Brian Reed: Hey — glad we're doing this today specifically.
Eliza Ward: Yeah, the timing is — it's genuinely live.
Brian Reed: Anthropic held a private meeting this morning — October 14th — with institutional investors. First look at Q3 financials. No public filing, no S-1 yet. Just selected disclosure.
Eliza Ward: Hold on — CNBC reported this yesterday, the 13th. They're calling it potentially the largest IPO in U.S. history.
Brian Reed: Larger than SpaceX, which filed targeting $1.75 trillion in May. So Anthropic's going above two trillion.
Eliza Ward: And Reuters already leaked the prospectus — September 28th. So we have the 2025 numbers. $4.6 billion revenue, twelve times year-over-year growth.
Brian Reed: And a $42 billion net loss.
Eliza Ward: Right. So the confirmed event is: today's meeting happened. The numbers are out. And the gap between those two figures — that's actually what needs explaining.
Brian Reed: The gap — right, but here's the part that doesn't fit: that $42 billion loss sits next to a $65 billion annualized run rate as of July. That's the number that's been circulating. But where did it come from?
Eliza Ward: That's — okay, this is the thing. It came from an unaudited leaked document. Not a filed S-1. Anthropic submitted a draft registration statement to the SEC on June 1st, and the $65 billion figure is from internal reporting as of July, sourced through unnamed backers. That's not the same as an audited filing.
Brian Reed: So — wait, let me just say this plainly. It's like a restaurant telling you it's on track for a record year based on January receipts. The trajectory might be real. But you haven't seen the books.
Eliza Ward: And you don't know if they're counting the catering arm the same way as the dining room.
Brian Reed: That's — yeah, that's exactly the unresolved piece. Because we watched OpenAI's cloud-partner revenue methodology get disclosed and it rattled markets. That wasn't abstract. It moved numbers.
Eliza Ward: Right, and the question for Anthropic's S-1 is whether they count cloud-partner revenue the way OpenAI did. We don't know yet. That methodology footnote — actually, that's not a minor detail, that's the detail. Because the SEC already flagged heightened scrutiny of hyperscaler AI partnerships in its review of the June 1st draft.
Brian Reed: So this is a regulatory question, not just an investor-confidence question.
Eliza Ward: Both. And then the backers — unnamed — are projecting $100 to $120 billion by year-end. That's not from Anthropic. That's from people who already have money in and need the valuation to hold. I mean, that jump from $65 billion in July to $100-plus by December requires assumptions about margin and pricing power that nobody has confirmed.
Brian Reed: And the thing we actually can't answer yet — will the S-1, when it's final, show the same methodology? That's the missing fact. Not the run rate number. How they got there.
Eliza Ward: And that methodology question is actually what breaks the take I keep hearing — that twelve-times revenue growth proves the model works. It doesn't. Not when the cost structure is this decoupled.
Brian Reed: That's the one I want to correct. Because demand existing and unit economics working — those are completely different claims.
Eliza Ward: What about the customer base, though? Anthropic's investor materials cite eight of the Fortune 10, roughly seventy percent of Fortune 100. That's not nothing.
Brian Reed: No, it's real — I'm not disputing the customer list. But those contracts, however large, don't close a $42 billion annual loss. I mean, think about what's sitting on the other side of that revenue line. $518 billion in signed compute commitments. Locked in. That number doesn't move if revenue slows.
Eliza Ward: Wait — locked in regardless of near-term revenue. So even if Claude adoption flattens in Q4—
Brian Reed: The compute bill doesn't flatten with it. That's the thing. You can't point at Fortune 100 penetration and call it proof of breakeven when your fixed obligations are eight times your annualized revenue.
Eliza Ward: And then whatever Anthropic shows investors today — any adjusted operating profit figure — you have to ask what's been excluded to get there. Because the GAAP loss is $42 billion. That gap is a disclosure risk, not a rounding error.
