AI Investment Chill? Top Firms Slash Spending Per Employee in August

The adoption rate of artificial intelligence tools by businesses experienced a notable slowdown in August, according to spending data collected from 70,000 companies by the payments firm Ramp. The most recent survey revealed that 56% of Ramp's customers were paying for AI products during August, marking only a marginal 0.4% increase from the preceding month. This isn't an unprecedented trend in Ramp's metrics, as the company's AI index previously showed minimal to no growth in adoption between August and October last year, before seeing a resurgence by the year's end.

Despite previous patterns, the current pause in growth raises concerns given the extraordinarily rapid pace of AI infrastructure development. The massive investments in AI by frontier labs and hyperscalers are predicated on the expectation of substantial future revenue to recoup these costs. While usage has generally grown steeply, particularly among software engineers adopting agentic coding tools, a deceleration in this adoption could directly translate to slower revenue generation for AI providers.

It's important to contextualize Ramp’s figures, as its clientele is predominantly tech-oriented, potentially overstating overall market adoption. For comparison, an ongoing US Census Bureau survey on AI adoption, updated in August, indicates that only 22% of businesses report using AI. Nevertheless, Ramp’s survey remains one of the few direct spending datasets available and is considered a potential leading indicator for the industry.

Several factors might contribute to this August slowdown. One potential explanation is the typical vacation period within the industry during this month. However, Ramp economist Ara Kharazian points to additional warning signs for companies heavily reliant on 'token spend'. A significant observation is a nearly 10% decline in AI spend per employee among the top 1% of firms in his sample, falling to $7,205. This reduction could be influenced by the vacation effect, but also strongly suggests a trend of decreasing token costs. As major AI players like OpenAI and Anthropic have implemented price reductions, average token costs have fallen to $0.68 per million tokens, a considerable drop from a previous peak of $1.15 per million tokens observed in March.

This data implies that AI labs have yet to offset these price cuts with a corresponding increase in volume of usage. Furthermore, these cost incentives are prompting many customers to opt for older, more economical models, such as OpenAI’s ChatGPT 5.6-Terra and Anthropic’s Sonnet, over the more advanced frontier releases. Employees at frontier labs have noted that a significant portion of training costs for new models is typically recouped within the initial weeks of release; thus, slower adoption rates could jeopardize this crucial dynamic.

Despite ongoing discussions about open-weight models posing a threat to frontier labs, only 6.4% of businesses currently spending on AI were utilizing model-serving or inference platforms in August. While this share is steadily growing, it is not expanding rapidly enough to significantly influence the broader business adoption dynamics. Kharazian summarizes the situation by stating, “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.” This trend also helps to explain the strategic focus of AI labs on attracting non-technical users for AI co-working tools.

While this 'blip' in adoption could signal challenges for model-builders and hyperscalers with substantial chip orders, Kharazian offers a nuanced perspective: “it depends on who you are in the market. If your company is using AI, it’s great.” This suggests a bifurcated impact, with users benefiting from increased accessibility and lower costs, even as providers face pressures on their growth models.