The headline changed in 2026: Anthropic passed OpenAI in revenue. A year that started with OpenAI comfortably ahead now has Anthropic at a higher run rate and a higher valuation. Here's the definitive money scoreboard as of August 2026 — from the frontier labs down to the absurdly efficient small teams — with the caveats that make the numbers honest.
August update: Anthropic's run rate reached ~$47B (May) and kept accelerating, OpenAI closed the largest private round in history ($122B at $852B post-money), and we've corrected the GitHub Copilot line — the widely-quoted "$2B ARR" didn't survive contact with analyst work. Details below.
The Top 10 by ARR (Annualized Run Rate), August 2026
| Rank | Company | ARR (run rate) | Valuation | Headline |
|---|---|---|---|---|
| 1 | Anthropic | ~$47B | $965B | Passed OpenAI in April; first operating profit projected for Q2, not yet confirmed* |
| 2 | OpenAI | Est. $24–33B | ~$852B | Closed a $122B round — the largest private raise in history |
| 3 | Databricks | $5.4B | $134B | The data-and-AI infrastructure workhorse |
| 4 | Cursor (Anysphere) | $2B+ | ~$50B | 300+ staff as of Nov 2025 — about $6.7M per head |
| 5 | xAI | Est. $1B+ | $50B+ | Grok's enterprise and consumer push |
| 6 | Scale AI | Est. $1B+ | ~$14B | The data-labeling backbone of the industry |
| 7 | GitHub Copilot (Microsoft) | Est. $0.9–1.1B | Part of MSFT | 4.7M paid seats; corrected from the "$2B" headlines |
| 8 | Mistral AI | Est. $600M+ | ~$10B+ | Europe's frontier-model champion |
| 9 | Midjourney | Est. $300–500M | Private (est. $5B+) | Reported profitable; no outside funding; headcount not reliably known |
| 10 | ElevenLabs | Est. $500M | $3.3B | Voice AI's runaway leader |
*Anthropic reports revenue on a gross basis — it counts total end-customer spend through cloud resellers (AWS, Google, Microsoft) as revenue and books partner payouts as expense. OpenAI reports closer to net, which is why its range is wide: OpenAI has confirmed ~$2B in monthly revenue (≈$24B annualized), while earlier gross-leaning reports cited ~$33B. That accounting difference means the top two aren't perfectly apples-to-apples. Both are real, enormous, and growing faster than any software company in history.
A correction we owe you on GitHub Copilot. Earlier versions of this leaderboard — like most of the press — carried "GitHub Copilot $2B+ ARR." Microsoft has never disclosed Copilot's ARR. What it disclosed (FY26 Q2 earnings) is 4.7 million paid subscribers, up ~75% year over year; independent analyst work built on that seat count and tier mix now puts ARR at roughly $0.9–1.1B. We've corrected the row. On this page, honest numbers beat big ones.
Note: Figures for private companies are estimates based on public reporting and disclosed funding rounds. Run rates move fast at this stage — treat them as the August 2026 snapshot, not gospel.
What the table adds up to
Numbers computed from the ten rows above, using the midpoint of any stated range:
- Combined run rate: roughly $87.5B across the ten companies.
- The top two hold about 86% of it. Anthropic and OpenAI together run ~$75.5B; the remaining eight share ~$12B.
- Median run rate: about $1B. The distribution is not a ladder, it is a cliff — half the board sits at or below a billion while the leader is at forty-seven.
- Six of ten are at or under $1B. A "top ten AI company by revenue" is, more often than not, a company doing hundreds of millions.
- Revenue per employee: we no longer publish a ranking for it. Only one company here has a headcount we can date and source — Cursor, 300+ as of November 2025, giving roughly $6.7M per head. For the rest, the figure circulating online is guesswork. The section below explains why.
Key Takeaways
Anthropic overtook OpenAI — and the growth rate is the real story
Anthropic went from roughly $9B in annualized revenue at the end of 2025 to a ~$47B run rate by May 2026 — the crossover analysts predicted for August happened back in April — and filed confidentially for an IPO at a $965 billion valuation. It projected its first-ever operating profit (~$559M) for Q2 — a projection the company has not since confirmed as a reported result, and Q2 has now closed. OpenAI is no slouch — it confirmed ~$2B in monthly revenue and closed a $122B round, the largest private raise in history — but for the first time, the challenger is ahead on both revenue and valuation. The business-model split explains a lot: ~85% of Anthropic's revenue is enterprise, ~85% of OpenAI's is consumer subscriptions. We covered the dynamics in Anthropic's leap past OpenAI.
