Inside Anthropic’s S-1

**Anthropic has kicked off the road to one of the biggest IPOs of all time, and its 300 page S-1 filing has given the credit market plenty to chew on.

In this episode of Cloud 9fin, interim distressed debt editor Sasha Padbidri cuts through the noise with senior reporters Maria Heeter and Dan Geiger to unpack the biggest takeaways from the filing.

Tune in as they break down the scale of Anthropic’s massive AI infrastructure buildout, litigation risks, and the wildcards they’re watching next.

Have any feedback for us? Send us a note at [e-mail], and be sure to check out Dan Geiger’s Hot Stacks — his weekly newsletter tracking data center deals and the dealmakers behind them. The first three editions are free to read on 9fin’s Insights page. Thanks for listening!

Creators and Guests

Producer

Evie Rusman-Okonkwo

EMEA Podcast Producer for 9fin Limited

What is Cloud 9fin?

Leveraged finance, distressed debt, and private credit drive today’s markets. Cloud 9fin delivers expert insights on high-yield bonds, syndicated loans, direct lending, and debt restructuring. Join top analysts and investors as we explore credit markets, special situations, and private debt strategies shaping the industry.

From credit risk assessment to institutional credit trends, each episode provides actionable intelligence for fund managers, institutional investors, and financial professionals. Whether you’re tracking high-yield issuances, analyzing corporate debt, or uncovering distressed debt opportunities, we’ve got you covered.

Through its AI-powered data and analytics platform, 9fin provides everything you need to get your head around credit or win a mandate — all in one place. We help subscribers win business, outperform their peers and save time. Stay ahead in leveraged finance market trends—subscribe now for expert discussions on the forces moving global credit.

**Sasha Padbidri**

Hi everyone, I'm Sasha Padbidri, interim distressed debt editor with *9fin*. Welcome to this week's *Cloud 9fin* podcast, where I unpack **Anthropic**'s S-1 document with senior reporters Maria Heeter and Dan Geiger. Welcome, Maria and Dan.

First of all, thank you for taking time to do this podcast in what has got to be one of the most insane weeks we've had in a while. So kudos to you for making time for us. And this all kicked off earlier this week with the release of this highly anticipated document ahead of what could be one of the largest potential IPOs of all time, and it's a really hefty document.

Plenty to unpack here, so let's just start with Maria. You know, you did some reporting on the sheer scale of the infrastructure bill that was indicated in the document, and **Anthropic** has locked in hundreds of billions of dollars in non-cancelable compute agreements across suppliers like **Amazon**, **Google**, **Microsoft**, and neoclouds. Could you walk us through that?

**Maria Heeter**

Thanks so much for having me. It's very nice to be here. So, just for background, I used to cover venture capital, and I've followed **Anthropic** since it was a tiny baby startup about five years ago. But this time, with my credit hat on, I was very interested in how much Anthropic is spending.

In recent earnings reports, both **Google** and **Meta** have recorded negative cash flow in their recent earnings because of how much they're spending to finance their AI technology, particularly on infrastructure. That's buying compute, which involves leasing space in data centers, spending money on chips, and basically securing the energy and technology that they need in order to run their AI models, which power the chatbots that we use every day, like Claude, Gemini and ChatGPT.

When I looked at **Anthropic'**s S-1, which is the first time all of its financials are cobbled together, we've seen them drip, drip out for a series of about three years. But this is the first time they've had to show their investors what their bills are and how they plan to spend them. The main takeaway is that while Anthropic's compute bills are rising and a lot are not included in its balance sheet, Anthropic's revenue is pretty healthy.

So, according to its current financial statements, **Anthropic**'s revenue reached $16.3bn in the first half of 26, which is up from $1.2bn from the first half of 2025. That is a huge increase, but at the same time, **Anthropic** spending is also increasing. **Anthropic**'s S-1 showed that it could spend more than $800bn in infrastructure costs over the next decade. That is $807bn to cloud providers, including **Microsoft**, **SpaceX**, **Nscale**, **Lambda,** and other companies, including cloud providers.

