AI purchasing by businesses barely changed in August. In a payments dataset covering 70,000 companies, 56% made AI-related purchases, up just 0.4% from the previous month. The headline adoption figure looks stable, but the most important detail is less reassuring: spending per employee among the top 1% of firms fell almost 10%, to $7,205.
A flat adoption rate with two different readings
The 56% figure describes companies in a technology-heavy payment sample, not every employer. A broader government survey updated Aug. 23 found that only 22% of businesses said they used AI. The gap is not necessarily a contradiction. One tracks actual AI invoices from a commercially oriented customer base; the other captures self-reported use across the wider economy.
That distinction matters. The invoice-based data may be useful as an early signal of enterprise demand, while the broader survey is better for understanding how common AI has become. Neither should be treated as a complete portrait of the market.
Why the biggest customers deserve close attention
The top 1% of companies in the spending sample are important because they can represent a disproportionate share of paid usage. A near-10% decline in their per-employee expenditure does not prove they are leaving AI behind. It does show that billings from these accounts slowed sharply during the month.
That is a meaningful warning for model developers and cloud providers. Their facilities, chip purchases, and engineering investments require high levels of utilization to produce attractive returns. If the most intensive customers reduce paid consumption, revenue can grow more slowly even while AI adoption continues to expand.
Cheaper tokens are reshaping the bill
One major factor is price. The average cost of a million tokens dropped to $0.68 in August from a 2026 peak of $1.15 reached in March. Lower prices make AI more accessible, but they also reduce the amount a company pays for a given volume of use.
This creates a difficult balance for model providers. They can win customers by cutting prices and offering capable tools, yet they must generate enough additional usage to compensate for the lower price per token. Customers also have an incentive to route workloads to older, less expensive models when those systems meet their needs.
Alternative delivery models remain niche
Only 6.4% of businesses spending on AI used model-serving or inference platforms in August. The share is rising, but it remains too small to redirect the market on its own. That leaves leading model providers with room to compete, even as buyers become more selective about which systems they use.
The result is a more fragmented AI economy. Companies may combine general-purpose assistants, internal tools, older models, and specialized services rather than concentrating all activity with one provider. For sellers, winning a trial is no longer enough; they must earn a durable place in everyday workflows.
Seasonality could explain part of the slowdown
August is a weak month for many business purchasing cycles because employees are on vacation and approval processes slow down. The spending data showed a similar pause from August through October last year, followed by stronger activity later in the year. That history makes it risky to interpret one monthly dip as a permanent reversal.
Still, seasonality does not erase every concern. If lower token prices are encouraging more experimentation but not enough paid volume, providers could see demand become less profitable. The key question is whether cheaper access produces a larger usage base or simply lowers the cost of the same work.
What executives and AI vendors should do next
-
For business buyers: measure AI by productivity, workflow completion, and return on investment—not by the number of tools purchased.
-
For model providers: focus on practical collaboration features and clear business outcomes, especially for non-technical teams.
-
For investors: watch usage volume, customer concentration, and unit economics alongside adoption rates.
August should be read as a cautionary note, not a verdict. AI is still entering companies, but the path to sustainable revenue is becoming more complicated. Lower prices and wider access are good for users; whether they are enough to support the industry’s ambitious buildout remains the test.
