Ramp says AI spending nearly stalled for two months, right before Anthropic's IPO
Token spend barely moved in August and September, Ramp's economist says, and the slack is showing up right where Anthropic needs growth most.
Ramp’s monthly AI spending index is usually a quiet little chart nobody outside finance Twitter looks at. This one landed differently. According to a TechCrunch report published November 6, token spend across Ramp’s customer base grew just 1.5% combined over August and September. In July alone, it had grown 42% month over month, and Anthropic is in the middle of pushing its own IPO story on the back of growth numbers like that one.
That’s not a small deceleration. That’s a near stop, right as Anthropic is trying to convince public-market investors it still belongs in the hypergrowth bucket.
| Cohort | Per-employee spend, summer peak | What changed, Aug-Sep |
|---|---|---|
| Top 1% (“AI-pilled”) | About $10,000/month | Stuttered, came in just under $10,000 |
| Top 10% | $600-$1,000/month | Rose, not enough to offset the top 1% |
| Median firm | $12-$15/month | Rose, same story as top 10% |
The number that matters isn’t total spend
Ramp’s lead economist, Ara Kharazian, told TechCrunch the slowdown shows up most clearly among what Ramp calls the “AI-pilled” cohort: the top 1% of firms by AI spend, mostly venture-backed tech companies spending close to $10,000 per employee per month at their summer peak. In August and September, that cohort’s spend stuttered, coming up just under that $10,000 mark.
My read: that detail is the whole story. Median firms spend $12 to $15 per employee a month. The top 10% spend $600 to $1,000. Neither group is anywhere near big enough to move the aggregate number. Kharazian put it plainly: a slowdown in the top 1% is “sufficient to slow down revenue entirely, despite the fact that we’re seeing rising volumes.”
Token volumes were still rising the whole time. Ramp’s own October index, titled “Cracks in the AI Thesis, Part 3,” says per-employee spend climbed to a new peak in October, about 1.5% above the prior August peak, after two months of near-flat growth. TechCrunch adds that token spend rose 8.9% between September 23 and October 23, so the stall wasn’t permanent. It just happened to land in the two months before Anthropic needs its growth story to look its sharpest.
Why volume up, spend flat isn’t a contradiction
This is the part I think gets lost in the headline. Kharazian’s explanation isn’t “demand dried up.” It’s that AI got cheaper to use at almost exactly the same time companies got better at not overspending on it.
In September, OpenAI cut API prices in half for its GPT-6 Sol and GPT-6 Luna models. Anthropic has run its own round of price competition this year too. Kharazian told TechCrunch that’s doing real work: “OpenAI and Anthropic price cuts have…just made it cheaper to use AI.” Cheaper tokens mean more volume without proportional spend growth, which is good for users and bad for anyone modeling AI-company revenue off of usage trends.
The second factor is more mundane and, I’d argue, more durable: companies are getting better at AI procurement. Kharazian described customers setting per-employee budgets, routing routine tasks to cheaper light models, and restricting access to the most expensive models for long-running jobs. None of that requires anyone to use AI less. It just means the bill stops scaling in a straight line with usage, which is exactly what a company spending $10,000 a head would want once the novelty wears off.
Notion co-founder Akshay Kothari posted internal numbers that line up with Ramp’s read, summing it up as “tokens go brr. Costs go ahhh.” It’s a glib line, but it matches what Ramp’s data is actually showing: more usage, flatter bills.
The timing problem for Anthropic
None of this would be a blog-worthy news story if Anthropic weren’t mid-IPO process. The company filed its prospectus in October, disclosing a loss that later widened to $80.7 billion across six months once a full accounting line was included. Investors pricing that filing need a growth rate that justifies the loss, and “the top 1% of AI spenders plateaued for two months, right when you needed them not to” is not the kind of data point a prospectus wants floating around.
It’s already showing up in the number that counts most. Bloomberg reported Friday that Anthropic’s expected IPO valuation has fallen from roughly $2 trillion in early October to $1.2 to $1.3 trillion now. Ramp’s data doesn’t prove causation here, and nobody involved is claiming it does. But a market that’s already nervous about AI spending durability doesn’t need proof, it just needs a reason to mark the number down, and this is a plausible one.
I don’t think this is a sign the AI spending boom is over. Volumes are still rising, October reaccelerated, and a two-month plateau in one cohort’s per-employee number is thin evidence for a thesis that big. What it does show is that the growth is lumpier and more price-sensitive than the “AI spend only goes up” story assumed a few months ago, and that lumpiness landed at the single worst moment for Anthropic’s roadshow.
Ramp's data covers its own customer base, not the whole US economy, and skews toward venture-backed tech companies that adopted AI tools early. It's a real, large sample, but it isn't the full market.
Related: Anthropic lost $42 billion last year. About $34 billion of that never left the building.
Sources
Note on sources: I read the TechCrunch article and Ramp’s October AI Index in full. I did not open the underlying interactive token-spend chart Ramp links from its site, so any numbers from that chart are taken as TechCrunch reported them.
- Ramp data shows AI spending stalling just as Anthropic nears its IPO (TechCrunch, November 6, 2026)
- October 2026 Ramp AI Index: Cracks in the AI Thesis, Part 3 (Ramp)
- Anthropic’s IPO valuation expectations pressured after AI rout (Bloomberg, November 6, 2026)