The largest commitment was to **Broadcom**, the chip maker, for about $106bn in rent for five-year lease payments. What jumped out to me about the **Broadcom** lease was that the leases are classified as finance leases. That makes them look more like debt, and they will be added to the balance sheet as equipment moves online. My main takeaway is that **Anthropic** will be spending a lot, and a lot of this is not reflected on its current P&L statement.

So **Anthropic** is going to spend $45bn to **Microsoft** before May 2033, $115bn to **Google** before 2033, $110bn to **Amazon**, and that's just a few of their multi-billion-dollar payments that they cannot walk away from. This underscores that **Anthropic** needs its revenue to grow, and it's going to be interesting to see what happens as investors become more skeptical, as evidenced by how debt is trading, of whether AI spending is actually going to pan out.

**Sasha Padbidri**

Awesome, yeah, no, certainly, especially after all of the headlines from last week. But moving away from that, you also mentioned that the S-1 document flagged, you know, litigation and regulatory action as one of AI's biggest risk. Can you walk us through that, especially since you know you've been covering the **Character.ai** bankruptcy quite a bit?

**Maria Heeter**

Yeah, I mean, what kind of jumped out to me was that **Anthropic** has to detail all of its risks to investors, and those can be as minute as we might be hacked, and we might not be able to continue as a business of a going concern. What was interesting about this S-1 is that **Anthropic** warns in extreme detail that litigation is one of its biggest risks.

So for background: **Meta**, the owner of **Facebook** and **Instagram**, had about two decades without being targeted by state attorneys general before it had to pay serious cash over allegations of user harm. Just last week, a New Mexico judge ordered **Meta** to pay a $25bn penalty over violations of consumer protection law, and in August, **Meta** agreed to pay $17bn to settle lawsuits, which were brought by 51 attorneys general.

In the bankruptcy of **Character.ai**, which I have been reporting on, 46 attorneys general, which are bipartisan across many states, have been investigating Character, and the only reason that they didn't secure a settlement, in my view, is that character declared bankruptcy before that. So, **Anthropic** is going public just as state attorneys general really get their act together and decide the way to hold Big Technology companies accountable for harming users is through litigation.

So turning to **Anthropic**'s S-1, Anthropic warned investors that suicidal or violent people using its chatbot could be a risk. Children bypassing age limits and also cybersecurity incidents caused by its chatbots could expose it to lawsuits. State attorneys general are beginning to target other companies for similar violations, and it remains to be seen what the effect will be on Anthropic's balance sheet.

**Sasha Padbidri**

Got it. That is certainly something that we're going to watch very closely. Thank you, Maria. Now moving over to Dan, who, by the way, listeners has an amazing column called Hot Stacks, which is published on Thursdays. You put out an article detailing **SpaceX**'s 120 and a half billion dollar computing contract with **Anthropic**. Can you walk us through, you know, the mechanics of that contract?

**Dan Geiger**

Yes, very nice to be here. Thank you. It's it's funny because **Anthropic** is usually like I'm covering the nitty gritty of these data center developers and infrastructure providers, and **Anthropic**'s usually the off taker. So the S-1 was this interesting chance to kind of get under the hood. The thing that jumped out about this arrangement between **Anthropic** and **SpaceX** is not only the magnitude, right? It's this $120.5bn contract, but the fact that it's cancelable on 90 days' notice, so you have this huge block of spending, this huge amount of compute that could be kind of torn up by either side, **Anthropic** or **SpaceX**, at any time.

And I guess that jumped out at me in the sense that is this a risk, right? In the S-1 hiding in plain sight, because of course, you know, as Maria mentioned, right, it has to continue to scale from the first quarter to the second quarter. It scaled its revenue tremendously, and that's all based on continuing to ramp up its compute. And about 30% of its planned spend, **Anthropic**'s planned spend on compute is with **SpaceX**.

