Ramp data shows AI spending stalling just as Anthropic nears its IPO

Companies are consuming more AI tokens, but they are not spending proportionally more. Ramp reports token spend grew only 1.5% across August and September, after a 42% month-over-month jump in July. October showed a partial re-acceleration — up 8.9% from September 23 to October 23 — but Ramp's Ara Kharazian says the headwinds on spend growth remain strong.

The slowdown is concentrated where it hurts model vendors most: the top 1% of "AI-pilled" firms that had been approaching $10,000 of AI spend per employee. Median firms still sit around $12–$15 per employee per month; the top 10% around $600–$1,000. Rising volumes without rising dollars means price cuts, enterprise discounts, routing to cheaper models, and thriftier customers are capping revenue.

That lands as Anthropic nears an IPO narrative and Bloomberg reports its potential valuation sliding from about $2 trillion in early October toward $1.2–$1.3 trillion, with a high junk credit rating expected. OpenAI has also cut proposed valuation and fundraising goals.

We're treating Ramp's series as the clearest public window into whether the AI spend thesis still compounds. Volumes up, dollars flat is not a demand crash. It is competition and competence showing up on the invoice.

OpenAI halved API prices for GPT-6 Sol and GPT-6 Luna in September; Anthropic answered with cuts on Fable 5.1 and Opus 5.5 that a filing said sliced about $600 million off revenue growth. Discounts meant to lock in enterprise spend are hitting targets instead. Add Meta and NVIDIA cheap or open enterprise offerings, Chinese and U.S. open models, and tools like Jev that route simple yes/no work onto even lighter models, and the market stops looking like a two-player tollbooth.

Kharazian's point about concentration should scare anyone underwriting hypergrowth: almost all AI spend still comes from venture-backed high-growth tech. If that cohort stutters together, cascading effects follow. Notion's Akshay Kothari put the customer-side win in one line — tokens go brrr, costs go ahhh — and that is genuinely good for useful AI adoption. Cheaper, better-routed inference is how coding agents and everyday workflows spread beyond the top 1%.

The IPO risk is the other face of the same coin. Investors betting on unbroken revenue rockets need another coding-agent-scale use case. Personal agents are the hopeful next act; Kharazian is skeptical they are as economically productive. I'm watching for whether October's bounce becomes a new trend or a dead-cat reprice after summer cuts. Useful AI can thrive in a buyer's market. Trillion-dollar narratives that assume spend never learns thrift cannot.

TechCrunch published Rebecca Bellan's piece November 6, 2026, drawing on Ramp economist Ara Kharazian's "Cracks in the AI Thesis" work and contemporaneous Bloomberg reporting on Anthropic and OpenAI valuation resets.