The median company spends $11.38 a month on AI. You can beat that this afternoon.
The median American business spends $11.38 per employee per month on AI.
That figure comes straight from the Ramp AI Index for June 2026, built on actual card and bill-pay data across more than 70,000 businesses. That’s the real ledger, not a survey where people guess what they spend. (TechCrunch covered it too, if you want a second set of eyes on the math.)
$11.38, per person, per month. That’s about one seat on ChatGPT or Claude, and not a premium one. Round it up and it’s a large pizza. The typical company in America has decided that its entire per-person AI budget for the month is a large pizza.
We acknowledge, and will talk a little later about a better approach, spending is an imperfect metric. But spending is a place to start.
The gap everyone’s worried about
There’s a lot of hand-wringing about a widening gap between the companies that “get” AI and the ones that don’t. And the gap is real. The top 1% of firms, the ones Ramp politely calls “AI-pilled,” spend about $7,450 per employee per month. Against a median of $11.38, that’s a gap Ramp pegs at roughly 680x.
Six hundred and eighty times. Read that again. That’s almost three orders of magnitude between the leaders and the middle of the pack. We’re not talking percentage points.
And $11.38 is the median, which means half of all companies spend even less than that. The climb from there is steep and lopsided. The top 10% of firms spend about $611 per employee a month, and the top 1% spend roughly twelve times that again. Almost the entire gap is stacked inside that last percentile. You don’t need to live up there to be ahead of nearly everyone. (Ramp doesn’t publish how many companies spend exactly zero, but a median this low means a very large share are barely spending at all.)
When most people hear “680x,” they assume it’s a money problem. The big players have the budgets, the rest of us are priced out, the gap only grows. That story mostly gets it backwards, and the same dataset tells you why.
$11.38 buys an afterthought
A 30-person company spending the median is laying out about $341 a month on AI for the whole shop. A year of that is roughly $4,100. That’s not a strategic investment. That’s the kind of line item that shows up on the card statement and nobody can quite remember approving.
The constraint here is not money. A business that can make payroll for 30 people can find $341 a month behind the couch cushions. The median sits at $11.38 per employee because most companies haven’t really decided to use this stuff. A seat here, a free trial there, nothing pointed at actual work. Nobody got priced out. Nobody got started.
Which means the bar to clear the median is comically low. You don’t need a budget. You don’t need a data team. Spend more than a large pizza per head, on purpose, and you’re already ahead of half the country.
What $11.38 actually buys vs. what “trying” looks like
Let’s make this concrete. The following numbers are illustrative: a hypothetical shop, invented figures to show the shape of the thing.
Picture a 30-person HVAC company sitting in the bottom half of that distribution. Right now the office manager has a single ChatGPT seat she uses to reword the occasional customer email. Call it $20 a month, total. Across 30 people, that’s well under a dollar a head.
Now say the owner actually decides to try. He buys frontier-model seats for the 15 people who spend their day at a keyboard, about $30 each. That’s $450 a month, or roughly $15 per employee across the whole company. He’s now spending above the median, he did it on purpose, and it cost him about what the shop spends on coffee.
What does that $450 buy in a month? The dispatcher stops hand-typing the same five customer texts. The bookkeeper drafts the monthly statements in a tenth of the time. The owner stops writing proposals from a blank page on Sunday nights. None of that is exotic. None of it requires an integration, a consultant, or a “roadmap.” It requires somebody to hand out the seats and say “use these.”
Say each of those 15 people saves three hours a month, a conservative guess. At a loaded labor cost of, say, $40 an hour, that’s 45 hours, or about $1,800 of recovered time, against $450 spent. The math isn’t subtle. Those figures are made up to illustrate the shape (your shop’s will differ), but the ratio is the point, and the ratio is lopsided.
The leaders aren’t spending crazy money either
Even the AI-pilled top 1%, spending $7,450 per employee a month, are spending less than half of what one typical software engineer costs, about $16,000 a month all-in. And their spend grew 14.1% in a single month, so they’re leaning in hard.
So even the most aggressive AI spenders in the country are still spending less than one extra hire. The “leaders” are spending hire-adjacent money and getting more than one hire’s worth of output back. Nobody’s betting the company. That’s the actual frontier, and it’s nowhere near as far away as $11.38 makes it feel.
The distance between you and the median is one decision. The distance between the median and the leaders is a few more decisions, made repeatedly, over time. At no point on that curve does the money become the hard part.
Spending the money is the easy part
Catching the median is cheap. Pulling ahead is a different question, and it has little to do with the size of the number on the card statement. Spend and return are not the same thing.
Two companies can both jump from $20 to $450 a month and get completely different results. One hands out seats and tells everyone to “use AI.” The other points the same tools at one specific workflow, the month-end close or the quoting process or the service-intake queue, and rebuilds that workflow around what the tool is actually good at. Same spend. The second one compounds.
That’s what the dollar figures hide. Spending a little catches you up to the pack. Spending a little the right way, aimed at process instead of just access, is what turns a subscription into an advantage. The money gets you into the room. What you point it at decides whether you win it.
The gap is a decision, and deciding is cheap
None of this is that different from what we’ve been saying for a while now. We wrote that You can just install Claude Code tonight: the tools are sitting right there and the move is to start using them. We wrote that process beats raw capability, that the win comes from putting the tool into the workflow rather than owning the fanciest model. And we wrote that you can just decide to be AI native. The Ramp data is the receipt for all of it. We weren’t speculating. The median company really is spending $11.38.
The 680x gap looks terrifying from a distance. Up close, it’s a budget of $11.38 sitting next to a budget that hasn’t been set yet. One of those is fixed. The other is a Tuesday-afternoon decision and the price of a few subscriptions.
The median spends a large pizza. Spending two, pointed at the right work, is somehow a competitive advantage.