Our Take
The real story isn't the headline number—it's that the top 1% spent 14.1% more last month alone, suggesting AI costs are on a collision course with headcount spend.
Why it matters
CFOs and engineering leaders need to understand the trajectory: token costs are rising faster than salaries fell, and at current acceleration, the cost-per-employee question shifts from rhetorical to budget crisis within 18 months for firms with heavy model usage.
Do this week
Finance: Map your current monthly AI spend per FTE against Ramp's 1% benchmark ($7,500) and 10% baseline ($611) this week so you can forecast when model costs will exceed your next budget cycle.
The top 1% are already spending heavily on AI
Companies classified as "AI-pilled" by Ramp's AI Index are spending approximately $7,500 per employee per month on AI infrastructure, tokens, and compute (per Ramp AI Index). For context, that falls short of the $16,000 monthly salary for an average software engineer, but the gap is narrowing. The top 10% of firms spend about $611 per employee monthly, while the median spend sits at $11.38—roughly the cost of an enterprise SaaS seat.
More importantly, the trajectory is steep. Among the highest-spending cohort, monthly spend per employee grew 14.1% in the most recent measurement period. Most of these firms are mixing commercial frontier models from multiple vendors with cheaper open-source alternatives to manage costs.
Burn rate is accelerating, not stabilizing
The conversation has shifted from "will AI cost more than people?" to "when?" An Nvidia executive recently stated that compute costs now exceed employee salaries at his firm. Mercor's CEO reported spending more on internal agent tokens than on headcount. These aren't outliers anymore; they're data points in a broader compression.
At 14.1% monthly growth among heavy users, the math gets uncomfortable fast. If that rate holds even half the time, firms in the top 1% will match the $16,000 engineer baseline within 18 to 24 months. Budget committees are still treating AI spend as a line item; CFOs should be modeling it as exponential.
Audit your token velocity now
If your firm is in the top 10% on spend ($611+), start separating your AI costs by use case: inference on proprietary data (RAG), fine-tuning, experimentation, and production agents. Most companies have no visibility into which workload is actually driving the bill. Once you know, you can negotiate volume discounts, lock in pricing before further rate hikes, or migrate expensive workloads to open-source models with acceptable latency tradeoffs.
The firms spending $7,500 per employee are using that spend strategically—bouncing between models to exploit price differences and capability tiers. That's a hedge against lock-in; it's also a signal that they expect prices to keep moving and that single-vendor strategies will look expensive in hindsight.