The top 1% of US firms are now spending $7,500 per employee per month on AI, according to fresh data from the Ramp AI Index, which tracks corporate AI adoption across American businesses. Ramp labels that cohort 'AI-pilled.' For the heaviest users, monthly AI spend per head is closing in on half the roughly $16,000 monthly salary of an average US software engineer.
The gap between the top and the middle is enormous. The top 10% of firms spend about $611 per employee per month — less than a tenth of the leaders. The median firm spends just $11.38, roughly the cost of a single enterprise seat on a chatbot product. Most American companies, in other words, are still buying AI by the head, not by the workload.
The framing is no longer hypothetical. An Nvidia executive recently said the cost of compute at the company is now greater than the salaries of its employees. Last week, Mercor's CEO said the startup is spending more on tokens for internal agents than on employee headcount. Both data points come from companies whose entire workflow is wired into model inference.
“the cost of compute is now greater than the salaries of his employees”— Nvidia executive, Nvidia executive
Key facts
- 01The top 1% of US firms spend $7,500 per employee per month on AI, according to the Ramp AI Index.
- 02The top 10% spend $611 per employee per month; the median firm spends just $11.38 — roughly one enterprise seat.
- 03Per-employee AI spend among the top 1% grew 14.1% last month alone.
- 04For comparison, the average US software engineer earns about $16,000 per month in salary.
- 05Heavy users mix frontier models with cheaper open-source platforms rather than committing to a single vendor.
Ramp's numbers suggest those anecdotes are leading indicators rather than outliers. Among the AI-pilled top 1%, per-employee spend grew 14.1% last month. That is a single-month figure, not annualized — sustained at that pace, the cohort would more than quadruple its spend over a year, though Ramp does not project that the rate will hold.
The composition of that spend is also shifting. Ramp finds that the top 1% of firms tend to mix and match, bouncing between multiple frontier models and platforms that give them access to cheaper open-source models. That is consistent with what enterprise buyers at OpenAI and Anthropic customers have been doing all year: routing the expensive reasoning calls to frontier APIs and pushing high-volume, lower-stakes traffic to open-weight models hosted on cheaper infrastructure.
“the startup is spending more on tokens for internal agents than on employee headcount”— Brendan Foody, CEO of Mercor
The Ramp AI Index measures actual corporate card and bill-pay flows running through its platform, which gives it visibility into what companies are paying for rather than what they say they are using. That makes the $7,500 figure a spending number, not a survey number — it reflects invoices from model providers, inference platforms, and AI tooling vendors that hit corporate accounts.
The number also reframes the 'AI versus jobs' question that has dominated coverage for the past year. At $7,500 per employee per month, the AI-pilled cohort is spending less than half what it would cost to hire one additional engineer per existing employee — but the spend is per existing employee, layered on top of payroll, not in place of it. The substitution math only kicks in if AI spend keeps climbing while headcount holds flat or falls.
That is precisely the scenario several AI-native startups are now describing. Mercor's comment that token spend has overtaken headcount spend is a signal that, at small companies built around agents, the cost structure has already inverted. Whether the same inversion shows up at larger enterprises depends on whether middle-of-the-pack firms — the ones still spending $11 per head — start climbing the curve.
There are reasons to think the 14.1% monthly growth rate will cool. Model prices have been falling across the board, with cheaper frontier tiers and open-source alternatives putting downward pressure on per-token costs. If the top 1% successfully shifts more workloads to cheaper models, the dollar figure could plateau even as usage keeps rising. Ramp itself notes it is 'not yet clear if that trend will continue.'
The harder question for the rest of the market is whether the gap between the top 1% and the median closes or widens. A $7,500-versus-$11.38 spread implies the AI-pilled cohort is running a fundamentally different operating model — one where compute is a primary input rather than a software line item. If that model produces measurable productivity gains, the median will be forced to chase it. If it does not, the top 1% will look like a cautionary tale by the time the next index drops.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
