One Percent of Firms Pay 80% of the AI Bill. That Cohort Just Stopped Growing.

Ramp published its AI Index numbers on November 6, and the summary line going around was that AI spending has stalled. That framing is defensible and it misses the mechanism. The index is built from card, invoice and ACH transaction data across more than 85,000 US businesses, and what it shows is a market where consumption and revenue have come apart, driven by a very small group of buyers who stopped spending more.
Start with the two headline figures. In the week ending October 23, total business AI token volume was up 37.1 percent compared with the week ending September 25. Total token spending over the same period rose 8.9 percent.
The gap has a price tag on it
The arithmetic is not mysterious. The blended price per million tokens across OpenAI and Anthropic fell from $0.76 to $0.56, a drop of 26.8 percent. The two vendors moved differently: OpenAI's effective price fell 42.4 percent, Anthropic's 15.7 percent.
Those are list-price cuts working their way through. OpenAI halved API prices for GPT-6 Sol and GPT-6 Luna in September. Anthropic cut prices on Fable 5.1 and Opus 5.5, and said in a filing that the decision reduced its revenue growth by roughly $600 million.
Six hundred million dollars is the cleanest number in this whole story. It is what a lab pays to buy volume, stated by the lab, in a document with legal weight behind it.
There is a kernel of genuine demand softness in the series too. Token spend grew only 1.5 percent across August and September, after a 42 percent jump in July. Then it re-accelerated to 8.9 percent from late September to late October. A two-month plateau followed by a resumption looks more like buyers getting better at the same job than like a collapse of demand.
Ramp's lead economist Ara Kharazian points at three forces: frontier price cuts, enterprise discounts, and buyer skill. The third one is the one that lasts. Teams route work to cheaper models, set hard budgets, pick cheaper defaults. Notion co-founder Akshay Kothari's summary of the dynamic was that tokens go brrr and costs go ahhh, which is funnier than it is precise but does describe a real change in purchasing behaviour.
The concentration is the story
Here is the number that explains why a 1.5 percent monthly change in spend can flatten a provider's revenue line.
About 1 percent of firms account for roughly 80 percent of all spending on OpenAI and Anthropic combined. In September, that cohort spent about $9,568 per employee per month. That was up 11.7 percent from the prior month and still 3.1 percent below the July high of $9,873.

The median firm paying for AI spends $15.70 per employee per month. The top decile spends about $813. Kharazian's read is that the top 1 percent had been clearing roughly $10,000 per employee per month through the summer, and that slipping below that mark was enough on its own to flatten total revenue even while median and top-decile spending kept growing.
The composition of that 1 percent matters as much as its size. Kharazian says nearly all AI spending comes from venture backed, high-growth technology companies. Coding agents are the workload that justifies the number. A software company can put an agent on every engineer and see the output. A dental practice cannot.
That is the uncomfortable part of the thesis. If the top of the market is a cohort defined by venture funding and software margins, then the spending curve is tied to a funding environment, not to the general adoption of AI. Adoption did keep climbing: 55.9 percent of tracked businesses in September, up 0.6 points from August and 12.1 points year over year. Ramp notes its adoption figure is probably understated because free tools and personal accounts used for work do not show up in transaction data.
So you have broad, cheap adoption and narrow, expensive adoption. The first is growing steadily. The second is what the revenue line depends on.
Market share and the IPO problem
In the week ending October 23, Anthropic held 60.5 percent of business token spending and OpenAI 35.1 percent. By paid adoption in September, Anthropic sat at 43.8 percent and OpenAI at 40.6 percent.
Anthropic has the larger share of the bigger spenders and a smaller share of the long tail. That is a good place to be if the top 1 percent keeps growing, and an exposed place to be if it does not. The company is also the one that just told investors its own price cuts cost it $600 million in growth.
Which brings in the valuations. Anthropic's expected IPO valuation slid from roughly $2 trillion in early October to a range of $1.2 trillion to $1.3 trillion. OpenAI reportedly lowered its proposed valuation from $1.4 trillion to $1.25 trillion and trimmed its raise by about $10 billion. Both moves landed in the same window as this spending data.
You do not need to believe the index is perfect to see the shape of the problem. A company priced on hypergrowth needs either volume growth that outruns price declines, or a new category of buyer who spends like a software company. Price cuts are rational for winning volume. They are also, arithmetically, a transfer from revenue today to revenue later, and the later part is a bet.
What the data does not say
Ramp sees card, invoice and ACH transactions. It does not see usage inside a prepaid credit that was bought last quarter, or work running on committed cloud contracts, or anything billed through a reseller. Its index is a view of how businesses pay, which is close to how providers earn but not identical.
The token volume figure is also a proxy. Different models tokenise differently, so a shift toward cheaper models with larger token counts can make volume look healthier than capability delivered. A 37 percent volume increase against a 27 percent price decline is consistent with a real increase in work done. It is also consistent with a mix shift.
And the index covers one country's business payments, weighted toward companies that use Ramp. The 85,000 number is large, and it is a sample of one country's card and invoice flow.
The question behind the number
The thing worth watching is not next month's spend figure. Look instead at whether AI finds a second category of heavy buyer outside venture funded software.
Kharazian is openly skeptical that a doctor's office or a car dealership will ever spend like a coding-heavy startup, and he is probably right that they will not. But they do not have to. They have to spend like a business that found one workflow worth automating and left it running. Fifteen dollars per employee per month is not nothing if enough employers sign up.
Until that happens, the business model for frontier labs rests on a narrow base of customers whose own budgets depend on someone else's willingness to fund them. That is a strange foundation for a trillion-dollar valuation, and it is why a 1.5 percent monthly change in a spending index is being read as a warning rather than noise.
Related articles
METR Spent 11 Weeks Inside Anthropic. The Number Is 1.9, and Most of the Evidence Is Blacked Out.
A policy threshold with a number in it, an auditor who says it may have been crossed, and evidence nobody outside can check.
Claude Can Now Read Your Bank Statements. Anthropic Deliberately Gave It No Hands.
The read-only design is the product, and unlinking does not retract what is already in a chat.
Adobe Stopped Asking Marketers to Open Adobe
For a decade the strategy was own the interface. Adobe just bet the opposite.
CrowdStrike Shipped a Kill Switch for AI Agents. The Hard Part Is Knowing What to Kill.
A kill switch turns a guess into something enforceable. Enforcement requires a list, and the list is the part nobody has.