The AI compute boom is adding customers, but the money remains extraordinarily concentrated.
New Ramp data highlighted Thursday by Apollo Chief Economist Torsten Slok shows the top 10% of Ramp customers account for 99.5% of spending on model-serving and inference providers and 99% of neocloud spending, a divide that matters for Nvidia Corp. (NASDAQ:NVDA) and CoreWeave Inc. (NASDAQ:CRWV).
Ramp provides corporate cards and spend-management software to businesses, giving it a window into how companies are allocating money across software and technology services.
More Businesses Are Buying AI Compute
Roughly 10% of software-spending businesses on Ramp now pay a GPU-compute vendor, up from under 4% two years ago.
Most of that growth came from model-serving and inference providers, which run AI models for customers. Adoption rose from 2.4% to 8.7%, while neocloud adoption increased from 2% to 3.3%.
The new users, however, aren’t where most of the money is. The figures reflect businesses using Ramp rather than all U.S. companies.
That concentration comes as the infrastructure bet reaches historic proportions. Economist Stijn van Nieuwerburgh estimates U.S. data-center and related AI investment could total $10.3 trillion from 2025 through 2032.
CoreWeave Shows the Concentration
Ramp lists CoreWeave in its neocloud category, where the Ramp data shows the biggest 10% of buyers generate 99% of spending.
CoreWeave’s own results show a similar pattern. Its three largest customers, which the company does not identify in the filing, accounted for 36%, 26% and 10% of second-quarter revenue, or 72% combined. CoreWeave separately names Microsoft and OpenAI as significant customers and has disclosed major commitments from Meta.
The concentration cuts both ways for investors. About 93% of CoreWeave’s year-over-year revenue increase came from existing customers spending more, showing how quickly a handful of large AI buyers can drive growth. But it also leaves the company heavily exposed if one of those customers slows spending.
For now, demand remains strong. Revenue jumped 112% to $2.58 billion, while CoreWeave ended June with a $104 billion backlog.
What It Means for Nvidia
For Nvidia, the Ramp data offers both validation and a warning. CEO Jensen Huang said last month that AI compute demand had broadened significantly from a year earlier, when "one lab alone was driving the buildout."
Nvidia’s second-quarter results showed Data Center revenue surged 117% year over year to $89 billion.
Ramp’s figures support the idea that more businesses are buying AI compute. But they also show how far those newer buyers have to go before they meaningfully reduce the industry’s dependence on its biggest customers.
Prediction-market traders remain confident in Nvidia. Polymarket currently puts its chance of ending 2026 as the world’s largest company at around 75%, with roughly $7.5 million wagered on the market.
CoreWeave CEO Michael Intrator said last month that customer demand was accelerating as enterprise adoption broadens. Ramp’s figures suggest that broader adoption has yet to produce an equally broad spending base.
Image: Shutterstock
Login to comment