Priced for a Monopoly

Within a week of the BIS naming circular AI financing as a systemic risk, two of the entities most exposed to that structure filed to list. OpenAI at a reported valuation near eight hundred and fifty two billion dollars. Anthropic at a reported valuation near nine hundred and sixty five billion dollars. Both filings are confidential, and both are draft. The SEC has acknowledged receipt. The reported numbers come from wire reporting on term-sheet material shared with underwriters and select institutional accounts, not from published S-1 documents. That distinction matters for what a reader can responsibly conclude, and it is the frame for everything that follows.

What the reporting establishes

Three things are on the record with a reasonable degree of confidence. First, both firms have submitted confidential registration statements. Second, both are being marketed at valuations near a trillion dollars. Third, both are being marketed to public markets alongside disclosed multi-hundred-billion-dollar forward compute commitments to a small set of counterparties. What is not yet on the record, because the filings are confidential, is the actual revenue table, the actual commitment schedule, the customer concentration, or the governance structure being proposed.

A responsible reader treats every number in the news cycle over the coming weeks as provisional until the S-1s go public. The exercise here is not to price either name. It is to identify what those documents will have to show for the reported valuations to be defensible.

What a defensible price requires

We laid out a three-layer framework in the SpaceX package, and the logic transports. Value the cash engine on comparables. Value the strategic optionality on probability-weighted present value. Whatever remains between the sum of those layers and the ask is narrative premium, and the honest treatment is to name it and size it rather than absorb it into the base case.

For OpenAI at eight hundred and fifty two billion dollars, that framework demands one of two things. Either the disclosed run-rate revenue is materially larger than the reported twenty-five billion dollar figure, or the outside-revenue trajectory in the next five years is credible enough at that scale to justify a valuation multiple that no software company at that revenue base has ever supported. Both are possible. Neither is currently visible in the material that has been shared.

For Anthropic at nine hundred and sixty five billion dollars, the ratio is more strained. Reported run-rate revenue near seven billion dollars implies a revenue multiple that only makes sense if the entity is priced as a near-monopoly beneficiary of a specific model of enterprise AI adoption. That is a possible outcome. It is not the only possible outcome. A responsible public-market disclosure has to price for a range, not for the single scenario in which it works.

A responsible public-market disclosure has to price for a range, not for the single scenario in which it works.

The circular exposure the filings will have to disclose

Both filings will be read against the BIS report. Underwriters know this. The commitment schedules that appear in the S-1s will be scrutinised for intra-circle concentration in a way that pre-BIS filings would not have been. Two specific disclosures matter more than the rest. First, the share of forward compute commitments held with counterparties that are also equity holders. Second, the share of run-rate revenue that originates from counterparties inside the same commitment book. If the first is high and the second is meaningful, the filing will attract governance and disclosure comment letters from the SEC in a way that materially delays the listing timeline.

The FCA and ESMA are expected to require analogous disclosures for any secondary listings or European institutional marketing. There is a small but non-trivial chance that the two filings become the first test cases for a new class of counterparty-concentration disclosure that regulators have been circling for two years.

The SpaceX precedent, cleanly stated

The SpaceX listing at a fixed one hundred and thirty five dollars per share and roughly one point seven seven trillion dollars implied valuation created a live precedent for founder-controlled, narrative-heavy, retail-tilted listings. Two features of that deal will now be studied by the underwriting teams working on the AI filings. The fixed price structure that removed book-build discovery. The elevated retail allocation that pushed early flow toward accounts that had not seen the full disclosure. Whether either filing adopts elements of that structure will be visible when the S-1s go public.

For a public-market reader, the SpaceX experience the day of listing is the closest available precedent for what a heavily-hyped AI listing will look like on day one. Volatile, retail-heavy, with a price that references the deal terms more than any cash-flow model for the first several weeks. Anyone entering these listings should size for that reality and use the lock-up expiry, not the listing day, as the earliest date at which the price begins to converge on any fundamental anchor.

Governance signals to read for

Three governance elements from the filings will determine how minority shareholders should think about the price. Whether either filing includes a dual-class share structure with super-voting rights. Whether either includes founder compensation packages with milestone tranches similar in structure to the SpaceX award. Whether either includes information-rights limitations that reduce what minority holders can require the company to disclose after listing.

Any of those elements should attract a governance discount to the offered price. The magnitude of the discount is a matter of judgement, but the direction is not. Public academic work on dual-class listings puts the observed governance discount in the mid-to-high single digit percentage range for the first two years post-listing, widening in periods of stress. That is a real number, and it is the reader's to apply.

What we do not yet know, and will not pretend to

The confidential nature of the filings means several things a normal pre-IPO analysis would rely on are not yet available. Actual revenue breakdown by customer segment. Actual gross margin trajectory. Actual compute cost curves and their sensitivity to model architecture and hardware generation. Actual customer concentration. Actual capitalisation structure post-listing. Every one of these will be visible when the S-1s go public. Nothing said before that publication should be treated as a substitute for that reading.

The two things this piece does claim are that the valuations reported so far imply a specific set of assumptions about the outside-revenue trajectory, and that the BIS Annual Economic Report 2026 has changed the regulatory language against which those assumptions will be read. Both claims are supportable from primary sources today. Neither claim depends on any particular scenario for how either company evolves.

How to think about this if you are the buyer

Three practical steps. First, wait for the S-1s. The reported term sheets are not the filing. Second, when the S-1s go public, run the coverage-ratio arithmetic before reading the valuation section. The commitment schedule and the customer concentration table together determine whether the reported valuation is prudent or aspirational. Third, if you take a position at listing, size it as a retail allocation to a founder-controlled, narrative-driven listing with elevated volatility for at least one lock-up cycle. That is what the class of deal has looked like historically and there is no reason to expect this one to look different.

The circle and the filing are the same story. The commitment book is the input. The listing price is the output. When the S-1s make both public, the arithmetic that has been done privately can be done publicly. Until then, the responsible read is that the reporting establishes intent, not price.