Ramp data reveals corporate spending slowdown: price war pressures AI model valuations
Listen to this article
Read by Anchor
Companies consume a unprecedented volume of AI outputs and code, yet accounting books record a trajectory that is entirely opposite to the expectations of endless growth. A recent report from the expense-management platform Ramp shows that the pace of spending on AI technologies has stabilized and slowed sharply in late summer, despite a continuous rise in actual consumption volumes. The data revealed that the growth rate of spending on code did not exceed 1.5 percent over the combined months of August and September, compared with a jump of 42 percent in July alone. This slowdown comes at a highly sensitive moment, as Anthropic prepares to launch its shares for an initial public offering, putting the hypothesis of continuous rocket-like growth before a stark financial test.
Ara Khrazin, chief economist at Ramp and author of the series "Cracks in the AI Thesis," explained that the vast majority of spending is concentrated among venture-backed tech firms. He noted that the top one-percent most AI-using segment saw a slight decline in expenditures during August and September, stabilising below a level of ten thousand dollars per employee per month, which was the peak reached in the summer. According to Khrazin, "the slowdown of this small elite is enough to stall the overall revenue trajectory of the entire sector, even if October data shows an acceleration of 8.9 percent; the average spending of ordinary companies remains modest, between 12 and 15 dollars per employee per month, while the top ten-percent segment spends between 600 and a thousand dollars. Khrazin does not expect doctors’ clinics or car agencies to require expenditures on the scale of software firms.
This shift is mainly due to the market entering a buyer-led pricing phase, after a fierce discount and price-cut war among major developers. OpenAI cut the prices of its GPT-6 Sol and GPT-6 Luna API interfaces by fifty percent in September, and Anthropic responded with parallel reductions on its Fib 5.1 and Obus 5.5 models, a decision the company disclosed in a recent statement that it had cut about 600 million dollars from its revenue growth. This coincided with Meta and Nvidia entering the market with open models available at very low cost, and the emergence of lightweight tools such as Jeff, which allow simple dual-task routing and option settlement at the lowest possible cost, making cheap products even cheaper.
The price war runs parallel to cautious operational maturity in organisations; engineering teams have shifted from random experimentation to consumption optimisation and budget control. Khrazin confirms that companies now impose strict financial limits on employees, use code prompts that shift routine queries from expensive standard models to smaller-sized ones, and prohibit the use of large models for long tasks. This shift was underscored by Akshay Kothari, co-founder of the Notion platform, when he published matching internal data summarising the scene of rising code consumption alongside shrinking cumulative invoices, which directly impacted investment valuations. Potential valuations for Anthropic fell from a trillion dollars to about 1.2 or 1.3 trillion dollars according to Bloomberg, with expectations of a non-investment credit rating, while OpenAI cut its proposed valuation to 1.25 trillion dollars and reduced its funding targets by ten billion dollars.
This pricing curve collapse carries direct implications for organisations and tech teams in the Gulf, Egypt, and the Levant;the technical power balance shifts from global suppliers to the hands of local procurement managers and engineers. This path allows banks and startups in the region to build their services without accepting open invoices or succumbing to monopolistic prices, by restructuring their cloud infrastructure using smart routing tools and low-cost models, and negotiating service contracts at competitive discount rates. Alongside this shift, the engineering efficiency requirements in the regional market change; value no longer lies in invoking the heaviest model matrix, but in engineering systems capable of sorting tasks and leveraging open models to complete work at the lowest possible computational cost.
Developing companies bet on the emergence of new categories of personal agents to revive the spending cycle, after OpenAI, Meta, SpaceX AI, and Instinct released their own models, but current indicators suggest that personal agents will not achieve the high economic productivity that programming agents did. The financial reality proves that the era of extraordinary growth has hit its ceiling, and that AI usage is expanding while the returns of its creators shrink.