Newly filed accounts for OpenAI's UK subsidiary have produced two eye-catching numbers: an average salary of about £794,000 and a total staff cost of about £2.3m per head. Both come from the same filing, they measure different things, and neither describes a typical OpenAI employee. The figures were reported on October 15, 2026, by Business Matters and LADbible, among others, and were covered by the Telegraph under the £794,000 headline.
This post sets out what the accounts say, how the headline numbers are built, what they leave out, and why the AI talent market is relevant to anyone hiring, building a career, or tracking where AI research happens.
TL;DR
| Item | Figure (from the accounts, as reported) |
|---|---|
| Entity | OpenAI UK Ltd, London |
| Average employees in 2025 | 110 |
| Wages and salaries | £87.3m |
| Share-based compensation | £168m |
| Total staff costs | £258m (2024: £60.7m) |
| Implied average salary | About £794,000 (£87.3m divided by 110) |
| Implied total cost per head | About £2.3m (£258m divided by 110) |
| Turnover | £293m (2024: £79.2m) |
| R&D costs | £117.8m (2024: £48.2m) |
| R&D tax credit income | £4.7m |
How the headline numbers are built
Both numbers are simple divisions, which is why they should be read carefully.
£794,000. Wages and salaries of £87.3m divided by an average monthly headcount of 110 is about £794,000. That total can include bonuses and other cash pay, and a headcount that changes during the year affects the average. It is the mean of the whole payroll, so a handful of very large packages move it a lot.
£2.3m. Business Matters says total staff costs were £258m, made up of salaries, pension costs, share-based compensation and other benefits. Divided by 110 that is about £2.3m. The largest single component was share-based compensation at £168m, bigger than cash pay. Under accounting rules, share-based compensation is an expense recognised over time based on the fair value of awards, so it measures the cost to the company, not what an employee can sell or has received.
One outlet's version needs a correction: LADbible says the "average salary was £2,318,181," but that figure is the all-in cost per head, not salary. The cash average is the £794,000 figure. This is the sort of conflation that makes the numbers travel badly.
Means are not medians
The accounts do not give a median or a distribution. LADbible cites the Guardian as reporting a large discrepancy between the lowest and highest earners, with the biggest packages going to researchers who build OpenAI's models. A subsidiary of 110 people may include researchers, engineers, sales and policy staff, and finance and operations roles. A few multimillion-pound research packages would pull the mean far above what the typical employee earns. Nothing in the filing as reported tells us what the middle of the distribution looks like.
For context, LADbible quotes the Office for National Statistics median annual earnings for full-time UK employees of £39,039. Comparing a mean at a frontier lab with a national median is a vivid but loose comparison; it shows scale rather than fairness or typical pay.
What the subsidiary does
According to the accounts as reported, the UK company develops "UK-based commercial capabilities" for the OpenAI group and carries out research and development, which it charges to its parent. Turnover rose from £79.2m in 2024 to £293m, which the company attributed to expansion of sales, marketing and R&D activities. Administrative expenses rose to £264.1m from £64.6m. In other words, this is largely an intercompany arrangement: costs are recharged to the parent, so the profit and loss of the UK entity does not tell us about OpenAI's overall finances.
Business Matters also reports that OpenAI signed a lease in April on an 88,500 sq ft King's Cross office with capacity for 544 staff, as part of plans to more than double a London workforce of roughly 200 people at the time. That is larger than the 110-person average in these accounts, which implies the subsidiary's headcount was already growing during and after 2025. When OpenAI announced its London office in 2023, it said it wanted to "build dynamic teams in research, engineering and go-to-market functions."
How skewed could the distribution be?
The filing as reported gives no spread, but a simple illustration shows why the mean says little. These are hypothetical splits, not data from the accounts.
| Scenario (hypothetical) | Staff | Cash pay each | Total cash | Mean |
|---|---|---|---|---|
| Even split | 110 | £794,000 | £87.3m | £794,000 |
| 20 top researchers at £2.5m, 90 others | 20 / 90 | £2.5m / £414,000 | £50.0m + £37.3m | about £794,000 |
| 10 top researchers at £4m, 100 others | 10 / 100 | £4m / £473,000 | £40.0m + £47.3m | about £794,000 |
In the third case, 100 of 110 people average about £473,000, while ten people average £4m, and the headline mean is unchanged. Even the lower figures are very high by UK standards, which is the real point: the range of plausible distributions is wide, and the mean cannot tell you which one applies. The same logic applies even more strongly to the £2.3m cost figure, because share awards are typically concentrated among the most senior and sought-after staff.
