Ramp’s AI Index shows U.S. businesses bought 37.1% more AI tokens in the week ending October 23 than in the same week a month earlier, while total token spend rose 8.9% and the blended price per million tokens across OpenAI and Anthropic fell 26.8%. Ramp’s lead economist writes that customers hold the stronger negotiating position.
What Ramp measured
The index draws on card and bill-pay transactions from more than 85,000 firms on Ramp, with data through September 2026 and a weekly token-spend series for the latest complete week. Ramp reports that 55.9% of U.S. businesses on its platform paid for AI in September, up 0.6 percentage points from August and 12.1 points from a year earlier. Among those firms, Anthropic reached 43.8% adoption and OpenAI 40.6%, with OpenAI’s gain of 1.1 points outpacing the overall market. Ramp’s lead economist, Ara Kharazian, wrote in his October 14 letter that new-customer growth has slowed and that further market growth will have to come from deeper use inside firms that already pay.
Volume up, price down
The weekly series compares the week ending October 23 with the week ending September 25. Total token volume rose 37.1% and total token spend rose 8.9%. The blended price per million tokens for OpenAI and Anthropic together fell from $0.76 to $0.56, a drop of 26.8%. OpenAI’s blended price went from $0.80 to $0.46, down 42.4%, and Anthropic’s from $0.73 to $0.61, down 15.7%. Anthropic’s share of token spend moved from 51.3% to 60.5% over the same four weeks, while OpenAI’s fell from 44.2% to 35.1%.
Kharazian’s October letter notes that token spend was virtually flat across August and September even as volumes rose, and that spend reached a new peak only at the start of October, 1.5% above the August high. He attributes the slowdown mainly to price declines at the frontier, plus efficiency and quality gains at every model tier.
Where the spend sits
Ramp’s data is concentrated. The October letter repeats Ramp’s earlier finding that 80% of spend on OpenAI and Anthropic comes from 1% of firms. Among firms paying for AI, the top 1% spent a median $9,567.53 per employee per month in September, against $813.43 for the top 10% and $15.70 for the median firm. Kharazian’s letter puts the top 1% figure 3.1% below its July high of $9,873, though Ramp revised its estimates for that group upward after adding vendors, so the comparison carries a methodological caveat that Ramp states itself. AI’s share of business spend excluding payroll slipped from 5.96% to 5.77%.
What it means for the buying committee
For a CIO renewing a model contract, the figures describe the price a peer pays today, and they sit well below where the same tokens cost a month earlier. A commitment priced off a September rate card now sits above the blended market rate. Committees negotiating in the next quarter can ask both labs for a price-protection clause that tracks published list reductions during the term, and can benchmark any quoted per-token rate against Ramp’s $0.46 and $0.61 blended figures with the caveat that blended rates mix model tiers.
Kharazian also writes that the contracts he has heard about rarely assume above-normal AI spend by the customer, and that where they do, the labs offer discounts off sticker price. He frames this as hearsay from businesses that signed or saw such deals, and the index does not publish contract terms. It does give buyers a reason to test how strong their position is before accepting a volume commitment. Where a finance team set an AI budget in dollars, falling prices mean the same dollars buy more usage, and the budget conversation shifts to what the extra usage is for. Salesforce’s per-user pricing for its coworker product, covered in our report on Agentforce Coworker at $50 per user, shows a seat-based alternative to metered tokens. Introductory rates that expire, like those in our coverage of Gemini 4 Argon pricing ending November 30, carry a different risk, because the rate a team budgets on ends on a fixed date.
Limits of the data
Ramp’s sample is its own customer base, which skews toward U.S. firms that use its cards and software, and its own figures show 1% of firms driving 80% of spend on the two labs. The weekly token series covers spend routed through Ramp’s token spend management product, and the latest-week comparison is a single four-week window. Ramp also states that its estimates for the top 1% are volatile because the group is small. Enterprise buyers with negotiated cloud commitments may see different effective prices than card-based spenders.
What to do with it
Pull the last three invoices from each model vendor and compute your own blended price per million tokens, then compare it with Ramp’s $0.56. Where spend controls are already in place, as in Salesforce’s Agent Fabric spend limits, review whether the caps were set against last quarter’s prices. Ask each vendor in writing how a mid-term list price cut would reach your invoice. A committee that learns the market rate from its own invoices will negotiate with better numbers than one that waits for the next index release.
Source: Ramp AI Index
