Corporate spending on AI rose 1.5% across August and September, after monthly growth of 42% in July, according to Ramp, the payments platform whose AI index tracks transactions across more than 85,000 businesses. Token volume kept climbing over the same period. The money did not keep pace, and that gap is now the central fact in the pricing of every AI company preparing to sell shares.
The index shows concentration as much as deceleration. The top 1% of spending companies paid about $9,568 per employee in September, down 3.1% from their July peak. Median firms spend $12 to $15 per employee per month, and the top decile $600 to $1,000. Growth among those smaller cohorts was not enough to offset the pullback at the top, where a disproportionate share of model revenue originates. Ramp also recorded a partial recovery, with token spending up 8.9% between Sept. 23 and Oct. 23.
The cause is price rather than appetite. Ramp's lead economist Ara Kharazian told TechCrunch that usage kept rising while spending flattened, and attributed the divergence to price cuts. OpenAI halved API pricing for GPT-6 Sol and GPT-6 Luna relative to the earlier GPT-5.6 promotional prices. Anthropic cut prices on several Claude models, a move its filings say reduced revenue growth by roughly $600 million. Enterprise discounts negotiated for volume commitments, and internal controls that route routine tasks to cheaper models, finished the job.
Adoption is still broadening. Ramp found 55.9% of businesses on its platform paid for AI services in September. Anthropic reached adoption with 43.8% of those businesses and OpenAI with 40.6%, with substantial overlap between the two.
That is the backdrop for the largest listing of the decade. Anthropic's amended registration statement, filed Nov. 2, reported $17.0 billion of revenue for the quarter ended Sept. 30 and left the share count and price range blank. Reuters, citing four people familiar with the discussions, reported that prospective investors were treating the $1.5 trillion valuation as the upper end of an acceptable range rather than a starting point, and that the roadshow expected in the week of Nov. 9 could slip. Other coverage of those discussions put the working range nearer $1.2 trillion to $1.25 trillion.
Two variables are being priced simultaneously: the multiple and the trajectory. Investors are waiting for Anthropic's October revenue, which the filing left blank. If the September flattening was a pricing artifact, October should show it. If it reflected saturation among the customers that actually drive model revenue, the blank is the warning.
Kharazian's structural observation is the one enterprise buyers should take away. Almost all AI spending comes from venture-backed, high-growth technology companies, he said, which means that if spending slows for any reason, it slows across the cohort at the same time, and the effects cascade. He added that industries outside software and finance, naming a doctor's office or a car dealership, may never need to spend at the level the top 1% currently does. That is a statement about the ceiling of the market, not its floor.
The cost side explains the urgency. Anthropic's compute costs consumed about 58% of third-quarter revenue, according to analysis of its disclosures. Price cuts designed to win volume reduce revenue per token, while compute obligations are fixed by multi-year contracts. Growth has to outrun the obligation, which is why the IPO pricing conversation and the Ramp data are the same conversation conducted in two rooms.
The selloff forming that backdrop was itself extreme. More than 50 technology stocks in the S&P 1500 fell at least 10% in the Nov. 5 session, and Vanda Research measured retail selling at the strongest level since the pandemic, according to Reuters coverage. Keith Lerner, chief investment officer at Truist Advisory Services, said technology had underperformed the S&P 500 by 10.7% over the preceding month, the most extreme one-month gap since May 2002.
Private-market demand has not disappeared, which is why the negotiation is about price rather than appetite. Rick Meckler, a partner at Cherry Lane Investments, said he expects interest in large private technology companies to persist when they list, while allowing that valuations set in private rounds will come down. OpenAI, meanwhile, was reported this week to have raised an additional $11 billion at a valuation of $1.25 trillion, a private mark close to the range Anthropic investors are now discussing publicly.
For procurement and finance teams, four practical consequences follow. Renegotiate on committed volume rather than list price, because discounting is now structural and the published rate is a starting position. Route workloads by task complexity, since the price gap between frontier and mid-tier models is the largest controllable lever. Avoid multi-year prepaid capacity in models unless usage is genuinely inelastic, and prefer reserved discounts with termination rights. And measure token consumption internally every month, because the binding constraint has moved from access to budget.
The data does not show AI demand collapsing. It shows a revenue model shifting from usage-based expansion to price-based competition at precisely the moment a $1.5 trillion valuation requires the opposite. Whether Octobers 8.9% rebound continues through November and December is the number that will settle both the offer price and enterprise budgets for 2027.