**SpaceX** is the thing that's giving **Anthropic** its near-term immediate compute. Right? It's using Colossus. Elon Musk set up that gigantic data center installation in Memphis very quickly. So these other contracts that **Anthropic** has, they're not as near-term.

So it's incredibly important for **Anthropic** to have this compute, and you know it's funny because Elon has said some interesting things along the way recently, right? He has said that he has in the third quarter earnings call for **SpaceX** that he has tremendous ambitions for ****Grok Bot. Grok Bot, of course, is the **SpaceX** large language model that he's you know he's turning into an agentic AI partner for for customers.

They have Cursor. They just bought Cursor, which is a coding platform that was purchased for $60bn. You know they have tremendous AI ambitions themselves, right? And Elon hinted in this call that eventually they may need their own supercomputer, right, for their own stuff. And he's made sort of other statements that he could pull this back at some point. Obviously, that would be, you know, that that would be maybe problematic for **Anthropic**.

I mean, **SpaceX** itself said that it has gone out to the market, right? We're at this moment where there's a lot of planned compute coming online, but there's not a lot that's actually out there right now. And **SpaceX** has has kind of reinforced that by saying they've gone out to the market, right? They said in their earnings call, they looked around and there's really not a lot of cloud compute available for sale, right? So in in some senses, like they're the only game in town, so is that a risk? Yes and no, right? Because just as **Anthropic** is incredibly, you know, I found in my reporting as incredibly dependent **Anthropic** is on **SpaceX**. **SpaceX** too, although it may have these future ambitions, as we just sort of discussed.

On its own AI, you know, it's very dependent right now on Anthropic and this revenue that it's getting. 28% of **SpaceX**'s revenue guidance in the first quarter of 2027 is literally from Anthropic, and it's like 25% for the year for its entire revenue stream, which just shows you how incredibly important **Anthropic** is to **SpaceX**'s own balance sheet, right? And why is that? Why is it such an important part? Because it's up charging **Anthropic** because of this kind of short-term cancellable structure to this contract or this agreement. It's able to essentially, you know, continue to mark its compute to market, which right now you could say, hey, that's risky, right?

If if there's a pullback, this will be the first thing to go. But right now, there hasn't been a pullback. There's actually scaling compute needs and pricing keeps going up. I think in the **SpaceX** third quarter earnings, they were quoting that they're getting more than $50 per watt of compute, which is considered a very kind of lucrative sum. That's across their third party clientele, of which **Anthropic** is just a part.

They also lease out space, I believe, to **Google** and to others, and it's so lucrative that they say that like basically equates to a payback on this infrastructure, on new infrastructure, compute infrastructure, which we we of course know is very expensive within about a year. And Elon Musk chimes in during the call again just a few weeks ago, and he says that it was not impossible. Those are his words. Next year could get even more lucrative to the point where it would pay back infrastructure within six months, right?

I mean, we're talking about an incredibly profitable business that they're running by leasing this compute out to customers like **Anthropic**, right? So, so there's this mutual benefit that is also kind of underpinning this agreement, although it's a cancelable agreement, although it can be, you know, there's there's factors that could change this arrangement on short notice. For now, it seems like an agreement of mutual need.

**Anthropic** needs this compute; it needs to continue to scale its revenue. We saw from the first to the second quarter, it became profitable, right? And why did it become profitable? Became profitable because it's just ramping its revenue so much. It had a margin in the second quarter, as the first profitable quarter of 8% because it grew its its revenue from about $5bn to more than twice, right, to $11bn plus. And there's kind of expectations that it's going to continue to scale. It needs **SpaceX** clearly to do that, and **SpaceX** is is making a very very pretty penny right now in leasing this space to **Anthropic**. So for the time being, I think you basically have a lot of consensus out there in the market that this is a match made, if not in heaven, than in some digital version of that.

