Ramp data: AI spending growth stalls from 42% to 1.5%
Ramp's AI Index shows business AI token spending growth fell from 42% in July to 1.5% over August–September 2026, just as Anthropic nears its IPO.
Ramp, the corporate spending-card company, tracks how American businesses pay for AI with its own card and bill-pay transaction data, published monthly as the Ramp AI Index. According to figures TechCrunch pulled from it on November 6, month-over-month growth in business AI token spending collapsed from 42% in July to a combined 1.5% across August and September. The twist: token volumes kept growing at double-digit rates over the same stretch, so what stalled is the bill, not the usage. The timing lands awkwardly for Anthropic, which is preparing its IPO and did more than anyone to cause the price cuts behind the slowdown. “Almost the entire spend on AI is from venture-backed, high-growth tech companies,” Ramp’s lead economist Ara Kharazian told TechCrunch — which is why a slowdown at the top shows up immediately in the aggregate.
Key points
- The Ramp AI Index, built from transactions at 85,000+ US businesses, puts per-employee monthly AI spend at the top 1% of firms at $9,567.53 in September — 3.1% below the July peak of $9,873. The median firm spent $15.70.
- Token spending grew 42% month-over-month in July and just 1.5% combined across August and September; in the week ending October 23, spend rebounded 8.9% while token volume rose 37.1%.
- The blended price per million tokens across OpenAI and Anthropic fell to $0.56, down 26.8% in about a month — and 1% of firms account for 80% of what those two labs collect.
Background
Ramp sits on a transaction firehose most AI analysts lack: what tens of thousands of American companies actually pay model providers each month. Its AI Index has been running all year, and the October 14 installment of its “Cracks in the AI Thesis” series — the title is Ramp’s own economists talking their own book down — was the third straight month flagging decelerating dollar growth. We have covered adjacent pieces of this puzzle before: Anthropic’s leaked prospectus showed $4.6 billion of 2025 revenue with two customers at about 12% each, and UK card spending data had Anthropic overtaking OpenAI in SME share within a quarter. The price war started in September, when OpenAI halved API prices for GPT-6 Sol and GPT-6 Luna and Anthropic answered with significant cuts on Fable 5.1 and Opus 5.5 — about $600 million of revenue growth foregone, per its filing as reported by TechCrunch.
The facts
The stratified numbers from the Ramp AI Index (September 2026, updated through the week ending October 23):
| Firm tier | AI spend per employee, month | vs August |
|---|---|---|
| Median firm | $15.70 | +$0.96 |
| Top 10% of firms | $813.43 | +$72.96 |
| Top 1% of firms | $9,567.53 | +$999.36, still 3.1% under the July peak |
Other checkable figures: paid AI adoption reached 55.9% of Ramp businesses in September (up 0.62 points), with 43.8% paying for Anthropic and 40.6% for OpenAI. In the week ending October 23, Anthropic held 60.5% of token spend share (up from 51.3% two weeks earlier) against OpenAI’s 35.1% (down from 44.2%). AI’s share of total business spend excluding payroll slipped to 5.77%, down 0.19 points. The IPO-file side of the ledger agrees: per SaaStr’s reading of Anthropic’s amended S-1, Q3 revenue of $16.96 billion beat Q2’s $11.56 billion, but the dollar add shrank from $6.84 billion to $5.4 billion — growth intact, acceleration gone. Bloomberg, via TechCrunch, reports Anthropic’s IPO valuation expectations slid from roughly $2 trillion in early October to $1.2–1.3 trillion, while the Financial Times reports OpenAI’s expectations were cut to $1.25 trillion with fundraising goals reduced by about $10 billion.
What others say
Kharazian’s framing is that this is “not a demand problem”: the biggest customers are capping spend with model-default policies, per-employee budgets, and routing work to cheaper lightweight models. “For a lot of AI use cases, you just need a simple Yes/No,” he said — referring to Jev, a lightweight tool that picks among a few predetermined outputs and now ranks among the top token consumers on Ramp. He is skeptical personal agents will pick up the slack, calling them “not as economically productive as coding.” Notion co-founder Akshay Kothari posted “tokens go brrr. costs go ahhh.” on X on October 22, saying his own internal data matched Ramp’s. Jason Lemkin at SaaStr reads the Index against the S-1 and concludes that “a handful of accounts carry the revenue line” — once the top 1% plateaus, it shows up in the filing within a quarter. Unite.AI’s same-day report leans the other way, stressing that volume is outrunning spend and reading the 8.9%-versus-37.1% gap as adoption still accelerating. The readings are not contradictory: usage up, dollars flat, and the disagreement is about which one you bet on.
Our take
Our read: this is an S-curve top stacked with a price war, not demand evaporating. Ramp’s own framework says as much — over half of US businesses already pay for AI, so new-customer growth is thinning out while unit prices fell nearly 30% in a month; the growth rate was always going to cave. For buyers, the negotiating window is open: enterprise contracts rarely assume elevated spend, labs are discounting off sticker price, and switching or pushing back costs less than it did six months ago. For IPO watchers, the number to watch is the October 31 run-rate Anthropic left blank in its filing — SaaStr’s yardstick is roughly $75 billion: well above it and August–September was noise, hovering near $68–70 billion and the top-of-market plateau is confirmed. The honest catch is the data’s edge: this is one fintech’s customer base, the token subset is officially “not representative,” the top-1% sample is small, volatile, and was revised upward across all periods. Reading 1.5% as proof the AI bubble has burst goes further than the data does.