US companies spent more on AI in October than in any month before, but only 1.65% more than in September. A typical company pays $16.32 per employee each month. The top 1% of spenders pay a median of $10,200.
The same product, bought two different ways
Over a year, $10,200 per employee comes to about $122,000 for every person on the payroll, and it is 625 times the typical bill. No company pays that to give staff a smarter search box. A bill that size means AI is doing work that people used to be paid for, and the spending shows which work. Software teams account for 72% of all the AI spending in the data, and the biggest spenders are mostly venture-backed firms in tech, finance and technical services.
A coding agent reads a whole codebase, writes a change, runs the tests and tries again, and every loop uses far more text than a chat does. So the bill grows with the amount of work handed over, not with the number of people who have a login. The typical company pays $16 because it bought AI the other way: one assistant per person, used to draft an email or tidy a report.
A bill of $16 is what a bolt-on looks like
The factories that swapped steam for electricity but kept their belts were not foolish. The old layout worked well enough, and rebuilding meant tearing down the building. What the owners got was a cleaner and more reliable engine. The gains went to the managers who rebuilt around electricity a generation later.
The same logic explains why cheaper AI changes so little for most companies. The price of processing text fell 21% in October, and the typical bill rose by less than a dollar. A company paying $16 a head was never held back by price. Falling prices matter a great deal to a firm whose AI does whole jobs, and that is why usage keeps growing faster than spending in the data. For everyone else, doubling the bill to $32 for the same drafting help would change nothing.
Customer support is the first test outside software
Customer support and customer success teams went from under 1% of AI spending in June to 13.3% in October, while software teams slipped from their summer peak. Support work also tends to run on cheaper, faster models, so it will never look like the coding bills. It does not need to.
Take a 40-person online shop that gets 3,000 support messages a month, most of them asking where an order is or how to return something. If AI is connected to the order system and closes half of them on its own, at about $1 each (support tools charge between $0.05 and $1.25 per resolved ticket), the bill is $1,500 a month. That is $37.50 per employee, more than double the typical company's bill and a tiny fraction of the top tier's.
For $1,500, around 1,500 routine messages stop reaching a person. At about six minutes each, that is 150 hours a month, close to one full-time job. The people on the team do not disappear. They take the refunds, the angry customers and the cases the AI cannot settle, and the manager decides how much money the AI is allowed to give back without asking.
That is the difference between the two factories. In the first, every support agent has an AI tab open and the bill stays near $16. In the second, AI answers first and a person checks, and the bill is a little higher but the work is organised differently.
The number worth watching in your own company is therefore not what it pays for AI. It is the list of jobs where AI now goes first and a person reviews. At most companies that list is still empty, and support is the most likely first entry.