THE SHORT ANSWER
● Anthropic — S-1 expected today, $16.3B H1 revenue, $450B cloud obligations disclosed
● OpenAI — up to $30B at $1.25T, led by UAE investors, nothing disclosed
● The real difference — one becomes auditable, the other does not
● Matters if — you are picking a model to build a product on
Side by side
| | Anthropic | OpenAI |
|---|
| Method | IPO, S-1 expected today | Private round |
| Size | Listing, November possible | Up to $30 billion |
| Valuation | $1.8-2T banks, ~$1.5T Reuters, ~$150B New Constructs | $1.25 trillion |
| Lead | Morgan Stanley, Goldman Sachs expected | UAE investors |
| H1 revenue disclosed | $16.3 billion | Not disclosed |
| Compute obligations disclosed | ~$450B over 7-10 years | Not disclosed |
| Customer concentration | ~20% from two customers | Not disclosed |
| Ongoing reporting | Quarterly, audited | None |
What disclosure actually buys you
If you are an engineer picking an API, none of this looks like your problem. It is, in three specific ways.
Price stability. Anthropic has roughly $450 billion in mostly non-cancelable cloud obligations over seven to ten years, potentially $750 billion if pending hardware deals close. That is now a number you can hold them to. You cannot know OpenAI equivalent, which means you cannot reason about where either company has to take pricing.
Concentration risk. About a fifth of Anthropic revenue comes from two customers. If one leaves, the pressure lands somewhere, and historically it lands on the pricing of everyone else. OpenAI may have the same exposure or none. There is no way to tell.
The insurance disclosure. Anthropic states in its risk factors that catastrophic AI-related damages may be so large the company cannot obtain sufficient insurance. No AI company has put that in a filing before. Whether OpenAI shares that assessment is unknowable, because nobody has to ask it under oath.
The case for staying private
It is not a weak one. Quarterly reporting pushes companies toward decisions that look good in ninety days, which is a poor fit for research with multi-year horizons. Disclosure also hands competitors a map — Anthropic customer concentration is now something OpenAI sales teams can act on. And sovereign-linked capital is patient in a way public markets are not.
OpenAI has said it will not list until 2027. On this evidence that looks less like reluctance and more like a preference for raising at $1.25 trillion without explaining how the money gets spent.
Which should you build on
- If you need to assess vendor risk formally — procurement, compliance, a board that asks questions — Anthropic becomes the only one of the two you can answer about
- If you are optimising purely on model capability and price — this changes nothing today. Benchmark both
- If you are worried about pricing shocks — the disclosed obligations are a floor under what Anthropic must eventually earn. That is uncomfortable reading, but it is better than not knowing
- Either way, do not single-source — both are financing a bet at a scale where being wrong is expensive
Frequently asked questions
Does an IPO make Anthropic safer to depend on?
More legible, not safer. You learn the obligations; you do not remove them.
Will OpenAI have to disclose eventually?
Only if it lists. It has indicated 2027, which is a stated intention rather than a commitment.
Why is New Constructs valuation so much lower?
Because it values fundamentals rather than estimating placement demand. $150 billion against $1.8-2 trillion is the gap between what the business earns and what the market may pay.
Sources
BigGo Finance · CNBC