The OpenAI IPO: When Research Becomes a Stock

OpenAI, Anthropic, and SpaceX have all confidentially filed to go public within weeks of each other, putting more than $3 trillion of combined valuation on the market at the same time.

## Mechanics of the filing This is not the public S-1 yet — that drops after the SEC review and audited financials. OpenAI is working with Goldman Sachs and Morgan Stanley, with a listing possible as soon as September. SpaceX could list in the next 2–3 days. Three trillion-dollar-class companies hitting the tape in one window is, in the creator's framing, "Wall Street history."

## The structural conflict OpenAI started as a nonprofit, then adopted a capped profit structure, and is now preparing for standard public ownership. Public companies have a fiduciary duty to maximize shareholder value. Frontier research is "expensive, unpredictable, and occasionally requires you to stop." When a genuine safety inflection point arrives — slow down, publish less — the quarterly earnings call will not be sympathetic. The budget line will ask why safety research is not growing revenue.

## What the S-1 will force into the open The filing will publicly disclose the cost of training frontier models, the revenue split between consumer subscriptions and enterprise APIs, and whether the business is a software-margin product or a "compute-intensive commodity play in disguise." The creator's read: in a sector "disguised by secrecy, going public is like shooting themselves in the foot — that benefits everyone except the company itself."

## Where this points the company Public markets reward predictable recurring revenue. Frontier research rewards "unpredictable leaps." The incentive shifts OpenAI away from being a lab and toward being an infrastructural utility — API endpoints, team licenses, cloud partnerships. "You don't get a trillion-dollar valuation by gambling on paradigm-shifting breakthroughs. You get it by selling shovels." Expect OpenAI to start looking like the company that hosts the next frontier model, not the one that trains it.

## The cohort problem With Anthropic and SpaceX in the same window, OpenAI is no longer evaluated in isolation. Shared volatility and comparative multiples bind them together. When one stumbles, the others absorb the narrative damage. They are "each other's context now."