Employee spending on AI falls to $7,205, with price war preceding consumption growth

Corporate spending data recorded a noticeable slowdown in the pace of expansion toward AI tools during August, according to monitoring by the payments company RAMP of transactions from seventy thousand firms. The index showed that the share of companies paying subscriptions for AI products reached 56%, a modest increase of no more than 0.4% compared with the previous month. Although this summer-time stability repeated last year between August and October before rising again toward year-end, the sensitivity of tech markets to any decline has become amplified today, amid massive investments by major cloud-computing firms and advanced model labs in infrastructure and chips, hoping to recoup the shortfall quickly from corporate consumption revenues.

The most indicative index did not stop at the overall adoption rate stabilizing, but appeared deep within the budgets of the highest-spending segment.The average spending per employee on AI tools fell by roughly 10% to $7,205.Among the top 1% of companies in the sample. RAMP’s economic expert Ara Kharazian explains that this decline is partly due to seasonal holidays, but fundamentally reflects a more structural factor: the sharp drop in the cost of processing code and text tokens after the price-war escalation between OpenAI and Anthropic.

The average cost of one million tokens fell to $0.68 in August, compared with the March 2026 peak of $1.15. This accelerating decline means that the major labs have not yet managed to offset the price cut with a commensurate increase in usage volumes. Company behavior shows that they now prefer to rely on older, cheaper models, such as ChatGPT 5.6-tera and the Sonnet model, rather than immediately upgrading to the latest, expensive flagship releases. This shift poses a challenge to the labs’ business model, whose engineers have traditionally recouped most of the model-training costs in the first weeks after launch.

Regarding open-weight models, the figures show that the growing talk of their threat to closed-lab models remains grounded in a slow operational reality; only 6.4% of all AI-spending companies in August used model-hosting platforms or independent inference gateways. This huge gap explains why model developers are currently pushing to attract non-technical users and build collaborative work tools, in order to expand the subscriber base beyond programming teams that have become saturated with code-agent tools.

This shift redraws digital-spending priorities in operation rooms across the Gulf, Egypt and the Levant.If you manage a technology budget at a bank in Riyadh, a payments firm in Cairo or a services platform in Amman, a roughly 40% drop in token costs means API bills should automatically fall without compromising service quality, provided that routing paths are re-examined toward cost-effective models and that paying for flagship versions on routine tasks is avoided. The weak institutional uptake of self-hosting platforms (6.4%) also sends a clear message to regional IT managers: pause the purchase and provisioning of expensive on-premise servers to run models internally, and instead take advantage of the declining prices of ready-made cloud APIs that have entered a commodity phase.

The current equation reveals a stark divergence among market participants; while the slowdown in spending growth sounds an alarm for investors awaiting billions in chip and infrastructure returns, it simultaneously offers a golden opportunity for consuming companies to trim operating costs and benefit from the flare-up of price competition among model developers.