorrow.