Corporate AI spending stalled in August, with the top 1% of firms in payments company Ramp's dataset cutting AI spend per employee by nearly 10% to $7,205. Adoption across Ramp's 70,000 tracked customers rose just 0.4% month-over-month, with 56% paying for AI products — a near-flat reading that lands awkwardly against the hundreds of billions in infrastructure commitments hyperscalers and frontier labs are making on the assumption that revenue keeps pace.
The spend-per-employee decline at the heaviest users is the more consequential number. These are the firms Wall Street has been counting on to carry the next leg of AI revenue growth, and they pulled back materially in a single month. Ramp economist Ara Kharazian attributes part of the drop to falling token prices rather than falling usage, but the effect on lab revenue is the same either way.
Average token costs fell to $0.68 per million tokens in August, down from a 2026 peak of $1.15 per million in March. OpenAI and Anthropic have each cut prices repeatedly through the year, and the volume growth that would normally offset those cuts has not fully materialized. The result is a squeeze on the exact revenue line that pays back GPU capex.
Key facts
- 01AI spend per employee at the top 1% of Ramp customers fell nearly 10% in August to $7,205.
- 02Average token costs dropped to $0.68 per million, down from a March 2026 peak of $1.15 per million.
- 0356% of Ramp's 70,000 tracked customers paid for AI products in August, up just 0.4% month-over-month.
- 04The US Census Bureau's August 23 survey pegs broader business AI use at 22%, well below Ramp's tech-heavy sample.
- 05Only 6.4% of AI-spending businesses used model-serving or inference platforms in August.
Customers are also trading down. Ramp's data shows companies increasingly routing workloads to older, cheaper models like OpenAI's ChatGPT 5.6-Terra and Anthropic's Sonnet rather than the latest frontier releases. Employees at frontier labs have said much of the training cost for a new model is recouped in the first weeks after launch, which makes a shift toward the previous generation a direct hit to the economics of the release cycle.
Ramp's sample skews toward tech-forward companies, so its adoption figures likely overstate the broader market. The US Census Bureau's ongoing survey, updated on August 23, shows only 22% of businesses reporting AI use — less than half of Ramp's 56% reading. Ramp's numbers still matter as a leading indicator, because its customers tend to adopt earlier and spend more per employee than the median firm.
“We are showing that competition between OpenAI and Anthropic is making AI more accessible, and also driving the price down for companies—and not just driving the price down, but driving spend down at the top 1% of companies that previously the market was expecting to drive much of the growth going forward.”— Ara Kharazian, Ramp economist
The August slowdown has precedent. Ramp's AI index showed little to no adoption growth between August and October of the previous year, then reaccelerated into year-end. Much of the industry takes vacation in August, which affects both procurement decisions and per-seat token consumption. That seasonal read gives labs a plausible explanation for the softness — one that will be tested when September and October data land.
Open-weight models remain a small slice of the picture. Only 6.4% of AI-spending businesses used model-serving or inference platforms in August, a share that is growing steadily but not fast enough to explain the spending declines at top firms. The pressure on lab revenue is coming from price cuts and model downgrades within the closed-model market, not from customers defecting to self-hosted alternatives.
Kharazian frames the dynamic as bifurcated: buyers are winning, sellers are absorbing the compression. That competitive pressure between OpenAI and Anthropic is what is driving prices — and spend — down at the customers labs were counting on.
The counterargument is that August is a single data point in a noisy series, and Ramp's tech-heavy customer base amplifies any pause in enterprise procurement. A rebound in September that mirrors last year's pattern would reset the narrative quickly. Kharazian himself flags the seasonal caveat, and the Census Bureau's slower-moving 22% figure suggests the broader adoption curve has not turned.
For AI-market watchers, the number to track is not headline adoption but revenue per active user at the frontier labs. If token prices keep falling while top-decile spend keeps compressing, the payback math on the current infrastructure buildout gets uncomfortable — and it explains why OpenAI and Anthropic are pushing hard into non-technical co-working tools, consumer subscriptions, and enterprise agents. Volume growth from those categories is the wedge that has to close the gap the price war opened.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
