Ed Zitron on Anthropic's IPO (S-1), AI Debt, and Counterparty Risk

Jack Farley (0:00)

This episode is brought to you by Sarmaya Partners, who manages the Return To Tangibles ETF, Lens, ticker L-E-N-S, the Sarmaya Thematic ETF, built for a return to tangibles. Learn more at sarmayetf.com and see the show notes for disclosures. Let's get into it. I am joined once again by Ed Zitron of the Better Offline Podcast and the Where's Your Ed newsletter. Ed, great to see you again. Welcome back.

Ed Zitron (0:22)

Great to be back.

Jack Farley (0:23)

Ed, the first time I had you on, you talked about your work on OpenAI's financials, which you found in 2025 We now have the financials for Anthropic. They just were leaked. Reuters reported when they filed their confidential S1 for their IPO. What was your reaction to these numbers? How would you characterize them?

Ed Zitron (0:42)

Well, I had a good chuckle. That was the first thing I went. I had a chortle.

But then I went and actually did the maths. Anthropic, at least in 2025, was a worse business than OpenAI.

It spent $2.75 to make a dollar, and OpenAI did $2.60 based on my own reporting. Anthropic lost $8 billion and $4.6 billion of revenue. But it's really the finer details that make it interesting. About 47 percent of Anthropic's revenue came through sales from Google and Amazon reselling their models, which is a worrying concentration risk considering Amazon fed $50 billion to OpenAI this year and sells them models. And Google actually provides compute to OpenAI, but Google has their own models with Gemini. They sell Grog. They sell everything. So it isn't necessarily anything dodgy, but there is a real risk there they have that basically their largest customers by proxy are also their competition. But fundamentally, there's a really simple way of putting this. The myth of Anthropic has been, since the very, very, very beginning, that this was the more profitable, more sustainable, more stable company compared to OpenAI when the literal opposite was correct at least in 2025 We do not have the 2026 numbers, which I think would shut the boosters up big time. But I think that this is the thing. I don't know if it's possible to have a profitable AI lab. I don't think it's possible. At the very least, we know that at least in 2025, Anthropic was in a worse condition than OpenAI.

Jack Farley (2:26)

Just how much worse were those numbers compared to OpenAI?

Ed Zitron (2:31)

They spent $2.75 to make a buck. OpenAI did $2.60.

Kind of the same, but Anthropic just mildly worse.

The big thing is, and while I didn't have in the numbers I got about OpenAI, the breakdown of where the revenue came from, we didn't have the same percentage thing, because that 47% disclosure is strictly something that Reuters got as a result of the Prospectus. I don't think that half of OpenAI's revenue came from Microsoft Azure. I think that people need to realize, sure, it's a counterparty relationship, it's very standard, what have you, but it is still bad because remember, Google and Amazon can discount Anthropic's tokens all they want. I've actually heard of it going the other way where Anthropic will discount tokens through AWS. It's all very weird and unstable and also just odd. I think that 2026 numbers are going to be so important to see because if this revenue concentration has got worse, this is on top of the fact that 80 percent of OpenAI and Anthropic's revenue comes from one percent of their customers who are predominantly AI startups, it's really becoming the case where the success of these companies is not a result of their sales operations or any particular innovation but counter-parties being able to lose money on their tokens or indeed resell their tokens to captured customers.

Jack Farley (3:52)

So a very high percentage of Anthropic's revenue is not from people going to cloud.com or businesses going to cloud.com.

Ed Zitron (4:02)

It's through Amazon and Google's reselling. Now there's nothing, I must be clear, and I'm a hater, don't get me wrong, but I want to be fair here. That is very standard for cloud software companies.

That high of a concentration though, I don't think that there's any comparison I've seen. It also means that their success is contingent on Google and Amazon's ability to sell their stuff, which they seem to be doing fine with. It's just odd. It's very odd.

Jack Farley (4:31)

Yeah. The narrative was the exact opposite that the risk for OpenAI is concentration. The whole thing about Anthropic is that it's less concentrated.

That's a narrative.

Ed Zitron (4:41)

In theory, people are already saying, well, who cares about these numbers? 2026 will be better. We've had these stories about Anthropic hitting profitability, but it's on an adjusted basis, which means it is not profitable. People are going to push part, and I've heard too many boosters saying, oh, just ignore these 2025 numbers, 2026 will be better. The thing is, they better be amazing compared to this, because if they're not, this is a company in distress. I'm also very curious how, let's take them at their word, in these very clearly dodgy leak stories, Anthropic managed to become profitable on some level in Q2, and that was likely due to Elon Musk's discount in their compute. That is one of the most aggressive turnarounds of costs I have seen in history. You are telling me this company that went from losing $8 billion in a year, turn that into a profit? How?