Ramp sells corporate cards to more than 85,000 American businesses, and it reads every charge that crosses them. That ledger turned up something odd this fall. Businesses pushed far more AI tokens through their bills than a month before, while the dollars behind those tokens barely moved3. Usage, in short, is outrunning spending.
A million tokens from OpenAI or Anthropic cost $1.15 in March. By late October the price was less than half that, and still falling1. Ara Kharazian, Ramp's lead economist, tracks the swing every month. His numbers for the heaviest spenders tell a second story, one that moves on its own slower clock.
Tokens Cost Less Than Half What They Did in March
Ramp built the AI Index from its own card and bill-pay records3. The March peak held through spring. "Our latest index tracking the effective price per million tokens shows that prices have declined 41% to $0.68 as of this week, down from the 2026 peak of $1.15 in March," Kharazian wrote in early September5.
The price kept sliding after that, with a bump along the way. By the week ending Sept. 25 the blended rate stood at $0.76, higher than that early-September reading. A month later the price was $0.56, down 27 percent in four weeks3. OpenAI's own rate fell further than Anthropic's over that stretch.
A Million AI Tokens Fall From $1.15 to 56 Cents in Seven Months
Blended effective price per million tokens, OpenAI and Anthropic, U.S. businesses on Ramp's card and bill-pay data, March to October 2026
The line is the blended price a business paid for a million tokens; the September point sits above the October one, a bump along a mostly falling path.
The first two points are single-week readings from Ramp's own posts, a narrower measure than a monthly average, so the September rise is a real weekly swing inside a falling trend.
Sources: Ramp [1]; Ramp [3]; Ramp [5]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Period | Value |
|---|---|
| March | $1.15 |
| Sept. 4 | $0.68 |
| Sept. 25 | $0.76 |
| Oct. 23 | $0.56 |
Spending trailed usage by a wide margin, and that gap is the story Kharazian keeps returning to. "That volume may not be large enough to counteract the effective decline in prices," he wrote in September, naming the risk plainly5. Businesses, so far, keep buying more anyway. Ramp's own share of paying customers climbed again in September.
Who Pays, and How Much a Head
Ramp sorts its paying customers into three tiers. A median firm, a top 10 percent, and a top 1 percent that spends far past either. In September the gap between the smallest and largest of those tiers ran past 600 to 13.
The Top 1 Percent of AI Spenders Pay 609 Times the Median Firm's Bill
AI spend per employee per month, by tier, U.S. businesses paying for AI on Ramp, September 2026
Each bar is one spending tier; the top 1 percent's bar runs far past the other two, because the heaviest buyers run far more AI work through far more of their staff.
The three tiers span wildly different scales, from under $16 to nearly $9,600, so the bars compare dollar gaps between very different group sizes, separate from how well each tier uses AI.
Source: Ramp [3]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value |
|---|---|
| Top 1 percent of firms | $9,567.53 |
| Top 10 percent of firms | $813.43 |
| Median firm | $15.70 |
That top tier carries most of the dollars at stake. Torsten Slok is chief economist at Apollo Global Management. He studied the same spending data and found a market dominated by its heaviest users. Among the top 1 percent, most spenders trace back to venture money, a far higher share than among typical AI buyers9.
Adoption splits by provider, though spend splits further still. Anthropic and OpenAI run a close race for how many businesses sign up with each. Measured in dollars actually spent on tokens, Anthropic pulls well ahead of OpenAI, and its customers simply spend harder.
Ara Kharazian@arakharazianNEW: AI spend stutters in the latest Ramp AI Index.
Per employee spend rose in Sept for the top 1% of business spenders but remains below highs in July.
The top 1% of firms spent $9,568 per employee per month, down 3.1% from its July high of $9,873.
The median firm spent $16 pic.twitter.com/tt3c3eqRO3
The Case for Buying More, Paying Less
Kharazian reads the price collapse as a sign of real competition. He says it helps buyers more than it hurts sellers. "We are showing that competition between OpenAI and Anthropic is making AI more accessible, and also driving the price down for companies — and not just driving the price down, but driving spend down at the top 1% of companies that previously the market was expecting to drive much of the growth going forward," he told TechCrunch6.
That framing cuts two ways, and Kharazian knows it. "It depends on who you are in the market. If your company is using AI, it's great," he told the same outlet, drawing the line between a buyer and a seller of tokens6. Cheaper AI widens who can afford it. September's adoption figures back him up, spreading growth across firms big and small alike.
