Jordan Hale: Hey — did you see the SpaceX IPO hit two trillion on opening day back in June? Because I keep thinking about that as like, the backdrop for everything we're about to talk about.
Alex Mercer: I did, yeah. Two trillion on day one sets a strange benchmark. Why, where are you going with that?
Jordan Hale: Because four months later — October 20th — the Wall Street Journal and Reuters both publish detailed figures out of Anthropic's draft IPO prospectus, and the number that everyone grabbed onto was this eighty-point-seven billion dollar net loss in the first half of 2026. And I keep wondering if the SpaceX moment had just primed everyone to accept absurd valuations, you know, so the loss just felt like... whiplash.
Alex Mercer: That's an interesting frame. Though I'd actually separate those two things — the valuation story and the loss story have different mechanics. Anthropic filed its draft S-1 with the SEC on June 1st, right after a Series H that raised sixty-five billion at nine hundred and sixty-five billion post-money. So by the time October 20th hit, this wasn't some startup number people were reading. It was a near-trillion-dollar company reporting what looked like a catastrophic loss.
Jordan Hale: And the revenue side of that filing — sixteen-point-three billion in H1 2026, up from one-point-one billion the prior year — that part kind of got buried under the loss headline, which, like, that's the whole problem.
Alex Mercer: To some extent that's a financial literacy problem. To some extent it's also a real question about what the loss actually represents — and those are not the same issue.
Jordan Hale: Okay — so that's what we're untangling today. What that number really is, what the filing actually says, and whether Anthropic's path to the public markets is a story about triumph or something more... complicated.
Alex Mercer: What the loss actually represents — that's exactly where I want to start. Think of it this way: imagine a friend buys a house, the bank marks it up five hundred thousand dollars on paper, and then when rates shift, they have to revalue it back down. Their accounting shows a five-hundred-thousand-dollar loss. The house didn't burn down. No cash left. That's basically what's happening with Anthropic's convertible notes — roughly seventy-two billion of that eighty-point-seven billion loss is accounting charges from revaluing those instruments after earlier fundraising rounds. Not cash out the door.
Jordan Hale: Okay wait — so the sixty-five billion Series H in May, the nine-hundred-sixty-five billion valuation, that's what triggered the revaluation? Like, the financing model generated the scary headline?
Alex Mercer: Exactly that. Each high-valuation round forces the previous convertible notes to be marked at the new implied price. So the headline loss is basically a byproduct of Anthropic successfully raising money at increasingly high valuations. The financing structure produced the number people panicked about on X between October 15th and 20th.
Jordan Hale: That's — okay, that's almost funny. The better the fundraising goes, the worse the headline looks.
Alex Mercer: Right — but here's where I'd pump the brakes on any relief. Strip out the seventy-two billion and you're left with nearly nine billion in actual compute and infrastructure spending. In six months. H1 2026 alone.
Jordan Hale: Wait — so we traded one scary number for a different scary number?
Alex Mercer: Pretty much, yeah. And here's the part that I think actually lands harder — that nine billion in H1 2026 already exceeded the seven-point-three-three billion Anthropic spent on compute across all of 2025. I mean, the entire prior year. In six months they blew past it. That's not a ratio improving; that's absolute spend accelerating.
Jordan Hale: No way. The full year 2025 figure — gone by June?
Alex Mercer: Gone by June. And some of that is Nvidia GPU rentals — up to eighty-six billion cited in the filing. So the non-cash story is technically correct, the commentators who said the headline was misleading weren't wrong, but — I think the mistake is treating the correction as reassurance. The mortgage payment is still very real, even if the house didn't burn down.
Jordan Hale: And that mortgage metaphor is doing a lot of work, right, because — wait, no, let me back up — the revenue side of this actually is extraordinary. Like, Q3 2026 alone, Anthropic separately told prospective investors: sixteen-point-eight billion. One quarter. That's up roughly forty-five percent from Q2. Almost no software company in history has grown at that pace.
Alex Mercer: The Claude trajectory is genuinely hard to dismiss. That's real.
