Anthropic S-1 due diligence: three red flags acquirers meet

November 11, 2026 · 8 min · Free, with full transcript

What does Anthropic's S-1 teach an acquirer, as bankers push its IPO valuation toward $1.5tn? The episode argues that disclosed risk is not priced risk. The filing shows three structures private targets carry without writing them down: about $900bn of fixed compute commitments against consumption revenue, a run-rate metric built on 28 days of usage times thirteen, and a board majority elected by a trust with no economic stake. Listeners learn to test three ratios that should sit near one: fixed commitments against contracted revenue, headline metric against real revenue, and votes against economics. Examples include CoreWeave matching loan tenors to customer contracts, Nvidia's key person gap, and the Weil, Gotshal and Manges wind-down.

Listen to the podcast

8 min

Chapters

Key takeaways

  • Compare fixed commitments with contracted revenue, because the commitment schedule does not move when demand or price does.
  • Check how a run-rate is built and measure the gap to actual annualised revenue, since a large gap in a flat business means the number was built in a spreadsheet.
  • Hold your run-rate haircut in a hot auction instead of letting it shrink to a trim.
  • Map votes against economic ownership, person by person, and treat a minority stake at a majority price as giving up the control premium.
  • Read for missing risk factors, such as an unnamed key person, because a missing disclosure does not mean the risk is missing.
Read the full transcript

Host:MBA Training, CFO edition. We get into Read Anthropic's S-1 like an acquirer would. A company that tells you the truth about its risks is more dangerous to buy than one that hides them.

Expert:Oh, come on. That's backwards.

Host:Is it? Anthropic files a prospectus with ninety pages of warnings, and the bankers still push the price toward one and a half trillion. Candour didn't slow anybody down.

Expert:Fine, that part stings. Disclosure works like a vaccine for the seller. "We told you." Nobody reads the part they were told.

Host:So that's myth one. Disclosed risk is priced risk.

Expert:And it's wrong. The S-1, the registration document a company files before listing, has three structures in it that every acquirer meets eventually. The filing doesn't make them interesting. What matters is that private targets carry the same three and never write them down.

Host:Start with the scariest.

Expert:Revenue that floats, costs that don't. Most of Anthropic's revenue is consumption, meaning customers pay as they use. Meanwhile it has signed roughly nine hundred billion of compute and infrastructure commitments running for years.

Host:Wait, nine hundred billion. Committed.

Expert:Committed. And the first version of the filing left that line blank. It only showed up in an amendment.

Host:The most important number in the book arrived in the sequel.

Expert:Like finding out who the killer is in the director's cut.

Host:Okay, but I'll defend them. Compute went from more than their entire revenue to about half of it within a year. That's a business learning to breathe.

Expert:It is. And then in the third quarter revenue grew strongly and compute grew faster. Some of that was expensive short-term capacity. The trend is real, and the fixed bill doesn't care about the trend.

Host:Speaking of things that don't care, The Daily Upside had a piece last week. Finance leaders are putting AI agents to work and arguing over how much authority to hand them. FP&A vendors, the planning and forecasting software crowd, Anaplan among them, are rebuilding their products around these agents.

Expert:Vendor-led, so I'd hold the enthusiasm at arm's length. The question is the right one, though. Authority without accountability is the same problem as the Anthropic board.

Host:Picture the agent proposing a nine-hundred-billion compute contract at two a.m.

Expert:"Approved. Confidence: high."

Host:It's going to be the most decisive CFO you've ever worked with. Right, myth two. The headline number is a figure. It's actually an opinion.

Expert:Run-rate revenue. Anthropic takes the last twenty-eight days of usage, multiplies by thirteen, and adds subscriptions times twelve.

Host:Hang on, say that again. Twenty-eight days?

Expert:Twenty-eight days times thirteen. In fairness, they say plainly it isn't GAAP, the standard accounting rules, and isn't a forecast. They even warn about holiday dips.

