AI spend flatlined at +1.5% as Anthropic's $2T IPO shrinks

AI token spend grew 1.5% across August and September, after growing 42% in July. Usage kept rising — the money stopped following it.

That is the number in Ramp's new report, published today. Ramp tracks corporate card spend, so this is a proxy for the market, not the market itself — but it is the first hard dataset showing the AI revenue engine stalling while the usage engine accelerates. And it landed the same week Bloomberg reported Anthropic's IPO valuation talk falling from about $2 trillion in early October to $1.2–$1.3 trillion.

What Ramp actually measured

The core finding, from Ramp lead economist Ara Kharazian: token volumes rose through late summer, but token spend grew just 1.5% across August and September combined, versus 42% month-over-month in July. My arithmetic on those two numbers: the growth rate fell 28x in a single quarter.

The stall is concentrated at the top. The most "AI-pilled" 1% of firms — the ones spending near $10,000 per employee per month at their summer peaks — came in just below that level in August and September. Median firms spend $12–$15 per employee per month; the top 10% spend $600–$1,000. So the top 1% spends roughly 740x the median firm, and when that thin cohort stops growing, total spend stops growing with it. Kharazian's own warning: "Almost the entire spend on AI is from venture-backed, high-growth tech companies. That's fine as long as AI spend continues to rise, but if it starts to slow down... they will all start to slow down simultaneously."

The cause is price, not demand

This is not a demand story. Volumes are up. It is a price story, and the price war is doing exactly what price wars do:

• In September, OpenAI cut API prices in half for GPT-6 Sol and GPT-6 Luna.
• Anthropic cut pricing for Fable 5.1 and Opus 5.5 — and said in a recent filing that the cuts shaved about $600 million off revenue growth. Against H1 2026 revenue of $16.3 billion, that is 3.7% of half-year revenue, given up voluntarily to hold customers.
• Both companies are also offering enterprise discounts to lock in spend, which Kharazian says are "hitting their revenue targets" — the discounts cost more than they retain.

Customers are getting thrifty on top of it: routing work to cheaper models, setting budgets, and using defaults that keep expensive models away from long-running tasks. One new entrant Kharazian flagged is Jev, a lightweight tool for choosing between a few predetermined outputs — "for a lot of AI use cases, you just need a simple Yes/No." The cheap stuff is getting cheaper still.

Kharazian's Uber analogy is worth quoting because it fails in an interesting way. Subsidizing use to lock in customers worked for ride-hailing because the market consolidated. In AI it doesn't work, he argues, because "there's just too much competition" — Meta and Nvidia pushing cheap enterprise models, open models from China and the US, and now Jev. A buyer’s market, with no consolidation in sight.

The IPO math

Bloomberg reported Friday that prospective investors are discussing a $1.2–$1.3 trillion valuation for Anthropic, down from the roughly $2 trillion discussed in early October and $1.4–$1.5 trillion in mid-October. At the midpoint of $1.25 trillion, that is a 37.5% cut in about a month. The Wall Street Journal reported one investor thinks it could land closer to $1 trillion. No final decision has been made, no price range has been set, and Anthropic declined to comment — these are investor discussions, not terms.

OpenAI is getting the same treatment: its proposed valuation was lowered to $1.25 trillion from $1.4 trillion, a 10.7% cut, and its fundraising target was cut by about $10 billion. The context is a market rout, not just AI skepticism: the Nasdaq 100 fell 6.4% on Thursday, erasing about $2.3 trillion in a single day, the worst session on record in dollar terms.

Anthropic's own amended S-1, filed November 2, adds a second pressure point. Q3 revenue was $17.0 billion, but compute and infrastructure spending was about $9.86 billion — 58% of revenue, up from 48% the prior quarter — including $5.0 billion of short-term, high-priced commitments. Total committed spending stands at about $895.4 billion. And the filing leaves October's run-rate revenue literally blank: "our Run-Rate Revenue was $ as of October 31, 2026." The Journal reports it topped $80 billion, but the company chose not to print the number.

The counterpoint, honestly stated

October re-accelerated: token spend rose 8.9% from September 23 to October 23, nearly 6x the August–September pace. Kharazian says the headwinds are strong, but the stall is not a straight line down. And the bull case for hypergrowth has always been the next use case: coding agents emerged just 18 months ago as the product that made Anthropic and OpenAI trillion-dollar companies, and investors hope personal agents are next. Kharazian is skeptical — personal agents are "not as economically productive as coding."

Here is my read. The IPO thesis was never "people use a lot of AI." It was "people use a lot of AI and pay more for it every month." Ramp's data breaks the second half. When your best customers spend less per token every month because you cut prices to keep them, and the customers who could replace them spend $12 a month, the growth story needs a new chapter — a new product category, not a new pricing tier. Until that chapter exists, a $2 trillion valuation was pricing in a hypergrowth era that, by Ramp's numbers, may already be over.

Sources: TechCrunch, "Ramp data shows AI spending stalling just as Anthropic nears its IPO" (2026-11-06); Bloomberg via multiple outlets on the $1.2–$1.3T valuation discussions (2026-11-06); Anthropic S-1/A (2026-11-02) for Q3 revenue, compute share, and commitments; OpenAI's valuation and fundraising cuts as reported 2026-10-31. All ratios and percentages above are my own arithmetic on the reported figures. Ramp's data covers its own customer base — a proxy for corporate AI spend, not a census of the market.