Brian Reed: Right — and the Series H actually confirms this. Anthropic closed that round in May 2026 at a $965 billion post-money valuation. Investors aren't funding profitability. They're funding scale-up. Those are bets on a future that hasn't been modeled publicly yet.
Eliza Ward: Which is what makes the $2 trillion target the real question — and that's actually where the valuation math depends on things we haven't gotten to yet. The SpaceX benchmark, what Dario Amodei's own safety argument does to the story, OpenAI's competitive pressure — that's the part that makes this cost problem look different.
Brian Reed: The SpaceX number is what makes the $2 trillion target feel less like a valuation and more like a position statement. SpaceX filed in May 2026 at $1.75 trillion — and that already folded in xAI via an all-stock deal that closed February 2nd. So Anthropic isn't just naming a number. They're saying: we're worth more than a rocket company that absorbed Elon Musk's AI division.
Eliza Ward: And analyst estimates — what's the range on IPO-day market cap?
Brian Reed: One-point-five to four trillion. That's not a range — that's, I mean, that's a fundamental disagreement about what this company is.
Eliza Ward: Wait, $1.5 to $4 trillion? Okay, picture a pension fund manager. She's sitting in her office right now, looking at that spread, looking at a $2 trillion ask, and the underlying run rate is unaudited. There's no settled number to anchor to. She can't build a position on a range that wide.
Brian Reed: And then the Dario Amodei piece lands on top of that. He's on record — publicly — calling for the AI industry to scale back new capability launches on safety grounds. And simultaneously he's leading the push toward the largest IPO in history specifically to fund... more capability scaling. I don't have a clean way to resolve that.
Eliza Ward: It's not just a tension in the abstract — GPT-6 Astra dropped right before this investor meeting window and it's gaining ground with business customers right now. That's Anthropic's core market. So Amodei is making a safety argument while OpenAI is actually moving faster, and Anthropic is asking investors to price in safety leadership as a premium. That's a hard sell when the competitive clock is running.
Brian Reed: And Sam Altman called going public right now 'ill-advised' — that was September 2026. Safety concerns. Then Sarah Friar told CNBC OpenAI is targeting 2027 for its own IPO. So Altman is publicly tapping the brakes while his CFO is already setting the calendar.
Eliza Ward: Right — and Anthropic is structured as a public benefit corporation. Which is actually the thing nobody's priced. Because public benefit corporation obligations and maximizing return for that pension fund manager — those can pull in opposite directions. The S-1, when it's final, has to address that directly. That's the disclosure signal to watch.
Brian Reed: So the question isn't whether the $2 trillion number holds. It's whether the methodology footnote, the PBC structure, and the safety-versus-scale contradiction — whether any of that surfaces before investors commit. Because once the S-1 is public, those heroic assumptions stop being assumptions.
Eliza Ward: And the public benefit corporation piece — it's a governance footnote, but it's not just that. That's the thing that could legally constrain how Anthropic responds once public shareholders have actual fiduciary claims on a $2 trillion investment. Dario Amodei staked the company's identity on safety-first positioning. But once those investors are in — does that framework stay a binding constraint, or does it become a reputational asset that justifies the premium without actually limiting anything?
Brian Reed: That's — I mean, I don't think anyone has answered that. Not in the materials we've seen.
Eliza Ward: We don't have the S-1. That's the honest state of it. What we have is an investor meeting that happened today, numbers that aren't audited yet, and a valuation the market hasn't confirmed.
Brian Reed: And the institutional investors who built conviction at this morning's private meeting — if the final S-1 shows the GAAP picture is materially worse than whatever adjusted figures circulated today, they don't get a do-over. That's actual exposure. And it's the same structural question for OpenAI's 2027 timeline — if Anthropic prices at $2 trillion and then resets, does that compress how aggressively Sarah Friar can bring OpenAI public? I genuinely don't know. That feels like a live variable, not a settled one.
Eliza Ward: The S-1 is where this either holds or it doesn't. That's the next real fact.
