A new analysis from Apollo Global Management, released on September 24, highlights a significant concentration of spending in the AI infrastructure sector. Chief Economist Torsten Slok reports that the top 10% of AI customers account for 99.5% of model-serving expenditure and 99% of neocloud spending.
The findings are based on data from the Ramp platform, which monitors corporate purchasing across thousands of businesses. While AI adoption is growing, the spending remains highly centralized compared to other enterprise software categories, where the top 10% of firms typically account for 91.8% of SaaS spending and 84.2% of CRM costs.
Over the past two years, the share of businesses on Ramp engaging GPU vendors rose from under 4% to approximately 10%. Participation in model-serving and inference increased from 2.4% to 8.7%, while neocloud utilization grew from 2.0% to 3.3%.
Most companies currently treat AI compute as an operating expense through cancellable subscriptions rather than investing in owned hardware. This approach allows firms to scale down quickly without sunk costs, but it also creates risks for infrastructure providers who may become overly dependent on a small group of clients.
