Until 26 October, every figure about frontier-lab economics was an estimate. Anthropic's S-1 changed that for
exactly one company. OpenAI remains private, with no filing in public view and a 2027 window.
| What | Anthropic (S-1, 26 Oct 2026) | OpenAI |
|---|
| Filing status | Form S-1 filed with the SEC | None filed |
| Ticker / exchange | ANT, Nasdaq Global Select | Not applicable |
| H1 2026 revenue | $16.3 billion | Not disclosed |
| 2025 revenue | $4.6 billion | Not disclosed |
| Q2 2026 operating income | Positive $974 million | Not disclosed |
| H1 2026 net loss | $80.7 billion, incl. $72.4bn non-cash | Not disclosed |
| Compute commitments | Up to $120.5 billion to 2029 | Not disclosed |
| Headcount | ~5,300 as of 30 September | Not disclosed |
| Revenue per employee | Over $6.6 million | Not disclosed |
| Expected listing | Process started | 2027 window, no S-1 |
What the disclosure actually changes
The single most useful line is Q2 operating income of $974 million, against a $1.7 billion
operating loss in Q1. That is the first evidence from any frontier lab that the unit economics can turn positive
at scale, and it is checkable rather than briefed.
The second is the compute obligation: up to $120.5 billion under arrangements running into 2029,
roughly $3.1 billion a month once capacity is fully delivered, expected December 2026. That is a cash commitment,
unlike the convertible-note charge, and it is the number that decides whether the operating profit holds.
Why the comparison is uneven on purpose
Every OpenAI figure in circulation - revenue run-rates, valuations, loss estimates - comes from reporting on
private conversations. Those may be accurate. They are not the same category of information as a document filed
with a regulator under liability. When OpenAI files, this table can be completed. Until then the right reading is
that one company's economics can be audited and the other's cannot.
Sources
Anthropic newsroom. Yahoo Finance live coverage. Yahoo Finance on the accounting.