Revenue per employee is the most abused number in AI coverage
This page used to carry the line everyone else carries: Cursor doing $2B with about fifty people, roughly $40 million per employee. It is wrong, and we had it wrong too. Anysphere reported a team of over 300 in November 2025. Against a $2B run rate that is about $6.7 million per employee — still extraordinary by any pre-2023 software benchmark, and still the highest figure we can actually evidence on this board, but a sixth of the number in circulation.
Midjourney's famous "$12.5M per head" has the same problem from the other direction. Public headcount estimates for it range from roughly 40 to 163 people, and revenue estimates run from Forbes's ~$300M for 2024 to the ~$500M widely quoted since. Divide an estimate by a guess and you get a headline, not a fact. We removed the calculation rather than pick the flattering pair.
The insight survives without the arithmetic: AI-native companies earn multiples per head that traditional software never reached. Most of the specific numbers you see quoted — including the ones we quoted — do not survive a check of when the headcount was published. The same physics drives the solo founders and tiny teams building real income with AI.
Foundation models still own the top of the board
The top two labs alone — Anthropic and OpenAI — run a combined ~$70–80B in annualized revenue. The model layer captures the lion's share, even as the application layer (Cursor, ElevenLabs, Copilot) scales underneath it. For how the model tiers are priced and who leads on capability, see the LLM model leaderboard.
Bootstrapped can still beat funded
Midjourney remains the efficiency legend: somewhere between $300M and $500M in revenue on $0 raised, and Forbes reports it as profitable. Its financials are private, so nobody outside can verify the margin, and its headcount is not reliably published — which is why we no longer print a per-employee figure for it. The point stands without the arithmetic: you don't need billions to build a profitable AI company, though you do if you want to train frontier models and land on the top half of this table.
Which AI companies are actually profitable?
Revenue rankings quietly dodge this question, so here is the straight answer: almost none of them disclose a profit, and most of the largest are still losing money on purpose. Revenue and profit have come apart further in AI than in any software market before it, because training runs and inference are paid for up front while revenue arrives monthly.
What is actually known, as of August 2026:
| Company | Profitability status | What that is based on |
| Anthropic | First operating profit projected for Q2 2026 (~$559M) | Company projection. Q2 has closed with no public confirmation of the actual result |
| Midjourney | Reported profitable; private financials unavailable | Forbes reports profitability. No audited figure is public, and headcount estimates range from ~40 to ~163 |
| OpenAI | Not disclosed; widely reported to be lossmaking while scaling | No published figure; inference from reporting and from the size of its raises |
| Databricks | Not disclosed | — |
| Cursor (Anysphere) | Not disclosed | — |
| xAI | Not disclosed | — |
| Scale AI | Not disclosed | — |
| GitHub Copilot | Not separable — reported inside Microsoft | Microsoft does not break out Copilot economics |
| Mistral AI | Not disclosed | — |
| ElevenLabs | Not disclosed | — |
Two honest conclusions follow. First, the most profitable AI company is probably not on this list at all — the reliable money in this cycle has been made selling infrastructure, not models. Second, when you see a headline calling a lab "profitable", check whether it means operating profit, gross margin, or a projection. In 2026 it is usually the third.
If a company on this list publishes an audited profit figure, we will add it here with the source and the date.
What this means
For entrepreneurs: The market is unambiguously real. When the #4 company does $2B with 50 people, the ceiling on what a small team can build has moved. Opportunity exists at every level — the cheapest on-ramp is still selling a service, as I argue in why AI belongs in your services, not just your back office, and if you run a small business, start with the AI implementation playbook for small business — no CTO required.
For job seekers: Companies doing tens of billions in AI revenue are hiring hard and paying up. The premium goes to people who can wield AI, not just describe it — see the highest-paying AI jobs of 2026.
For investors: Two companies are compounding faster than any software businesses on record. Even at $852B and $965B valuations, the run-rate growth is what's driving the multiples — and the AI IPO race is about to test whether public markets agree.
Frequently Asked Questions
Which AI company makes the most money in 2026?
As of August 2026, Anthropic leads on revenue with a run rate around $47 billion, ahead of OpenAI's estimated $24–33 billion — though Anthropic reports on a gross basis, which inflates the comparison. By valuation, Anthropic ($965B) also sits just above OpenAI (~$852B).
