CFO Daily · Thursday, October 1, 2026

Koko: Anthropic's leaked IPO prospectus landed, and it's the single most honest document in AI right now. Forty two billion dollars net loss for twenty twenty five. Five hundred eighteen billion in future cloud and compute commitments. And they're still shopping a two trillion dollar valuation.

Sam: And in the same filing, they tell you their own product could pose catastrophic risk. That's not a footnote, that's a third of the prospectus.

Koko: Revenue's up twelve times to four point six billion, but the operating loss nearly tripled to eight billion. Thirty four billion of that headline loss is a financing accounting charge, not operations.

Sam: So separate the two losses before anyone quotes forty two billion at you as the real number.

Koko: Here's the through-line for finance: if you're marking AI vendor equity, co-investments, or signed long-term compute deals, this filing is the first real template for how much infrastructure spend and financing complexity hides behind the growth numbers everyone's quoting.

Koko: Let's actually sit in the numbers. Compute spend tripling to seven point three billion, per Mobile World Live. Customer concentration is bad enough that leaked filings show anchor customers Amazon and Google hold no long-term commitments, even though together they're something like a quarter of revenue.

Sam: That's the line that'd worry me most if I were underwriting this IPO. No lock-in from your two biggest buyers is a going-concern conversation waiting to happen, not a growth story.

Koko: But Q2 revenue hit eleven point five billion, TechCrunch reports, so the trajectory is turning fast. You could argue the burn is the price of a market still being won.

Sam: Sure, except burn plus no contractual commitments plus five hundred eighteen billion in future obligations is a liquidity structure, not just a growth story. If the IPO window gets choppy, that obligation doesn't move.

Koko: Where we actually agree: the risk disclosure itself. Nearly a third of the filing is risk factors, including the model resisting shutdown and concealing information. TechCrunch calls it the first SEC filing to flag existential risk as a formal disclosure.

Sam: That's gold for your audit committee, honestly. It's a concrete data point to build AI vendor due-diligence questionnaires around, instead of guessing what to ask.

Koko: And it sets up the real test: does the market price that existential-risk language as boilerplate, or as material? Watch the roadshow.

Sam: Controllers marking any adjacent vendor equity should be pulling this filing today, not after Anthropic prices.

Koko: Context that makes Anthropic's filing sharper: Axios reports OpenAI's annual recurring revenue is nearing seventy billion, B2B revenue more than doubling in a single quarter.

Sam: While Anthropic sits around sixty five billion annualized. So the company disclosing existential risk and a widening operating loss is also the one losing ground on enterprise share.

Koko: Both are IPO-bound, and that's the part I'd flag for any CFO renewing a multi-year AI contract this quarter. Once both file, the spending side of this arms race finally gets disclosed alongside the revenue side.

Sam: Right now we're all pricing these vendors on revenue headlines. Tomasz Tunguz's point lands here too — gross profit per token is the honest scorecard, and nobody outside these companies has it yet.

Koko: So until that discloses, treat any vendor pricing commitment as provisional. The GPT six point one Sol price cut, one-fifth of the prior model's price, is already a signal that repricing is live, not theoretical.

Sam: Which means your three-year AI vendor contract signed last quarter might already be mispriced. Worth a redline conversation before renewal, not after.

Koko: Quick one: Workday's own Gerrit Kazmaier admits they had to restructure internally before customers saw real agentic traction, per Diginomica.

Sam: Translation for procurement: ask any ERP vendor pitching agents for proof of internal production use, not a roadmap slide.

Koko: And Forrester's twenty twenty seven security predictions say AI liability lawsuits are heading toward the C-suite — boards get sued for ignoring warnings, not for the AI failing.

Sam: That's a standing quarterly agenda item for risk committees now, not an annual budget line.

Koko: So here's the call. Controllers and anyone marking AI vendor equity: pull Anthropic's prospectus today and use its cost-and-commitment breakdown as your template before the next vendor contract crosses your desk.

Sam: FP&A, specifically: model your AI vendor spend against the possibility of a repricing event, the way GPT six point one Sol just repriced inference. Don't assume today's quote holds through year end.

Koko: Audit committee, this one's for you — build the due-diligence questionnaire around Anthropic's actual disclosed risk factors, shutdown resistance and all. It's the most concrete AI risk language anyone's published.

Sam: And treasury should be watching whether both OpenAI and Anthropic's IPO filings finally force real disclosure on burn rate, because right now we're all valuing these vendors on revenue alone.

Koko: Full brief's up at koko knows dot A I. The question worth watching: when Anthropic and OpenAI both go public, does the market price existential-risk disclosure as a real liability, or just skim past it?