The Case Against Betting on the Top 1 Percent
Slok's worry sits one level up, past the single index Ramp publishes. "As long as agentic AI and token spending are confined to a small part of the economy, the macroeconomic effects of AI will be very limited," he said, after laying out how few firms carry the spending7. His data names the top 1 percent, rather than broad adoption, as the group AI labs need to keep growing.
That group's own numbers moved under reporters' feet this fall. On Sept. 9, Kharazian wrote that the top 1 percent "spent $7.2K per employee per month, down 10% from a July peak ($8K)"48. By Oct. 14 his own account of July had changed: "The top 1% of firms spent $9,568 per employee per month, down 3.1% from its July high of $9,873"12. Ramp's live index now puts August well above the figure first reported that month3.
The direction stayed the same each time: spend at the top fell through the summer, though its size moved with every retelling. A reader who took September's headline figure at face value saw a sharper drop than Ramp's own later data supports. By September, the top tier's per-employee spend had closed to within 3 percent of its July high. That counts as a recovery, rather than a further slide.
What to Do
Budget against the tier your company sits in, rather than the headline average. A business running AI lightly looks far different from the firms driving most of the dollars Slok counted7. That median-size buyer should expect a bill closer to Ramp's small end than to the top 1 percent's3.
Watch your own vendor's price chart before renewing a contract sized to March's rates. A 51 percent drop in the blended token price means a volume commitment signed in the spring is probably overpriced today13.
Treat any single month's spend-per-employee figure as provisional. Ramp's own numbers for July and August moved after the fact145. A board update built on one month's reading deserves a second look once the next month's data lands.
By the numbers
- A million tokens from OpenAI or Anthropic cost $1.15 in March. By the week ending Oct. 23 it cost $0.56, a 51 percent drop13.
- Token volume on Ramp's platform rose 37.1 percent in a week. Token spending itself rose just 8.9 percent in that time3.
- AI adoption among Ramp's business customers reached 55.9 percent in September. That was up from 55.2 percent in August3.
- Per-employee AI spend, September: the median firm paid $15.70; the top 10 percent, $813.43; and the top 1 percent, $9,567.533.
- Anthropic took 60.5 percent of AI token dollars in the week ending Oct. 23. OpenAI took 35.1 percent3.
- Torsten Slok found the top 10 percent of firms hold 99.5 percent of model-serving and inference spending7.
- Venture-backed companies made up 80.4 percent of the top 1 percent of AI spenders. The median spender's venture-backed share was 13.1 percent9.
What to watch
Ramp's November release will show whether the top 1 percent's spend cleared its July high. It will also show whether the token price kept falling past its October reading. Anthropic's widening lead in token dollars sits against a tighter race in plain adoption. Both are worth a second look once Anthropic's own revenue reporting catches up to Ramp's weekly reads.
Sources
- Ara Kharazian, "October 2026 Ramp AI Index: Cracks in the AI thesis, part 3," Ramp, 2026-10-14, https://ramp.com/data/ai-index-oct-2026
- Ara Kharazian, "Post on X," X, 2026-10-14, https://x.com/arakharazian/status/2110356024368476203
- Ramp, "Ramp AI Index," Ramp, 2026-10-23, https://ramp.com/data/ai-index
- Ara Kharazian, "Post on X," X, 2026-09-09, https://x.com/arakharazian/status/2097706961584140645
- Ara Kharazian, "September 2026 Ramp AI Index: Cracks in the AI thesis, part 2," Ramp, 2026-09-09, https://ramp.com/data/ai-index-sept-2026
- Tim Fernholz, "AI spend per employee slumped at top firms in August — summer doldrums or a warning sign?," TechCrunch, 2026-09-09, https://techcrunch.com/2026/09/09/ai-spend-per-employee-slumped-at-top-firms-in-august-summer-doldrums-or-a-warning-sign/
- Stocktwits, "Apollo's Torsten Slok Warns AI Has Delivered 'Zero Financial Payoff' Outside Tech," TradingView, 2026-10-24, https://www.tradingview.com/news/stocktwits:398c53c88094b:0-apollo-s-torsten-slok-warns-ai-has-delivered-zero-financial-payoff-outside-tech-ai-story-will-be-decided-in-the-s-p-493-not-the-mag-7/
- Cryptobriefing, "Ramp reports 10% drop in AI spending among top businesses," Cryptobriefing, 2026-09-09, https://cryptobriefing.com/ramp-ai-spending-drop-top-businesses/
- Diego Almada Lopez, "Apollo reports top 10% of AI customers account for nearly all spending," Cryptobriefing, 2026-09-27, https://cryptobriefing.com/apollo-ai-spending-concentration-risk/
