Corporate AI Spending Growth Stalled in August, Ramp Data Shows

Payments company Ramp released data showing that the growth rate of enterprise spending on artificial intelligence tools slowed sharply in August 2026. Among the 70,000 companies analyzed, the share of customers paying for AI products rose by just 0.4 percentage points month over month, to 56 percent. That is the weakest result in months for a segment that had grown almost uninterrupted for two years.

The analysis was authored by Ramp economist Ara Kharazian, who has spent months tracking corporate AI spending as part of the recurring Ramp AI Index. He called the August data a warning sign, though he noted that part of the slowdown could be explained by seasonality - August is traditionally a vacation month in the US and Europe, when corporate purchasing activity drops across all categories, not just AI.

Decline Among Top Spenders

The most worrying part of the data does not concern the average company, but the market's leading edge. Among the top 1 percent of businesses by AI spending, cost per employee fell by nearly 10 percent in August, to $7,205. These are precisely the companies - large tech organizations and corporations deploying AI at scale - that have so far driven most of the global revenue for labs like OpenAI and Anthropic.

The spending decline among market leaders matters because these were the very customers expected to drive further growth in token revenue, the foundation of the multibillion-dollar investment plans behind computing infrastructure buildouts. If even the most sophisticated customers are starting to cut spending, it becomes harder to justify the pace of building new data centers financed by debt and investor capital.

Cheaper Tokens, Same Volume

The second piece of the puzzle is pricing. The average cost of AI tokens fell to $0.68 per million, down from a peak of $1.15 in March 2026 - the result of a price war between OpenAI and Anthropic, which have repeatedly cut rates for access to their models in recent months. The problem is that these price cuts have not translated into a proportional increase in usage volume that could offset the labs' lost per-query revenue.

Ramp's data also show that companies are increasingly choosing older, cheaper models over the latest flagship versions. Instead of the priciest newly launched models, customers are opting for cheaper variants such as OpenAI's ChatGPT 5.6-Terra or Anthropic's Sonnet, rather than paying more for the newest, most advanced releases. That is typical behavior for a market where customers no longer treat the latest model as a necessity and instead start optimizing for cost.

Competition between OpenAI and Anthropic is making AI more accessible and pushing prices down, but it isn't being offset by volume growth - Ara Kharazian, Ramp economist

Adoption Still Narrow

Ramp's data need to be read in the broader context of AI adoption across the economy, which is still far from saturated. According to US Census Bureau data cited in the analysis, only 22 percent of all businesses in the economy use AI tools - showing that even with the slowdown among Ramp's customer segment, the theoretical potential for further market expansion remains large. At the same time, only 6.4 percent of companies investing in AI use platforms built on open-weight models, and that share is growing too slowly to meaningfully affect global spending trends.

The vendor market also remains highly concentrated. According to July data, Anthropic strengthened its lead among US companies paying for AI, with a 43.5 percent share, while OpenAI grew by just 0.23 percentage points, to 39.7 percent. That means that despite the slowdown in overall spending, the battle for corporate customers between the two largest labs continues, with the stakes being who captures a bigger slice of an already slower-growing pie.

What This Means for the Industry

For investors and tech market analysts, Ramp's August data is the first hard signal that the growth rate of corporate AI spending may be leveling off rather than continuing to rise exponentially, as assumed by the forecasts underpinning the wave of investment in data centers, chips, and energy infrastructure. Economists cited in the analysis caution that one month of data, skewed further by the summer slowdown, is not enough to call this a lasting trend - September and October readings will show whether August was a temporary lull or the start of a more durable slowdown.

For Polish companies deploying AI, the practical takeaway is this: price pressure between OpenAI and Anthropic means cheaper access to models, but it also signals that providers will look for revenue elsewhere, for example in new service layers, enterprise subscriptions, or agentic features billed differently than per token. Companies planning their 2027 AI budgets should keep in mind that token prices may keep falling, but that does not necessarily mean lower total deployment costs if providers offset it with fees for additional features.