← Back to the map

Topic board1Map station ①|DemandDone

AI Revenue: Where the ARR Comes From

This board answers three things: how much AI is earning, how fast it is growing, and which product drove each step up.

Read from the map: Already happening, and each quarter adds more than the one before

Anthropic run rate

$81B

Nov 2026|End-October run rate from the company's S-1/A (Nov 10, 2026); about $71B at end-September

OpenAI run rate

$50B

Oct 2026

Anthropic monthly growth

13.8%/mo

Nov 2026

Anthropic Q3 2026 revenue

$16.96B

3Q26|S-1/A quarterly table, up 47% quarter on quarter; Q2 was $11.6B. Compute spend $9.86B, GAAP operating loss $0.57B, adjusted operating profit $2.46B

Ramp enterprise adoption

43.5% / 39.7%

Jul 2026|Anthropic / OpenAI, share of paying companies

Time range: All charts rescale together

1ARR path and product launches

The log axis shows growth rates: the same slope means the same multiple. Vertical dashed lines are product launches; shaded bands are the main growth bursts, orange for Anthropic, blue for OpenAI. Figures are annualized revenue as disclosed by the companies or in the press. From 2026 on, the ARR the two companies add in a year is several times the previous three years combined.

y axis:

ARR Path ($B, log) vs Public Launch Events#

On a log scale both companies' ARR climbs in steps, and each step lines up with a product launch.

How to read this chart

Figures are annualized revenue as disclosed by the companies or in the press, on a log axis. Shaded bands mark the main growth phases; vertical dashed lines mark product launches. Anthropic's roughly $71B at end-September and $81B at end-October 2026 are the company's own run rate from its S-1/A (Nov 10, 2026): the last 28 days of usage fees times 13 plus the month's subscriptions times 12, including sales made through cloud partners. OpenAI's $50B at end-September excludes those, so the two are still on different bases and the OpenAI piece runs low.

Source: FinSight compilation and estimates · Updated 2026-11-16

2New ARR per month: what data centers have to absorb

Each step is the average monthly addition between two disclosures. In three and a half years neither company has turned negative; the worst case was only slower growth. The dashed segment is OpenAI's July stretch with no disclosed figure, backed out from company comments, directional only.

New ARR per Month ($B) = What Data Centers Have to Absorb#

How to read this chart

Each step is the average monthly increase between two consecutive disclosed figures. The dashed line is a breakdown of OpenAI's July, inferred from management's 20% month-on-month comment, for direction only.

Source: FinSight compilation and estimates · Updated 2026-11-16

3Monthly growth rate: the step detector

Left, Anthropic: the Claude Code burst ran 20% to 30% a month, the Cowork burst 37% to 63%. Right, OpenAI: after the consumer burst it ground along at 8% to 12% a month for two and a half years; the roughly $50B run rate the company gave investors at end-September lifts the latest stretch to about 20% a month, and whether that is a new step waits on Ramp's September figures. Anthropic's S-1/A (Nov 10, 2026) adds a new stretch: only about 4.5% a month in August and September, then 14% in October alone.

ARR Growth per Month, Anthropic (%/Month) = Step-Up Detector#

Anthropic grew only about 4.5% a month in August and September 2026, then jumped 14% in October, adding $10B in a single month, another step up (company run rate, S-1/A of Nov 10, 2026).

How to read this chart

Monthly growth is the step-function detector: the Claude Code step in 2Q25 ran 20% to 30% a month, the Cowork step in 1Q26 ran 37% to 63%.

Source: FinSight compilation and estimates · Updated 2026-11-16

ARR Growth per Month, OpenAI (%/Month)#

How to read this chart

The blue dashed stretch is the February 28 to September 30 breakdown: the roughly $30B on July 15 is not a company disclosure but a back-calculated starting point from what the company said on October 8 (nearing $50B at end-September, up more than 70% in the quarter), shown for direction only. The solid line over the same period is the interval average. The earlier $70B (September) and $40B (August) figures were investors grossing up for cloud-partner sales and have been withdrawn. The five-company total is still on a mixed basis: Anthropic now uses the company's own run rate from its S-1/A (Nov 10, 2026), about $71B at end-September including sales made through cloud partners, while OpenAI's $50B at end-September excludes them, so the OpenAI piece runs low.

