Brian Reed: Hey — did you get through the whole thing?
Eliza Ward: The S-1? Most of it. It's a lot.
Brian Reed: The part that stopped me — and I want to check this against what you read — is that the valuation number everyone's treating as settled isn't actually in the filing. Like, the one-point-five to two trillion figure.
Eliza Ward: Correct. The prospectus Anthropic filed on October 27th leaves price and share count completely blank. What's in there is the revenue: sixteen-point-two-nine billion for the first half of 2026.
Brian Reed: That's — okay, that number is genuinely wild to me. Half a year of Claude revenue.
Eliza Ward: Right. And the company — Anthropic, PBC — is listing on Nasdaq. Dario and Daniela Amodei are the co-founders. That's the settled part. What isn't settled is what anyone's actually paying for a share.
Brian Reed: So the two trillion we've been seeing everywhere — that's someone's inference, not the company's stated number.
Eliza Ward: Press reports, unnamed sources. The roadshow is going to be the first real test of whether that number holds.
Brian Reed: But here's what the revenue number doesn't tell you on its own — it's not just that Anthropic grew, it's *how* the existing customers behaved. The net revenue retention is around 900%. Which is, I mean, let me just say that plainly: if your tenants don't just renew the lease but rent out the whole building the next year, that's 900% NRR. That's what's actually happening inside that sixteen-billion number.
Eliza Ward: Right — and H1 2026 alone beat full-year 2025 revenue. Full year was about four-point-five-nine billion. The first *half* of this year was sixteen-point-two-nine. That gap is not a rounding error.
Brian Reed: No. But who are these customers actually expanding that fast?
Eliza Ward: Wait, that's exactly the problem. There are over 11,900 customers above a hundred-K threshold, which sounds broad. But then 250 of them are at ten million or more annually. And those 250 are probably where most of that 900% NRR is actually living.
Brian Reed: So it's concentration risk dressed up as a retention metric.
Eliza Ward: That's — yeah, basically. And here's where it gets actually fragile, not just theoretically fragile. OpenAI is moving. Google has its own models competing against Claude. And then Moonshot — their open-weight model Kimi K3 — is explicitly cited as a low-cost threat to premium Claude pricing. So the 900% NRR is real, I think the number is real, but it was generated in an environment where Claude had pricing power. I don't — actually, I'm not sure that environment survives 2027.
Brian Reed: Hold on. So the growth trajectory that made H1 2026 look like fourteen times H1 2025 — that assumed customers kept paying premium rates for Claude while Kimi K3 exists and OpenAI is cutting prices?
Eliza Ward: That's the inference, yeah. The S-1 flags those competitors. It doesn't model what happens to NRR if gross margins compress because Claude has to reprice to stay competitive. That's the number I want to see that isn't there.
Brian Reed: So the growth is real — the fourteen-times year-over-year jump from about 1.16 billion to sixteen-billion-plus, that happened — but whether that *rate* holds is a completely open question, and the filing doesn't answer it.
Eliza Ward: Right, and that open rate question is actually — wait, let me pivot, because there's a take circulating that I think is genuinely wrong and it's making this harder to read. The take is: eighty-point-seven billion dollar net loss, company's a disaster, full stop.
Brian Reed: Yeah, that's the one. And it's too crude.
Eliza Ward: It's too crude, but the correct version is actually more unsettling. Because those same six months — H1 2026 — produced three-point-nine billion in positive operating cash flow. Up from negative one-point-five billion in H1 2025. So the company flipped to cash-flow positive while posting an eighty-point-seven billion accounting loss simultaneously.
Brian Reed: Actually, hang on. How does that work? Like, what is the eighty-point-seven billion even measuring?
Eliza Ward: That's exactly it. The S-1 doesn't fully reconcile it. The loss almost certainly includes non-cash items — mark-to-market on compute commitments, convertible note accounting — but the filing doesn't break out how much. We don't know what percentage of eighty-point-seven billion is actual cash leaving the building versus accounting treatment on those Google and Amazon obligations.
