AI spending growth nearly stalled in August. Should the big labs be worried?
New data from 70,000 companies shows businesses are still buying AI tools, but the pace of growth has slowed sharply, and prices are falling faster than usage is rising.
Key points
- 56% of businesses tracked by payments firm Ramp paid for AI products in August 2025, up just 0.4% from July.
- AI spending per employee at the heaviest-spending companies fell nearly 10% in August, to $7,205 per worker.
- The average cost of running an AI model dropped to $0.68 per million tokens (the basic unit AI charges for), down from a peak of $1.15 in March 2025.
- A separate US Census Bureau survey, updated on 23 August, found only 22% of American businesses report using AI at all.
August was a quiet month for the AI industry, and not just because everyone was at the beach.
Ramp, a corporate payments company that tracks spending at around 70,000 businesses, found that the share of its customers paying for AI products crept up by only 0.4 percentage points last month, landing at 56%. That is the kind of number that looks fine in isolation but makes investors nervous when the companies selling AI have spent hundreds of billions of dollars building data centres to meet expected demand.
So is this a real problem or just a summer blip?
Probably a bit of both, but the price picture deserves attention regardless.
Ramp economist Ara Kharazian points out that August is when much of the tech industry goes on holiday, which naturally trims usage. This happened before: Ramp's own data showed adoption flatline between August and October last year before picking up again in the final months.
But there is a second force at work. OpenAI and Anthropic, the two dominant AI companies, have been cutting prices aggressively. The average cost of running a query through their models has fallen to $0.68 per million tokens from $1.15 just five months ago. A "token" is roughly three-quarters of a word; companies pay for every token their employees or software send through an AI system.
Cheaper tokens sound like good news, and for businesses using AI they are. The trouble is that the labs have not yet replaced the lost revenue with higher volumes. Fewer dollars are flowing in per query, and August suggests volume is not growing fast enough to make up the difference.
| Metric | Figure | When |
|---|---|---|
| Ramp customers using AI | 56% | August 2025 |
| Month-on-month growth | +0.4 percentage points | July to August 2025 |
| AI spend per employee, top 1% of firms | $7,205 (down ~10%) | August 2025 |
| Average token cost | $0.68 per million | August 2025 |
| Average token cost at peak | $1.15 per million | March 2025 |
| US businesses using AI (Census Bureau) | 22% | 23 August 2025 |
Prices falling also nudge customers toward older, cheaper models. Many businesses are sticking with previous-generation tools rather than upgrading to the newest releases, which is a problem because AI companies reportedly recover a large portion of a new model's training costs in the first weeks after launch.
What does this mean for ordinary workers and businesses?
If your company already uses AI, lower prices mean the same budget buys more capability. That is straightforwardly good.
If you work at an AI company or a tech firm that supplies the computing power behind it, flat adoption numbers and shrinking revenue per user are the early signals worth watching. Kharazian put it plainly: "If your company is using AI, it's great."
Ramp's figures, first flagged by TechCrunch AI, likely skew toward tech-savvy businesses, so 56% adoption is almost certainly higher than the real economy average. The Census Bureau's 22% figure is probably closer to the full picture. Either way, a large majority of businesses have not yet bought in, which means the growth story is not over. It just needs a second act.
