Apollo has released the results of a study indicating that 10% of clients in the artificial intelligence sector are responsible for nearly all expenses in the industry. The analysis was published on September 24 by Thorsten Slok, chief economist at Apollo Global Management. According to the data, the top 10% of clients account for 99.5% of model-serving expenses and 99% of neocloud expenses. This highlights the extreme concentration of financial investments among a limited number of clients, posing risks for companies involved in AI infrastructure.
Slok relied on data from the Ramp platform, which tracks corporate spending for thousands of companies. According to this data, about 10% of companies spending on software are currently working with GPU vendors, compared to less than 4% two years ago. Participation in model-serving and inference has increased from 2.4% to 8.7% of firms over the same period, while the use of neoclouds has risen from 2.0% to 3.3%. The title of the report reflects the paradox: "AI Adoption is Widespread. AI Spending is Concentrated."
In comparison, in the non-AI SaaS category, the top 10% of companies account for 91.8% of expenses, while in CRM it is 84.2%. With 99.5%, model-serving expenses in AI appear not just concentrated, but almost a two-sided market.
One reason for this concentration, analysts suggest, is the consumption model: most companies prefer cancellable software subscriptions over owning hardware. By viewing AI computing as an operational expense rather than a capital investment, companies can quickly reduce usage without incurring non-recoverable costs.
Related articles
