Corporate AI Spending Stalls in US Summer, Exposing Revenue Ceiling Ahead of Anthropic's IPO

Corporate use of artificial intelligence has never been higher. The amount companies pay for it, however, has stopped keeping pace. Data from corporate expense management platform Ramp shows that token spending grew only 1.5% across August and September in the United States — a sharp brake compared with the 42% expansion recorded in July over the previous month.

The figure arrives at a delicate moment: Anthropic is preparing to go public, and the thesis that AI revenues would grow at exponential speed is now being questioned by concrete numbers.

The cheap-token paradox

The slowdown does not stem from a lack of demand. Processed volumes remain on the rise. What changed was the price. The discount war between OpenAI and Anthropic made access to models far cheaper, compressing revenue per unit consumed.

In September, OpenAI halved API prices for GPT-6 Sol and GPT-6 Luna. Anthropic responded with significant reductions on Fable 5.1 and Opus 5.5 — a decision that, according to the company itself in a recent document, subtracted about $600 million from its revenue growth. Both companies also began offering corporate discounts to try to lock in customer spending, but the practical effect is pressure on revenue targets.

The concentration that worries

Ramp chief economist Ara Kharazian's diagnosis points to a structural risk. Almost all AI spending comes from high-growth, venture-backed tech companies. As long as this group expands its outlays, the market sustains the narrative. If it brakes, the chain reaction is simultaneous.

The signs already appear in the most intensive layer. The top 1% of the most "AI-pilled" companies — those that have embedded AI most deeply into operations — slipped to slightly below $10,000 per employee per month, short of the US summer peaks. For Kharazian, this isolated retraction is enough to slow total revenue, even with volumes rising.

The distribution is deeply unequal. The median company spends between $12 and $15 per employee per month. The top 10% disburse between $600 and $1,000. The gulf between these levels and the $10,000 ceiling of the most advanced 1% raises doubts about the generalizability of use cases.

Where demand really exists

AI coding agents concentrate the most robust demand, in sectors such as software and finance. Kharazian notes that it is unclear whether medical practices or car dealerships will ever spend the equivalent of the top of the curve.

The series of analyses the economist publishes under the title "Cracks in the AI Thesis" argues that the era of hypergrowth may be behind us. October data, measured between September 23 and October 23, recorded a reacceleration of 8.9% in token spending — but, according to him, headwinds against spending growth remain strong.

The buyers' market has consolidated. For model providers seeking to justify billion-dollar public valuations, the challenge is no longer winning users but extracting revenue from a base that has already learned to negotiate.