A second caution is time. Accounting for share-based pay spreads the cost of an award over its vesting period. A large cost in 2025 can reflect grants made in earlier or later periods, and a rising valuation raises the value of new grants. A fourfold increase in staff costs in one year, from £60.7m to £258m, therefore reflects headcount growth, new grants at higher valuations and any catch-up charges together. The accounts as reported do not allow us to separate those effects.
The R&D tax credit
The accounts record £4.7m of income from a UK research and development tax credit, according to Business Matters. This is a standard government scheme for companies doing qualifying R&D, and its appearance in the accounts is not unusual. It is nevertheless the kind of detail that becomes political when set beside pay headlines. A similar debate is under way in the US over a different lab; see our post on Senate Democrats and Meta's $6.6 billion R&D tax credits.
The 2025 context: a talent war
LADbible notes that the figures are from 2025, "when there was a bit of a talent war amongst AI companies," and says Meta was reportedly offering multi-year packages worth hundreds of millions of dollars to some people. It adds that OpenAI has been hiring for robotics, with base pay ranges of $177,000 to $500,000 before equity on listings seen in September, as reported by Business Insider. Those are reported ranges, not accounts data.
Share-based compensation matters in this market because it ties pay to the company's valuation, which Business Matters says the OpenAI group is reportedly seeking to raise at around $1.4 trillion. If the shares are worth more, the accounting cost of new grants goes up, and the retention value of old ones does too. That is also why departures draw attention; see our report on OpenAI's compute chief leaving and the OpenAI and Apple trade-secrets dispute over a rehired employee.
What is claimed versus what is verified
| Claim | Status |
|---|---|
| 110 average employees, £87.3m wages, £168m share-based pay, £258m total | Reported consistently by multiple outlets from the filed accounts; we did not open the Companies House filing itself |
| Average salary about £794,000 | Arithmetic from those figures; a mean of cash pay |
| "Average salary £2.3m" | Incorrect framing; £2.3m is total staff cost per head, including share awards |
| Typical OpenAI UK employee earns this | Not supported; the mean is skewed and the distribution is not disclosed |
| OpenAI commented | Reported as declined to comment (via the Guardian, relayed by LADbible) |
| Group valuation of $1.4 trillion and funding target | Reported by Business Matters as OpenAI "seeking"; not part of the accounts |
To check the numbers yourself, search for OpenAI UK Ltd on the Companies House register and open the newly filed accounts for 2025. The notes to the accounts should show the staff-cost breakdown and the average number of employees.
Why this matters if you work in or around AI
- Hiring. Cash and equity offers at frontier labs are far above what most employers can match. Smaller teams compete on mission, scope, ownership and speed rather than pay.
- Career planning. The extreme tail is concentrated in research roles. For most engineers and product people, market pay is much lower than these means imply. Treat the averages as a ceiling signal, not a benchmark.
- Location. A £293m UK turnover and a planned 544-seat office show London is becoming a serious hub, which affects where AI jobs are, as in our look at OpenAI's EU agentic AI report. The labour side of AI in the UK is also changing in the other direction, with UK employers cutting entry-level jobs.
- Reading numbers. When a headline says "average salary," check whether it divides cash pay or total cost, and whether it is a mean or a median.
Limits of this analysis
We did not read the filing directly and rely on reporting by Business Matters and LADbible, plus a search summary of other outlets. Details such as the headcount definition, the split between bonus and base pay, and the valuation method for share awards are in the notes to the accounts, not in these articles. If a primary reading differs, we will correct this post.
Related reading
- OpenAI's compute chief leaves
- OpenAI agrees not to use Apple trade secrets
- OpenAI's EU agentic AI report
- Senate Democrats and Meta's R&D tax credits
- UK employers cut entry-level jobs
- OpenAI's next flagship: what is claimed
Figures are as reported on October 15, 2026 from OpenAI UK Ltd's 2025 accounts; they are averages and may be revised if the filing is read differently.