OpenAI and Anthropic Pay AI Engineers Hundreds of Thousands as Talent Race Intensifies

The sums paid in the contest for frontier AI talent have become difficult to ignore. An Anthropic engineering vacancy in London advertises a base salary of £325,000 to £390,000. OpenAI’s UK accounts, meanwhile, show an average of about £2.3 million in salary and share awards per employee for 2025. Those figures are dramatic, but they describe different things. One is an advertised salary range for a specific role; the other is an average across a workforce that includes equity and people on very different packages.

The Guardian’s analysis of OpenAI UK Ltd’s accounts says 110 staff shared £87 million in salaries and £168 million in shares during 2025. Dividing those two amounts by the reported headcount produces roughly £2.3 million per employee. It does not follow that a typical engineer took home £2.3 million in cash. Share-based awards account for most of the total, and a few exceptionally large packages can pull an average far above the median. The accounts do not disclose a median engineer’s pay.

Anthropic offers a clearer example of a role-specific cash figure. Its London listing for a software engineer in observability and profiling gives a £325,000 to £390,000 salary range. In the United States, an OpenAI compute-foundations vacancy lists $255,000 to $490,000 in base pay plus equity. These are advertised ranges, not evidence that every hire receives the maximum or that engineers across the industry earn similar sums.

Why pay so much? Frontier AI labs need a narrow mix of expertise in large-scale infrastructure, model training, performance engineering, safety and reliable products. A systems engineer who improves the efficiency of a vast compute fleet can affect costs far beyond an ordinary software budget. Competition for that expertise is intense, and equity can be used to persuade recruits to join or stay while private-company valuations climb.

There is a second story beneath the headline numbers. AI companies are simultaneously promoting tools that can automate parts of software development while spending heavily to recruit human engineers capable of building those tools. That is not necessarily a contradiction. Automating routine tasks may increase the value of people who can design systems, diagnose failures and make consequential technical judgments. It may also widen the gap between a small set of scarce specialists and the wider technology workforce.

For readers weighing an AI career, the right comparison is not between their current salary and OpenAI’s headline average. It is between the requirements, location, base pay, equity terms and risk of specific roles. The Guardian’s figures demonstrate the scale of the talent race; the job adverts show what some specialised positions actually offer. Keeping those measures separate makes the story more revealing, not less.