Brian Reed: Which means a retail investor who sees 'three-point-nine billion positive cash flow' in a headline could buy in thinking the losses are basically cosmetic — and only untangle the eighty-point-seven billion figure much later.
Eliza Ward: Hold on — that's the JPMorgan thing. JPMorgan Chase is directing fifteen to twenty percent of IPO shares to retail investors. Jamie Dimon is reportedly going to personally participate in that retail campaign.
Brian Reed: Wait, fifteen to twenty percent to retail — on a deal where the accounting this hard to parse? I mean, I don't know if that's confidence or — let me see — is that distributing the downside risk?
Eliza Ward: We genuinely don't know. And that question gets worse once you look at what the Google compute commitment and Amazon's convertible notes actually mean for Anthropic's structural independence — that's the part I think changes the whole picture of who's really exposed here.
Brian Reed: And that structural exposure — that's the part that actually changes the question, because it's not just 'will Anthropic grow.' It's 'who captures the upside if it does.' The Amazon convertible notes alone — six-point-two billion outstanding as of June thirtieth, with a conversion price as low as two-dollars-and-nine-cents a share.
Eliza Ward: Two-oh-nine. At a two-trillion-dollar target valuation.
Brian Reed: Right — that's the thing. If Amazon converts at two-oh-nine per share and the IPO prices anywhere near a valuation that implies, I don't know, fifty or a hundred dollars a share, that's massive dilution the public buyer is walking into. And Amazon already has a twelve-percent ownership cap, so it's capped, but still.
Eliza Ward: The Google side is actually — wait, this is the one that stops me. Anthropic committed to spend at least a hundred-and-fourteen-point-five billion dollars on Google compute, running through July 2033. That's seven years of current revenue, obligated to a single vendor who is simultaneously running Gemini against Claude.
Brian Reed: A competitor. Literally a competitor.
Eliza Ward: And some coverage puts total long-term compute obligations up to four-fifty, maybe five-eighteen billion when you add everything out. So if a model provider — say, Moonshot undercuts Claude on price in 2028 — Anthropic still owes Google that compute spend. The contract doesn't flex with the revenue.
Brian Reed: So the question for the roadshow — Goldman Sachs, Morgan Stanley, twenty-two banks in the syndicate, all of them pitching this — is whether public investors actually price in that obligation or whether they get anchored to the cash-flow-positive headline.
Eliza Ward: That's exactly what to watch. Spark Capital — only pure-play VC in the S-1, Yasmin Razavi on the board, roughly seven-seventy-seven million in since 2023 — they already priced in enormous upside at private valuations. The roadshow is the first moment public markets can actually push back. And the SpaceX comparison is — I mean, SpaceX valued at one-point-seven-seven trillion in June, and they had reusable rockets and a contract backlog. Anthropic is targeting two trillion on six months of cash-flow positive and a compute obligation it cannot unwind.
Brian Reed: The roadshow. That's the signal — whether those numbers hold when underwriters actually have to defend the compute detail in front of a portfolio manager who's read the full filing.
Eliza Ward: And that's actually the first time a public market professional is on record on any of this — Denny Fish at Janus Henderson commented on the S-1 for CNBC. That's not a VC with skin in the game. That's a portfolio manager looking at the filing cold.
Brian Reed: Which means — hang on — we're at the very beginning of public price discovery. Everything before Denny Fish was private capital's judgment.
Eliza Ward: The S-1 leaves valuation blank. That's not an oversight. So the one-point-five to two trillion is still, right now, an inference. The roadshow hasn't run. Public markets haven't spoken.
Brian Reed: So the actual open question — the one that doesn't resolve until next quarter at the earliest — is whether the compute commitments start showing up harder in the cash flow numbers. Because if they do, the three-point-nine billion operating cash flow story gets a lot more complicated. And I don't — I mean, I genuinely don't know if the margin story holds when those Google obligations start hitting at full weight against whatever revenue growth actually materializes.
Eliza Ward: Neither do I. That's the honest answer.
