Seven thousand five hundred dollars per employee per month is a number that stops you mid-scroll.
That’s the figure coming out of the Ramp AI Index, which tracks AI spending across American businesses. The top 1% of firms, which Ramp calls “AI-pilled,” are burning through $7,500 per employee every single month on AI tools, tokens, and platforms. To put that in context: the average software engineer pulls in roughly $16,000 a month in salary. So no, companies aren’t quite spending more on AI than on humans yet. But they’re getting closer than most people realize, and the trajectory is hard to ignore.
The 1% Problem
Here’s the thing about that $7,500 number. It’s an extreme. It represents the behavior of a tiny slice of companies that have gone all-in on AI infrastructure in a way most organizations haven’t even considered yet.
The top 10% of firms? They’re spending about $611 per employee monthly. That’s meaningful, but it’s a different planet from $7,500. And the median company? Eleven dollars and thirty-eight cents per employee per month. Basically the cost of one enterprise software seat. Translation: most businesses are still dipping a toe in the water while a small group is doing cannonballs off the high dive.
This kind of distribution isn’t unusual in tech adoption curves. Early adopters always look like outliers until they don’t. The interesting question isn’t whether $7,500 is a lot. It’s what those companies are actually doing with it, and whether that spending is producing anything worth the invoice.
Tokens Are the New Headcount
The backdrop here matters. An Nvidia executive recently said publicly that the cost of compute at his company now exceeds the salary costs of his employees. Mercor’s CEO said the startup is spending more on tokens for internal AI agents than on its actual human headcount. These aren’t hypotheticals anymore. They’re real operational realities at specific companies right now.
What’s a token, exactly? Think of it as the basic unit of text that an AI model reads and generates. Every time a model processes a document, writes code, answers a question, or runs an automated task, it’s consuming tokens. And tokens cost money, especially when you’re running large frontier models at scale across an entire organization. At low volumes, token costs are trivial. At enterprise scale, with agents running autonomously around the clock, those costs compound fast.
The AI-pilled firms aren’t just buying ChatGPT Plus for their employees and calling it a day. They’re building internal agent systems, running automated pipelines, and spinning up workflows that operate without human input. That’s where the token bills get serious. And that’s what separates the $11 median spender from the $7,500 outlier.
Mix and Match at the Top
One pattern worth understanding: the top 1% of firms aren’t loyal to a single AI provider. They bounce between multiple frontier models and use platforms that give them access to cheaper open-source alternatives depending on the task. Call it model arbitrage. Some tasks need the best available reasoning model. Others can run on something leaner and cheaper. The sophisticated operators know the difference and route accordingly.
This is actually a smart strategy, and it’s one that most mid-market companies haven’t figured out yet. The instinct for most organizations is to pick one provider, standardize on it, and move on. The power users treat AI infrastructure more like a cloud computing strategy, where you’re constantly optimizing for cost and capability across a portfolio of options rather than betting everything on one vendor.
That mix-and-match approach also creates some insulation against the pricing volatility that’s been a feature of the AI market. When one model gets expensive or hits capacity limits, you route to another. When a new open-source model drops that’s good enough for 80% of your use cases, you shift volume there. Flexibility is the whole game.
Growth Still Climbing
Despite whatever economic pressure companies are feeling right now, AI spending among the top firms isn’t slowing down. Among the AI-pilled group, spend grew 14.1% per employee just last month. Whether that pace holds is genuinely unknown. Ramp’s data doesn’t tell us what comes next, and it’s too early to call this a durable trend versus a burst of investment ahead of a plateau.
But 14.1% monthly growth is not a rounding error. That’s aggressive expansion of AI infrastructure at exactly the moment when some companies are tightening budgets elsewhere. It suggests the firms at the top of the spending curve aren’t treating AI as a discretionary experiment. They’re treating it as core operational infrastructure, the same way they’d treat cloud hosting or payroll software.
That’s a meaningful shift in how AI is being categorized internally. Not a pilot. Not a productivity perk. A line item that sits next to the foundational costs of running the business.
The Ramp AI Index is updated monthly, with the next data release expected in July 2026.
