Corporate AI usage continues to climb, but the revenue flowing into top artificial intelligence providers has hit an unexpected plateau. According to new spending data from corporate card and finance platform Ramp, enterprise expenditure on AI models flatlined across late summer even as total token consumption rose. The trend highlights a shifting market dynamic where aggressive price cuts and increasingly cost-conscious enterprise habits are curbing top-line growth for leaders like OpenAI and Anthropic.
The slowdown arrives at a sensitive moment for the sector, coming just as Anthropic prepares for an initial public offering and major model builders face cooling valuation expectations from private investors.
Token usage surges while spend levels off
Data from Ramp shows that enterprise spend on AI tokens grew by just 1.5% combined across August and September. That represents a sharp drop from July, when token spending expanded by 42% over the previous month. While spending showed a modest re-acceleration of 8.9% between late September and late October, overall spend growth remains significantly constrained compared to earlier quarters.
The core cause of the divergence between usage and revenue is efficiency. Businesses are running more automated workloads, but they are paying less per task due to price competition and smarter cost-management practices. Notion co-founder Akshay Kothari recently captured this reality in internal metrics shared online, summarizing the trend as "tokens go brr. Costs go ahhh."
Ara Kharazian, lead economist at Ramp, noted that spend is currently concentrated among a relatively narrow group of high-growth technology companies. He pointed out that almost the entire spend on AI comes from venture-backed firms, warning that if spend begins to stall across the market, it could cause broader cascading effects.
The top 1% hit a spending ceiling
Ramp's analysis highlights a vast gap between ordinary enterprise users and the most aggressive adopters. The median business spends between $12 and $15 per employee each month on AI products, while companies in the top 10% spend between $600 and $1,000 per employee.
At the extreme end of adoption, the top 1% of spending firms saw their monthly expenditure stutter slightly in August and September after approaching peaks of $10,000 per employee. Kharazian observed that a spending slowdown among this elite group is sufficient to flatten overall industry revenue growth, even while processing volume continues to climb.
It remains unclear whether broader industries will ever match the spending intensity of early software adopters. While AI coding tools have seen high demand in software and finance, typical businesses like medical practices or automotive dealerships are unlikely to require multi-thousand-dollar monthly expenditures per worker. As a result, the industry may be reaching the limit of how much revenue can be extracted strictly from top-tier power users.
Price wars and thrifty engineering
A major driver behind the spending plateau is intense price competition among model providers. Major companies have repeatedly cut API rates to retain corporate customers and lock in enterprise accounts.
In September, OpenAI reduced API pricing for its GPT-6 Sol and GPT-6 Luna models by half. Anthropic similarly enacted substantial price cuts for Fable 5.1 and Opus 5.5, a move that the company acknowledged in a recent regulatory filing had reduced its revenue growth by approximately $600 million.
At the same time, alternative options have proliferated. Meta and Nvidia have expanded their enterprise offerings with open-source and low-cost models. Emerging tools such as Jev, a lightweight decision-routing tool, allow developers to handle simple tasks at minimal cost instead of directing every request to premium frontier models.
Enterprise engineering teams have also grown far more sophisticated in controlling costs. Rather than defaulting to expensive, long-running model calls, companies are adopting automated routers, establishing strict employee budgets, and shifting routine workloads to lightweight models. Kharazian explained that as companies become more advanced in using AI, they get better at controlling their spend by using cheaper models and enforcing usage defaults.
Investor expectations face a reality check
The shift from a seller's market to a buyer's market is already impacting corporate valuations across the AI landscape. Bloomberg recently reported that market expectations for Anthropic's valuation have scaled back from $2 trillion in early October to between $1.2 trillion and $1.3 trillion, with reports suggesting credit rating agencies may issue low credit ratings following its public listing.
OpenAI has similarly adjusted its financial projections. The company recently lowered its target valuation to $1.25 trillion, down from $1.4 trillion, while scaling back its latest fundraising goals by roughly $10 billion.
To reignite hypergrowth, AI providers are betting on new product categories such as autonomous personal agents. OpenAI, SpaceX AI, Meta, and startups like Instinct have all introduced agentic tools designed to perform tasks on behalf of users. However, market analysts remain cautious about whether personal agents can match the clear economic productivity and massive compute spend generated by AI coding tools.
What to watch next
The era of effortless hypergrowth for foundation model providers appears to be transitioning into a more disciplined, price-sensitive phase. While surging token volumes prove that enterprise demand for AI capabilities remains robust, falling model costs and smarter deployment strategies mean that higher usage no longer guarantees exponential revenue gains.
As Anthropic moves toward its public debut and rival providers adjust their financial targets, investors will be watching closely to see whether next-generation AI agents can unlock new commercial workflows or if model infrastructure is settling into a lower-margin utility business.
Filed under: AI, TechNews, Startups, Software, EnterpriseAI