WHAT WAS SAID
● The commitment: OpenAI will be a public company in 2027, or sooner if the business continues to inflect.
● Who said it: CFO Sarah Friar, at an all-hands.
● What it resolves: a reported split — Friar pressing for 2027 at around a trillion, Altman pushing for 2026.
● Still missing: the public S-1. Confidential filing went in 8 June and nothing has appeared on EDGAR.
The timeline, as it actually stands
| Date |
Event |
| 8 June 2026 |
Confidential S-1 submitted to the SEC |
| 19 August 2026 |
CFO commits to 2027 at an all-hands |
| Not yet |
Public prospectus on EDGAR |
| Required |
Registration public at least 15 days before a roadshow |
| 1 October 2026 |
OpenAI vs Apple hearing |
A confidential filing in June and a 2027 listing commitment in August is a long runway. That gap is where the interesting information is.
Why the cautious side won
The reported disagreement was not about whether to list but when, and at what number. Friar was pressing for 2027 at around a trillion dollars. Altman was pushing for a 2026 debut.
A 2027 date buys three things a 2026 debut would not. It puts more audited quarters between the filing and the roadshow. It moves past the 1 October Apple hearing rather than pricing into litigation uncertainty. And it gives the loss profile more time to improve before public scrutiny — reported figures put OpenAI at roughly 2 billion dollars a month in revenue while losing about 1.22 dollars for every dollar earned.
THE ESCAPE HATCH IN THE WORDING
"In 2027, or sooner if our business continues to inflect" is not a date. It is a floor with an option attached. If growth accelerates, the 2026 debut is still live — the commitment removes the risk of listing late, not the possibility of listing early.
The number nobody else has published
Alongside the IPO news, OpenAI put a figure on something the industry usually discusses in adjectives. In its writing on pacing model development in an era of cyber-critical capabilities, the company estimates monitoring overhead at roughly 20 percent of the inference compute being monitored.
That covers all reinforcement learning training and every evaluation involving tools for GPT-5.6 Sol-class models and above, plus all inference on Astra.
Twenty percent is a real tax, and disclosing it ahead of a listing is a deliberate choice. For anyone building on the API it also explains something about pricing: safety compute is not free, it scales with capability, and it lands in the same cost base as the tokens you buy.
What actually matters if you build on OpenAI
| If you are... |
What changes |
| Running production workloads today |
Nothing immediately. No pricing or model change was announced |
| Planning 2027 budgets |
A public company faces margin pressure a private one does not. Price your downside |
| Single-vendor on OpenAI |
Route through an aggregator now so switching stays a config change |
| Watching for the real signal |
The public S-1 on EDGAR, not an all-hands quote. That is when the audited numbers appear |
Nothing here is investment advice, and none of the loss figures are audited yet. That is precisely the point of waiting for the prospectus.
FAQ
When will OpenAI actually go public?
CFO Sarah Friar told employees 2027, or sooner if the business continues to inflect. No filing date, roadshow date or exchange has been announced.
Has OpenAI filed an S-1?
Confidentially, on 8 June 2026. The public prospectus has not appeared on EDGAR. It must be public at least 15 days before any roadshow begins.
What valuation is being discussed?
Reporting puts Friar's preferred 2027 listing at around a trillion dollars. The March private round valued the company at 852 billion.
Is OpenAI profitable?
No. Reported figures put revenue near 2 billion dollars a month with losses of roughly 1.22 dollars for every dollar earned. These are unaudited and come from reporting rather than filings.
What is the 20 percent monitoring figure?
OpenAI's own estimate of safety monitoring overhead as a share of the inference compute being monitored, covering RL training, tool-involving evaluations for GPT-5.6 Sol-class models and above, and all Astra inference.
Should any of this change my model choice?
Not on its own. A listing commitment is a governance signal, not a product change. If it prompts anything, let it be routing through an aggregator so vendor risk stays a configuration decision rather than a rewrite.