Did Anthropic really pass OpenAI?
Yes — the crossover happened in April 2026, months earlier than analysts predicted. Anthropic grew from about $9B in annualized revenue at the end of 2025 to roughly $47B by May 2026, and projects its first operating profit for Q2. The caveat is accounting: Anthropic counts cloud-reseller end-customer spend as revenue, while OpenAI reports closer to net.
Which AI company has the highest revenue per employee?
On current evidence, Cursor (Anysphere): roughly $6.7M per person, from a $2B run rate against the 300-plus staff the company reported in November 2025. Treat every other per-employee figure in AI coverage with suspicion, including the "$40M per head" version of this one that we ourselves published until 7 August 2026 — it divided current revenue by a headcount that was already a year out of date.
Cutting through the noise
The scoreboard is clear: the revenue leader now does a ~$47B run rate, a 50-person company does $2B, and a bootstrapped studio of 40 does half a billion on no outside money. Mind the gross-vs-net caveat, mind that private figures are estimates — but the direction is not in doubt. These are real revenues from real customers. Not a bubble. A market, reordering itself in real time.
Methodology, sources and limits
What gets included. Companies whose revenue comes primarily from AI products they build and sell. That excludes the biggest AI earners of this cycle — Nvidia, the cloud providers — because their AI revenue is not separable from hardware and infrastructure lines. It is a leaderboard of AI-native businesses, not of everyone profiting from AI.
Where the numbers come from. Company statements and earnings disclosures where they exist; reporting by established outlets and independent analyst work where they do not. Every figure marked "Est." is an estimate, not a disclosure. Where a company reports a range or where credible sources disagree, we print the range rather than pick the flattering end.
Run rate, not annual revenue. ARR here means annualised run rate — the current monthly figure multiplied by twelve. For companies growing at these speeds, run rate and trailing revenue are very different numbers, and mixing them is the most common error in AI revenue coverage.
The gross-versus-net problem. Anthropic books total end-customer spend through cloud resellers as revenue; OpenAI reports closer to net. The two are therefore not perfectly comparable, and we flag it every time rather than burying it in a footnote.
Headcount and per-employee figures. We publish a revenue-per-employee number only where the company itself has stated a headcount and we can date that statement. Everywhere else the field is empty. This is the single most error-prone statistic in AI coverage: staff counts go stale within months while the revenue figure keeps being updated, so the ratio drifts upward on its own. We got this wrong on this page and corrected it — see the change log.
Confidence levels. The dataset marks each row high, medium or low. High means the company published the figure. Medium means credible independent reporting or analyst work. Low means an estimate we could not trace to a primary source; those rows are printed because omitting them would distort the ranking, not because we vouch for the precision.
Verification. All ten rows were re-checked on 7 August 2026. Figures move fast at this stage; treat the table as a dated snapshot.
What we will change. If a figure is wrong and you can show us, we correct the row and say so in the change log below — as we did with GitHub Copilot. Corrections are published, not quietly edited.
Change log
- 7 August 2026 (second correction) — Cursor's revenue per employee was wrong. We carried "~50 employees, ~$40M per head". Anysphere reported over 300 staff in November 2025, which puts the figure near $6.7M. We also removed Midjourney's per-employee calculation entirely: its published headcount estimates range from about 40 to 163, so any ratio built on them is invented precision. Midjourney's profitability is now stated as reported by Forbes rather than inferred from its lack of funding.
- 7 August 2026 — All ten rows re-verified. Added a profitability section, computed aggregates, this methodology, and a downloadable dataset.
- 4 August 2026 — Corrected GitHub Copilot from the widely-quoted "$2B+ ARR" to an estimated $0.9–1.1B, based on the disclosed 4.7M paid seats and independent analyst work. Added the correction notice in the body.
- August 2026 — Anthropic updated to a ~$47B run rate and $965B valuation; OpenAI's $122B round added.
Use this data
The table is free to reuse with attribution to AI Business (aibusiness.vc) and a link to this page. A machine-readable copy is available at ai-revenue-leaderboard.csv — same figures, same date stamp, ready for a spreadsheet or a chart.
If you are citing it in research or a newsletter and want to check a figure before you publish, write to info@aibusiness.vc and we will point you at the underlying source for that row.
