Companies keep blowing through token budgets — and after Nvidia and Mercor executives said AI costs now rival payroll, the big question is whether companies spend more on AI than on humans. Not yet, according to fresh Ramp AI Index research: even the most AI-obsessed firms spend $7,500 per employee per month, below the roughly $16,000 a month an average software engineer costs.
Who Counts as AI-Pilled
Ramp describes the top 1% of American firms by AI adoption as ‘AI-pilled.’ Their spending is the eye-popping number, but also a ceiling: it is not more than an engineer’s salary — yet.
The Spending Distribution
- Top 1%: about $7,500 per employee per month
- Top 10%: about $611 per employee each month
- Median firm: about $11.38 per employee monthly — roughly one seat on an enterprise plan
The gap between the median and the top tells the adoption story: most companies pay for a license, while a tiny fraction run heavy agent workloads.
Token Bills Keep Climbing
Despite the gap, AI spend is still rising — 14.1% growth per employee last month among AI-pilled firms. That momentum follows claims from an Nvidia executive that compute now costs more than his employees’ salaries, and Mercor’s CEO saying the startup spends more on tokens for internal agents than on headcount.
Mixed Models and Open Source
Power users do not commit to one vendor. The top 1% tend to bounce between multiple frontier models and platforms that offer access to cheaper open-source models, trading fidelity for cost as agent workloads scale.
What This Means
The $7,500 figure is a snapshot, not a ceiling. If agent workloads keep compounding at double-digit monthly rates, the AI-versus-humans line could be crossed sooner than the median firm expects — the debate is really about which cost wins the race, compute or headcount.