Back to Business & Economics

Enterprise AI spend flatlines into Anthropic's IPO, Ramp data shows

AI spend flatlined this late summer. A new report from Ramp, the payments and expense management company, found token spend across the businesses in its dataset grew 1.5% over August and September, compared with a 42% month-over-month jump in July. Volumes did not slow. Token usage kept rising. Ara Kharazian, Ramp's lead economist, reads the gap as a sign that price competition and savvier customers are capping the revenue of OpenAI and Anthropic.

The timing is the headline. Ramp published the report in the middle of Anthropic's IPO preparation, and the data runs against the growth story investors are being shown.

What the data shows

Almost all of the AI spend in the dataset comes from venture-backed, high-growth tech companies, Kharazian told TechCrunch. The concentration matters because the heaviest users are the ones stalling. The top 1% of firms, the cohort he calls the most "AI-pilled", came in just under $10,000 per employee per month in August and September, below their summer peaks. A dip in that group is enough to slow total revenue, he said, even while volumes keep rising.

Median firms spend $12 to $15 per employee per month. The top 10% spends $600 to $1,000. Kharazian is not confident the rest of the market will climb to the 1%'s level. A doctor's office or a car dealership does not necessarily need to spend that much on AI, he said.

October saw a reacceleration. Token spend was up 8.9% from September 23 to October 23, but Kharazian calls the headwinds on spend growth strong.

Why volumes climb while spend stalls

Kharazian says this is not a demand problem. It is price. OpenAI halved API prices for GPT-6 Sol and GPT-6 Luna in September. Anthropic cut pricing for Fable 5.1 and Opus 5.5, and the company said in a recent filing that the decision shaved about $600 million off revenue growth. Both companies are also running enterprise discounts meant to lock in spend. In Kharazian's reading, that has turned the market into a buyer's market.

The competitive set is getting cheaper at every layer. Meta and Nvidia are pushing enterprise offerings built on cheap or open source models, and open model companies from China and the US are adding supply. Jev, a lightweight tool that picks between a fixed set of outputs, has become one of the top performers on Ramp. A lot of AI use cases only need a yes or no, Kharazian said. Those run on already cheap models, at an even lower cost.

The second driver is thrift. As teams get more advanced, they set defaults that keep expensive models off long-running tasks, they impose budgets, and they route work to lighter models through routers. Notion co-founder Akshay Kothari posted internal data in late October that matches the pattern: "tokens go brrr, costs go ahhh."

What it means for the IPO

Bloomberg reported on Friday that Anthropic's potential valuation sank from the $2 trillion estimate in early October to $1.2 to $1.3 trillion. Credit agencies are also expected to hand the company a high "junk" rating. OpenAI cut its proposed valuation to $1.25 trillion from $1.4 trillion and trimmed its latest fundraising goal by about $10 billion.

The sector's growth case has always leaned on the next use case. Coding emerged a year and a half ago and became the most commercially viable product the labs have shipped. Investors are hoping personal agents play that role, and OpenAI, Meta, and startups like Instinct have all released their versions. Kharazian is skeptical. Personal agents are "not as economically productive as coding," he said, and the observed trajectory does not support nonstop growth. Based on what he can see today, AI spend has started to slow down.

For a company buying AI, the picture in the data is the opposite of bad. Usage is up, and the price of tokens is down. For the companies selling them, the number to watch is revenue, and for now it is growing slower than the volume it comes from.

Source: techcrunch.com

More news

Back to Business & Economics

Have something in mind?

Start a conversation