Source: FinSight compilation and estimates · Updated 2026-11-16

4Run rate vs. booked revenue: do they add up?

Full-year booked revenue comes to roughly half of the year-end run rate, and both companies' disclosures hold up against it. Anthropic's pace slowed in August and September; OpenAI's second-quarter figure confirms that the step up only arrived in July.

Anthropic: Company Run Rate vs. Quarterly Revenue × 4#

The company's own run rate (S-1/A, Nov 10, 2026) was about $71B at end-September and $81B at end-October: only about 4.5% a month in August and September, then a 14% jump in October. Our pre-filing figure of about $76B for end-September, solved from Q3 revenue, ran 7% high because the company defines run rate as the trailing 28 days, annualized.

How to read this chart

The orange line is the annualized revenue the company disclosed, interpolated log-linearly between points. The roughly $71B at end-September and $81B at end-October are the run rate from the S-1/A of Nov 10, 2026 (the last 28 days of usage fees times 13 plus the month's subscriptions times 12; end-September is backed out from 'about $10B higher at end-October'). Blue blocks are each quarter's actual revenue times four, all audited filing figures from 1Q25 on. The red dashed stretch is our pre-filing estimate for August and September solved from Q3 revenue, kept for comparison only. The hollow red point is the pace reported in the press in mid-September, not a company disclosure, and is left out of the calculation.

Source: Anthropic S-1/A (Nov 10, 2026), investor updates and public reporting; FinSight compilation and estimates · Updated 2026-11-11

OpenAI: Straight Interpolation vs. the Step Scenario#

How to read this chart

Both quarters of actual revenue come from public reporting. Full-year 2025 ($13.1B) and first-half 2025 ($4.3B) were checked the same way; both came in more than 10% above the interpolated estimate, so disclosed run rates are a floor.

Source: Company disclosures and investor updates, public reporting; FinSight compilation and estimates · Updated 2026-10-15

5Breadth check: how many companies are paying

Ramp measures the share of companies paying. The ChatGPT consumer burst (2.1% to 17.6% between January and March 2023) is still the largest on record. In the six months after Cowork, Anthropic added 24 points and passed OpenAI in May 2026; OpenAI added 0.23 points in July, ending its decline.

Ramp Enterprise Penetration (% of Companies) = the Breadth Layer#

How to read this chart

ChatGPT's consumer burst (2.1% to 17.6% between January and March 2023) remains the largest wave on record. OpenAI stopped falling in July 2025 with a 0.23-point uptick.

Source: Ramp AI Index (public data); FinSight compilation · Updated 2026-11-16

6Depth check: how much each company spends

Monthly AI spend per employee in three tiers: the top 1%, the top 10% and the typical company. Dashed lines are estimated from Ramp's published chart; dots are figures disclosed in Ramp's articles. The typical company pays about one seat; the top 1% are building their own AI workflows. In August the top 1% fell 9.7% month on month for the first time, which Ramp attributes to seasonality and lower unit prices; watch whether September recovers.

Top 1% (heavy users)#

Source: Ramp AI Index (public data); FinSight compilation · Updated 2026-11-16

Top 10%#

Source: Ramp AI Index (public data); FinSight compilation · Updated 2026-11-16

Typical company (median)#

Source: Ramp AI Index (public data); FinSight compilation · Updated 2026-11-16

7Extension: new ARR per new chip

Divide new ARR on the demand side by new compute on the supply side and you get the marginal yardstick. Marginal above average means the squeeze continues; marginal falling below average is the early warning of a top. The full supply and demand projection lives on the compute board.

New ARR per New Chip (Monthly Approximation, $/Chip/Year)#

How to read this chart

The orange line is a monthly approximation: OpenAI plus Anthropic monthly new ARR divided by market-wide monthly new compute. Dark dots are the exact quarterly values (five companies in the numerator). The grey dashed line is the average yardstick. 4Q24 was the trough of confidence.

Source: Epoch AI (CC BY 4.0); FinSight compilation and estimates · Updated 2026-11-16

8Event scorecard: which launches produced a burst

✔ produced a burst, ✘ did not, ⏳ still being checked. Orange is Anthropic, blue is OpenAI, grey is industry infrastructure. GA means general availability.

The event notes are kept in Chinese for now; an English version follows once the wording is settled.