Jordan Hale: And compute spending falling from over two hundred percent of revenue down to fifty-five percent in H1 2026 — that's the unit economics story, and it's directionally good. But then you hit the commitments number and it just... stops being a ratio story.
Alex Mercer: Because a ratio improves as revenue grows. A fixed liability doesn't.
Jordan Hale: Google committed a hundred and eleven billion. Microsoft thirty-one billion. Nvidia GPU rentals up to eighty-six billion. And total obligations potentially pushing to seven hundred and fifty billion over seven to ten years. That's — I mean, picture a CFO in November sitting with those numbers trying to set an IPO price against a two-trillion-dollar target. What is she even anchoring to?
Alex Mercer: SpaceX, basically. SpaceX opened at two trillion in June, and that became the reference point. Whether it should be the reference point is a different question.
Jordan Hale: Right — but the part that doesn't fit is, we don't actually know how the obligation schedule phases. Some of this might be conditional. So the seven-fifty isn't necessarily a cliff, it could be staged. I think? Though I'm genuinely not sure the filing spells that out cleanly.
Alex Mercer: I think that's the honest hedge. The scale still dwarfs current revenue by orders of magnitude regardless of phasing. And the part we haven't even gotten to yet — who's actually buying Claude, and what does that tell us about what the revenue is really validating — that reframes this whole picture.
Jordan Hale: Yeah. That's the thread I keep pulling.
Alex Mercer: Right — and who's buying is the part that breaks the clean revenue story. Two customers, one-fifth of revenue. The filing flags it as a material risk. That's not buried footnote language; that's Anthropic telling the SEC, hey, this is structurally fragile.
Jordan Hale: And those two customers are — wait, it's Google and Amazon, right? The same companies that are also the infrastructure lenders?
Alex Mercer: Google, Microsoft, Amazon — all three are simultaneously investors, committed cloud spenders, and the infrastructure Anthropic runs Claude on. So think about what happens if, say, a regulatory action — DOJ, EU, doesn't matter — forces Google to restructure its Anthropic relationship. Their revenue line and their compute capacity get impaired at the same moment. One trigger, two simultaneous hits.
Jordan Hale: That's — I mean, that's not a business risk, that's more like a pressure point where one squeeze affects everything at once.
Alex Mercer: Basically, yeah. And I think the honest question is whether the revenue growth — sixteen-eight in Q3, up forty-five percent — I mean, is that independent demand for Claude, or is it partly hyperscalers buying Claude because they need to own the model layer of their own infrastructure? Those are different things. One validates the product. The other validates the deal structure.
Jordan Hale: Okay but then — and this is the part that just floors me every time I come back to it — you have this IPO filing, the leaked version CNN reported on September 29th, and it literally says Anthropic's models pose an existential risk to humanity. In the document pitching a two-trillion-dollar valuation. How does that sentence survive legal review?
Alex Mercer: I mean — the public benefit corporation structure is probably part of the answer. Anthropic's legal mandate is literally to balance profit with societal benefit, so disclosing existential risk isn't contradicting the pitch, it's almost... required by the charter. But that doesn't resolve the SEC question. Could the commission treat existential risk language as an undisclosed material liability? I genuinely don't know. That's novel regulatory territory.
Jordan Hale: So it either gets filed and forgotten, or it becomes — like, actually live? As a liability question?
Alex Mercer: That uncertainty is real and unresolved. And with OpenAI also moving toward its own IPO, whichever company gets to public markets first sets the disclosure precedent for the whole sector. That makes the existential-risk language matter a lot more than a checkbox.
Jordan Hale: And that's sort of what keeps striking me — Wilson Sonsini hired in December 2025, Eric Schmidt circling as an investor reference point, the whole apparatus of a serious IPO in motion. And yet the document that apparatus produced contains a sentence about existential risk to humanity. Both things are just... true at once.
Alex Mercer: If it prices at two trillion, somebody read that filing — the existential risk clause, the seven-fifty billion in obligations, all of it — and bought in anyway. That's the only thing I'm actually certain of.
Jordan Hale: Yeah. I mean — I don't know if that's terrifying or just very human. Probably both.
