OpenAI's S-1 and its manifesto are the same document

OpenAI confidentially filed a draft S-1 (the registration statement a company submits before an IPO) with the SEC on Sunday, with Goldman Sachs and Morgan Stanley as lead underwriters. On the same day, Sam Altman and chief scientist Jakub Pachocki published a blog post titled "Built to benefit everyone: our plan," describing OpenAI's "third phase" as "giving everyone on Earth access to a personal AGI." ^{1 2}

I want to take the co-timing seriously. The S-1 is confidential, which means the prospectus is not yet public, which means the only document the market gets to read this week is the blog post. That is not an accident. The manifesto is doing prospectus work before the prospectus is available.

What was actually filed. Not much, from the public's point of view. A confidential S-1 under the JOBS Act discloses nothing externally. We do not yet know revenue, operating loss, cohort retention, compute commitments, or the share structure post-IPO. We know the company has hired the two underwriters you hire when you want the largest possible book, and we know it has filed one week after Anthropic did the same. ³

The headline number floating around the filing is Bloomberg's $3.6 trillion aggregate AI IPO pipeline. ³ That number is a sum of private marks, not a clearing price. OpenAI's last disclosed private valuation was roughly $157bn in October 2024. The gap between private mark and public-market clearing price is, at this stage, unknowable, and the IPO process exists precisely to discover it.

What the blog post is doing. The "Built to benefit everyone" document is structured as a three-phase narrative: build the technology, demonstrate safety, share prosperity. The Phase 3 language is the load-bearing piece: "accelerating science, sharing prosperity, and giving everyone on Earth access to a personal AGI." ²

Read this as a Public Benefit Corporation (a for-profit company with a chartered mission obligation) preparing to answer the most awkward question on the IPO roadshow, which is: how does a company that began as a nonprofit, converted to a capped-profit LLC, then converted again to a PBC with the OpenAI Foundation holding controlling equity, justify selling shares to the public without the Foundation having to explain what its fiduciary duty to a public float will be?

The blog post is the pre-emptive answer. The mission is framed as compatible with public ownership because the proceeds, in the document's phrasing, flow back via "broad distribution of benefits." That sentence has to do a lot of work in the prospectus. It is easier to make it land in a blog post first.

The Anthropic comparator. Anthropic filed its own confidential S-1 one week earlier. Both companies are going public as PBCs. The governance backstops are different.

OpenAI's backstop is the Foundation: a 501(c)(3) (a US tax-exempt nonprofit) that holds equity and a board seat. Anthropic's is the Long-Term Benefit Trust (a trust with defined beneficiaries and fiduciary duties that elects a majority of the board over time). A Foundation equity stake can be restructured, sold down, or out-voted; a trust with fiduciary obligations is harder to unwind. Whether institutional investors will price that difference is the most interesting structural question of the next eight weeks.

My read is that they probably will, modestly, and in Anthropic's favour. Governance-discount pricing is rarely large in hot IPO windows, but it exists, and the LTBT is the more legible constraint of the two. It is also the structure that more closely resembles what a public-markets governance lawyer would have designed from scratch.

Phase 3 as a revenue model. The interesting thing about "a personal AGI for everyone on Earth" is that, as a sentence, it is also a total addressable market claim. And no company reaches everyone on Earth on a $20-per-month subscription. The arithmetic does not work; the world does not contain enough $20 notes.

A mass-market AGI product has two plausible revenue models. One is consumer subscription tiered by usage, of which the high tier (current ChatGPT Pro at $200/month) carries the unit economics and the free tier carries the user count. The other is advertising — agent-mediated commerce, sponsored answers, or some new monetisation surface that has not been named yet. OpenAI has spent the last year hiring advertising executives and partnering with commerce platforms. The mass-market framing in the blog post and the hiring pattern point at the same revenue model.

The S-1, when it goes public, will have to disclose enough about consumer ARPU (average revenue per user) and free-to-paid conversion for the underwriters to defend the valuation. If the model is largely subscription, the valuation will be defended on conversion uplift; if the model is advertising-supported, the prospectus will need to describe an advertising surface that does not yet exist as a product. Either way the document will say more than the blog post does, which is the point of the blog post coming first.

The desperate-for-money read. The case made by Ed Zitron and others is that OpenAI is filing because private capital terms have become onerous, not because public markets are the natural next step for the mission. I think this read is partly right and partly under-priced. OpenAI's reported burn rate is large, its compute commitments to Microsoft and others are larger, and the private rounds of the last 18 months have included structured features (tender offers, secondary components, preferred stacks) that are typical of companies running out of clean equity to sell.

A public float resets the capital stack and gives the company a currency for acquisitions and retention. It is rational financing, not mission betrayal. But it is also not the "third phase" of a mission. It is the first phase of being a public company, which is a different kind of thing.

What to watch.

• The public S-1, when it comes. Specifically: disclosed ARR, segment split between consumer and API, Microsoft revenue concentration, and any disclosure about advertising product plans.
• How the Foundation's equity stake is described in the governance section. Voting structure, transfer restrictions, lock-ups, and any mechanism by which Foundation holdings can be diluted post-IPO.
• Whether Anthropic's prospectus arrives first, and whether its LTBT structure clears at a tighter governance discount than OpenAI's Foundation model.
• Hiring at the advertising and commerce surfaces. The product model that the prospectus will eventually have to describe is being built in plain view.

The blog post is the easy document to read. The S-1 will be the hard one. I would not assume the two say the same thing.

Footnotes

¶