Epoch AI report: Anthropic generates $9 million in revenue per employee, over 60% higher than OpenAI

According to data insights published by the AI research organization Epoch AI on May 8, researcher Luke Emberson estimates that Anthropic generates about $9 million in revenue per employee, while OpenAI generates about $5.5 million per employee. Both figures exceed the per-employee revenue levels of all large publicly listed tech companies on the Forbes Global 2000 ranking.

Data estimation methodology and key comparison metrics

(Source: Epoch AI)

Based on data insights from Epoch AI researcher Luke Emberson, the above estimates are not derived from companies’ publicly disclosed financial statements. Instead, they are calculated using publicly reported media coverage and the average year-over-year growth rate in employee counts. Epoch AI’s report notes that when companies reach an annual revenue scale of the “hundreds of billions” of dollars, they still manage to maintain high-speed growth in revenue per employee, which is rare in the history of tech industry development.

Main per-employee revenue comparison data:

Anthropic: about $9 million

OpenAI: about $5.5 million

NVIDIA: about $5.1 million (industry reference)

Epoch AI concurrent research report: Estimated AI chip smuggling and supply-chain analysis

In Epoch AI’s weekly roundup for the same period, researcher Isabelle Zuniewicz published an estimated report on AI chip smuggling, estimating that by 2025, approximately 290k to 1.6 million H100-class AI processors have been smuggled into China, with a median estimate of about 660k—roughly equivalent to one-third of China’s total AI compute capacity. The estimate draws on two types of evidence: data deviating from legitimate supply chains, and data from gray-market resale in China.

Epoch AI also launched an AI chip components data browser, tracking three key components in advanced AI chip supply chains since 2024: advanced-node logic, high-bandwidth memory (HBM), and chip packaging (CoWoS). In a related article, researcher Venkat Somala noted that high-bandwidth memory (HBM) has become the main cost driver and the main supply-chain bottleneck.

This is Epoch AI’s first weekly issue; previously it was published monthly.

Frequently asked questions

What methodology does Epoch AI use to estimate per-employee revenue for Anthropic and OpenAI?

According to data insights published on May 8, 2026 by Epoch AI researcher Luke Emberson, the estimate is based on publicly reported media coverage and the average year-over-year growth rate in employee counts, not on companies’ publicly disclosed financial statements. Anthropic is estimated at about $9 million, and OpenAI at about $5.5 million.

What level is NVIDIA’s per-employee revenue at in this report?

According to Epoch AI’s report, NVIDIA’s per-employee revenue is about $5.1 million, lower than Anthropic’s $9 million and OpenAI’s $5.5 million, but at a relatively high level among large publicly listed tech companies on the Forbes Global 2000 ranking.

What is Epoch AI’s estimated conclusion on AI chip smuggling into China?

According to the report by Epoch AI researcher Isabelle Zuniewicz, as of 2025, approximately 290k to 1.6 million H100-class AI processors have been smuggled into China, with a median estimate of about 660k, accounting for roughly one-third of China’s total AI compute capacity.