Host:Then where's the scandal?

Expert:There isn't one at Anthropic. Their run-rate sits roughly a fifth above last quarter's actual revenue annualised, and at their growth that's plausible. Now take a flat mid-market company that grabs its best four weeks and multiplies. It puts "exit-month annualised" on page three of the information memo, with no warning label.

Host:So the gap is the price of faith.

Expert:Exactly that. For a flat business, a gap that size means somebody built the number in a spreadsheet.

Host:Here's where I disagree. Every buyer knows run-rates are inflated. Everyone haircuts them. It's theatre both sides understand.

Expert:Everyone says they haircut. Then the auction gets hot, the banker says "covered," and the haircut turns into a trim.

Host:That's an argument about discipline. The metric isn't the problem.

Expert:The metric is a tool for wrecking discipline. That's what it was built for.

Host:Spoken like someone who's never had to sell a company.

Expert:Spoken like someone who's had to explain to a board why the bought revenue... evaporated.

Host:...in the first August. Yeah. Okay. Half a point to you.

Expert:I'll take half.

Host:Myth three is the one I actually hold. If you own the shares, you own the company.

Expert:Not here. A trust with no economic stake elects four of seven directors. A single founder share carries just over half the votes. Anthropic's own filing says the setup raises the risk of governance litigation.

Host:Wait, what? The company warns you it might get sued over its own structure?

Expert:In writing. Public investors get the economics without the control. In deal terms you're buying a minority stake at a majority price.

Host:And yet the order book is covered. People want in.

Expert:Because they've decided the control premium, the extra you normally pay to steer the thing, is worth giving up for the ride.

Host:That's rational. I'd do it.

Expert:Maybe. But this week's news shows control living in odd places. Nvidia's board hired Wachtell to investigate after Jensen Huang admitted an affair with a woman linked to Chinese intelligence. And Nvidia's annual report doesn't name him as a key person.

Host:The most key person in modern capitalism isn't listed as a key person.

Expert:A missing risk factor doesn't mean the risk is missing.

Host:And Weil.

Expert:Weil, Gotshal and Manges. Ninety-five years old. Partners left one after another, confidence went, and the partners voted to wind it down. Rescue talks with Vinson and Elkins and Willkie Farr went nowhere.

Host:That one lands differently. Retirees still owed money, people who built the place watching it close by vote.

Expert:Yeah. It's grim. And for anyone buying a consulting or advisory firm, the asset walks out the lift every evening.

Host:So underneath all three myths, what's the principle?

Expert:Ratios that ought to sit at one. Fixed commitments against contracted revenue. The headline metric against real revenue. Votes against economics, person by person. When one of them drifts far from one, the income statement stops being the story.

Host:Good operators already think like that.

Expert:The good ones look at CoreWeave. Its unsecured debt was yielding over sixteen percent, and its answer was to match loan tenors to customer contracts, so the debt is repaid when the contract ends. Commitment and revenue expire together.

Host:That's elegant.

Expert:It isn't perfect. Prepayments still trail the capex bill. But it's the shape you want.

Host:Which is where I still push back. Anthropic will probably price brilliantly on the eighteenth and every one of your ratios will look like pedantry.

Expert:Demand can move the price by hundreds of billions in a week. It doesn't move the commitment schedule by a dollar.

Host:Unless revenue grows into it.

Expert:Unless. That's the whole bet.

Host:So we're back at the opening. The honest company is the dangerous one.

Expert:Honest companies are fine. The danger is buyers who think reading it equals understanding it.

Host:Or buyers who let the AI agent read it.

Expert:"Confidence: high."

Host:Somewhere in every deal there's a nine-hundred-billion line. The only question is whether you go looking for it, or it finds you after closing. Sources: Techflowpost, Bloombergtax and Yahoo Finance. Every link is in the show notes. We'll stop there. The CFO tools for making the call are at mba-training.com.

Read and go deeper