OpenAI and Anthropic’s combined reported annualized revenue run rates reached $105B by late summer, up from about $30B at the start of the year, according to the State of AI Report 2026, the ninth annual edition by investor Nathan Benaich of Air Street Capital, published on 8 October 2026.
The report’s launch essay says the figures cover both companies’ whole businesses across subscriptions, coding products and APIs, and describes the frontier as “a three-lab race between Anthropic, OpenAI, and Google.”
Business Pill · THE COST PAID ON EVERY ANSWER
A one-minute explainer of inference: the cost of running a trained model every time it answers. It teaches the general idea only and says nothing about any company in this story.
The key insight: As we read it, the report’s numbers pull in two directions at once. The report says a fixed benchmark score has become roughly 13-fold cheaper each year since 2023, while OpenAI and Anthropic’s combined reported run rates rose from about $30B at the start of the year to $105B by late summer, so cheaper answers are coming with much larger total spending.
The essay argues that agents repeatedly calling trained models create recurring demand across APIs, subscriptions and infrastructure, and puts it this way: the industry is in a state of “either you die trying to get to the frontier, or you live long enough to serve inference.”
Citing Epoch AI, it says achieving a fixed score across five benchmarks has become 47% cheaper each quarter since 2023, roughly 13-fold cheaper each year.
It also cites Standard Metrics’ preliminary second-quarter data: AI-native companies with $1M to $20M in annualized revenue grew 256% year over year at the 75th percentile, against 90% for companies adding AI to existing software. The essay notes these comparisons do not establish that adopting AI causes faster growth.
OpenAI and Anthropic’s combined reported annualized revenue run rates, $B, as given in the State of AI Report 2026 launch essay of 8 October 2026: about 30 at the start of 2026 and 105 by late summer.
The Build and Who Pays for It
According to the report, four US hyperscalers guide to roughly $733B in total 2026 capex, and CoreWeave reported 1.5GW of active power at the end of the second quarter, with contracted power reaching 4.2GW by August 11.
By contrast, it says the EU’s initial AI gigafactory contribution is €1B, within a plan seeking up to €10B in public funding and at least €20B privately, and notes these are different measures.
Selected sovereign AI programs in the report pledge about $138B, and it notes NVIDIA reported more than $30B in sovereign AI revenue in fiscal 2026.
Share of Americans opposing a local facility in Gallup’s March survey, as cited by the State of AI Report 2026: 71% for an AI data center, 53% for a nuclear plant.
Local Consent
The report cites a Gallup survey from March in which 71% of Americans opposed a local AI data center, against 53% for a nuclear plant, and Data Center Watch’s count of at least 45 projects, representing nearly $68B in planned investment, blocked or delayed by local opposition in the second quarter.
“Local consent is already determining which projects can proceed,” the essay says.
Safety, and What It Predicts
On safety, the report leads with the OpenAI and Hugging Face incident, saying OpenAI’s technical report records code execution on 41 production workers, and that METR and Redwood found about 700 agents joined the attack.
Among its nine predictions for the next 12 months, the essay lists an agent halving its failure rate on new tasks after a month of customer work without a model upgrade, and US AI labs officially launching frontier cyberdefense products.
The Structural Read
Revenue is moving to whoever serves the calls. The essay describes agents calling models repeatedly, creating recurring demand across APIs, subscriptions and infrastructure.
The build is running into people as well as power. The report pairs roughly $733B of guided hyperscaler capex with Gallup’s 71% local opposition to AI data centers.
Sovereignty is being bought from the same supplier. The report says selected sovereign programs pledge about $138B, and that NVIDIA reported more than $30B of sovereign AI revenue in fiscal 2026.
Nathan Benaich, State of AI Report 2026
“Local consent is already determining which projects can proceed.”
Three Implications
TWO LABS AT $105B The report puts OpenAI and Anthropic’s combined run rates at $105B by late summer.
CHEAPER PER ANSWER Citing Epoch AI, it says a fixed benchmark score got roughly 13-fold cheaper each year.
CONSENT AS A CONSTRAINT Data Center Watch counted at least 45 projects, nearly $68B, blocked or delayed in Q2, per the report.
The framework’s starting point: “Training builds the brain. Inference is the brain working.”
As we read it, the report’s line that the industry must either reach the frontier or live long enough to serve inference is the same split: its revenue figures are inference revenue, and its capex figures are what that inference requires.
What Is Not Established
We read the report’s launch essay on stateof.ai; we did not read the full slide deck. Its figures are compilations of other organisations’ data, which we did not check against each original source, and its predictions are the author’s. We did not contact Air Street Capital.
Business Engineer Framework
The State of the Inference Economy
A Business Engineer framework on how AI moved from a training era to an inference era, and where the money goes.
The State of AI Report 2026 puts OpenAI and Anthropic’s combined run rates at $105B by late summer, up from about $30B at the start of the year. It also cites a roughly 13-fold yearly fall in the cost of a fixed benchmark score across five benchmarks.
On infrastructure, it reports that four US hyperscalers guide to roughly $733B in 2026 capex, and that at least 45 projects, nearly $68B in planned investment, were blocked or delayed by local opposition in the second quarter.
A note on sourcing. We read the State of AI Report 2026 launch essay on stateof.ai, published on 8 October 2026; its figures are the report’s compilations of the sources it cites. We did not read the full slide deck. We did not contact Air Street Capital. Nothing here is a forecast, and nothing here is financial or investment advice.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