**Sasha Padbidri**

Yeah, thank you for walking us through all of that, Dan. I actually just had like you know one more question for you. When you were like talking to your market sources, how did they feel about that? Because as you mentioned earlier, when you were saying this is one of the very first times you were actually getting a closer look at all these compute contracts.

**Dan Geiger**

I think that there were some that did acknowledge right that there's just naturally a baseline risk to this that if you have a counterparty like Elon Musk, who has these grand ambitions, you know, for his kind of his own business, his own stake in this AI race, that you know, sure, that could be a wild card here. This $120.5bn agreement between the two runs till 2029, you know it is a short term agreement by nature or by definition. But you know, I do think there is some acknowledgement, right, that there is some built in risk.

I do think though that the sort of takeaway that you hear from most observers right now, again, kind of comes back to the point I just made, which is that **Anthropic** needs this compute. It's not like they can just easily shift elsewhere. They're rapidly scaling. SpaceX is providing this capacity at this kind of very key moment.

For **SpaceX**'s part, they're earning this pretty penny by giving it to **Anthropic**, right? I mean, who else is going to be able to pay the freight? Maybe somebody else, but **Anthropic**'s got 130bn dollars of cash on its balance sheet. You know, they're pretty good customer right now, right? It's it's hard to replace an Anthropic right now. They can pay this freight, so it does seem to be a mutually very beneficial. The title of my article was "Mutually Assured Construction." I thought that was a nifty title. Yeah, I think they sort of they both. It's it's very symbiotic right now.

**Sasha Padbidri**

Awesome, good stuff. So both of you have said your piece, and there's plenty for readers to unpack here. But just to wrap things up, as we you know look past the initial headline noise and all. That what is the most important metric detail wildcard risk thing that you're tracking that you know our listeners should be aware of?

**Maria Heeter**

I think the biggest wildcard is litigation because we cannot see that reflected in **Anthropic**'s financial statements, like its balance sheet or its P&L, and the only way you can know how much attorneys general or appellate firms are going to win is by getting inside their heads.

**Sasha Padbidri**

Got it, and that's totally valid. What about you, Dan?

**Dan Geiger**

I mean, listen, I don't want to sound boosterish. Maria is certainly right in highlighting the fact that there's tons of risks here. There are tons of unknowns. I think that this whole AI build out, this this enormous bonanza, has been plagued by right uncertainty. It's it's been totally volatile. No one knows exactly how it's going to play out, and I think everybody's kind of bracing themselves for some kind of shoe to drop.

But with that said, I think what came out with this S-1 was a picture of a company that was a lot healthier financially than many people kind of expected. I know that's that's the feedback I've gotten from speaking to sources that this is a company that is earning extraordinary revenue, that's scaling its revenue extraordinarily, and that is becoming profitable fairly quickly.

People often, or some people, have made the the comparison to **WeWork**, right? Which was a company that scaled revenue, but as it scaled revenue, it scaled its losses, right? It was just unprofitable growth. That doesn't seem to be the case with **Anthropic**, and I think seeing that it's released a little bit of optimism.

I think in the AI infrastructure market, you've seen these bonds that have taken a beating; they have rebounded. These are debts tied to AI infrastructure companies, right? They've they've rebounded a little bit this week. Is it a you know 90 degree jump. No, but it's you've seen a bump, and I think that's because everybody just realized that this business can be profitable.

**Sasha Padbidri**

Awesome. All right, guys, thank you for joining me today. Fantastic reporting as always, and thank you, Dan, for making your *9fin* podcast debut. So good to have you. And for listeners who want full coverage and data breakdowns of this **Anthropic** S-1 that our entire team has been working on, or even a copy of Dan's weekly digest *Hot Stacks*. Please reach out to us at *[e-mail]* And thank you everyone for tuning in to this episode of *Cloud 9fin*. We hope to see you again next time!