AI use is soaring but AI spending stalled — right as Anthropic eyes an IPO

Companies are leaning on AI harder than ever. They're just not paying more for it. Fresh data from Ramp shows token spend crept up only 1.5% across August and September — a hard stop compared with July, when it jumped 42% month over month. Usage keeps climbing. Revenue doesn't.

The pain is sharpest at the very top. Ramp's numbers show the most AI-obsessed 1% of firms — the cohort that had been running near $10,000 in spend per employee at their summer peak — slipped just below that mark in August and September. Median companies shell out $12 to $15 per employee each month, and the top 10% spend $600 to $1,000. Growth in those tiers isn't nearly enough to plug the gap. Ramp lead economist Ara Kharazian told TechCrunch that almost all AI spending comes from venture-backed, high-growth tech companies, so if they slow at the same time, they all slow together.

Blame a price war. OpenAI halved API prices for GPT-6 Sol and GPT-6 Luna in September, and Anthropic answered by cutting rates on Fable 5.1 and Opus 5.5 — a move the company said in a recent filing shaved roughly $600 million off its revenue growth. Enterprise discounts meant to lock in customers are doing the same damage. Kharazian's read: it's a buyer's market, and the era of hypergrowth may be over.

There's cheap competition everywhere too. Meta and Nvidia are pushing cut-rate enterprise models and open source options, and open model shops in the US and China keep piling in. Kharazian also flagged Jev, a lightweight tool for picking between a few preset outputs, as a growing presence among Ramp's top performers. Translation, per Kharazian: plenty of AI jobs only need a simple yes or no, so you take already-cheap models, run them through Jev, and the cheap stuff gets cheaper still. Buyers are getting sharp — setting budgets, steering long tasks away from pricey models, and routing work to lighter ones. Notion co-founder Akshay Kothari posted internal data that lined up with Ramp's findings, summing it up as: tokens go brr, costs go ahhh.

All of this lands right as Anthropic heads toward an IPO. Bloomberg reported Friday that Anthropic's potential valuation has sunk to $1.2 trillion to $1.3 trillion, down from the $2 trillion floated in early October, and that credit agencies are expected to hand it a high junk rating. OpenAI has reportedly trimmed its own proposed valuation to $1.25 trillion from $1.4 trillion and cut its latest fundraising target by about $10 billion.

October offered a flicker of hope — token spend rose 8.9% from September 23 to October 23. Kharazian still sees stiff headwinds, and he's skeptical that personal agents, the buzzy follow-up to coding agents from OpenAI, Meta and startups like Instinct, will carry the load. His reasoning: they're not as economically productive as coding. Investors are betting another revolution arrives. The Ramp data says don't count on it yet.