Asked on a stage whether OpenAI would be ready to hold an initial public offering next year, Greg Brockman gave an answer of two words. The company’s president said he thought it could be. Then he offered the company’s own estimate, a year further out: 2027, when he expects OpenAI to be in a position to list. He was speaking on Nov. 5 at a technology conference the Wall Street Journal held in Napa, California.
The more consequential sentence in his remarks was about where the money now comes from. Most of OpenAI’s revenue arrives from business customers rather than from individuals paying for a subscription, he said. That is a different company from the one the public came to know, and a different set of numbers for anyone trying to put a value on it.
The mix matters because the two businesses are valued on different things. A consumer subscription is judged on growth and retention, and its costs are mostly infrastructure and app-store fees. An enterprise contract is judged on duration, renewal and how much more a customer spends in year two than in year one, and its costs sit in a sales organization that has to be built market by market. Investors have historically paid a higher multiple for the first and assigned a lower, steadier one to the second.
Brockman credited the shift to Codex, the company’s coding agent, and said OpenAI is competing with Anthropic for corporate accounts. The two companies are chasing the same buyers, which in practice means the same engineering organizations, the same procurement committees and the same security reviews. A coding agent is an unusually direct way into that budget, because the people who evaluate it can measure whether it works within a week.
Anthropic, according to the same account of the market, is expected to be in a position to list before the end of this year. If that happens, the ordering will matter more than either company would like to admit. The first comparable to trade gives the second a benchmark, and benchmarks cut both ways: a strong debut lifts the follower, a weak one sets a ceiling that the follower has to argue against with its own numbers.
There is a mechanical side to the 2027 date that gets lost in the headlines. Being ready to list is an accounting and governance condition before it is a market decision. It means audited statements, a board with independent directors, disclosure of related-party arrangements and a controls environment that survives a review by bankers and lawyers who are paid to find problems. A company can be ready and still choose not to file, and several large software companies have done exactly that for years.
OpenAI has delayed listing plans before, and it is now accelerating its fundraising to take advantage of the growth of the past year. Those two facts sit together awkwardly. Raising private capital at a higher valuation is easier when a public listing is close, because the private buyer can see an exit. But each private round also sets a price that the public market will eventually be asked to accept, and a public investor who is offered a discount to the last private round will ask why the last round was priced where it was.
The date also does work inside the company. A public listing is a retention instrument: equity that can be sold is worth more to an employee than equity that cannot, and a firm timeline is a way to hold senior engineers who have offers from competitors with liquid stock. That is a legitimate reason to name a year, and it is not the same as a commitment to file in that year.
For the enterprise customers Brockman described, the listing question is mostly indirect. A vendor that is about to be public has to show durable contracts and clean revenue recognition, which tends to make its sales terms stricter and its discounts smaller. It also has to disclose concentration, and a business that depends on a small number of very large customers will have to explain that dependence in a document that anyone can read.
The valuation question is the one that will be argued longest. Software companies with an enterprise base have traded at premiums to consumer businesses for most of the past decade, but the premium depends on the base being sticky and the sales cost being controlled. A company that grew on a consumer brand and is now selling to corporations has to prove the second half of that transition, and the proof arrives in renewal data rather than in an announcement.
None of this is settled by a sentence spoken at a conference. What the sentence does is fix a date that competitors, customers and employees can now plan around, and that journalists will hold the company to. The open question is not whether OpenAI can be ready in 2027. It is whether the market that greets it will price an enterprise revenue base attached to a consumer name the way it priced the last generation of software listings, or whether it will ask for a discount until the renewals have been observed for several years.
