Barclays Sees AI Inference Margins Surging as Cloud Giants Take 40%

Generative AI developers have faced persistent questions over whether revenue can outpace the enormous cost of training and running advanced models. Barclays’ research shifts the focus to how that revenue is divided: AI labs remain heavily dependent on three major cloud providers for computing capacity, making infrastructure charges a central constraint on profitability and a key source of leverage for the hyperscalers.

Barclays estimates that nearly $40 of every $100 earned by AI model companies currently flows to the three cloud giants. Yet gross margins from paid inference are rising sharply, challenging the view that AI is structurally uneconomic. The bank expects global AI-lab revenue to jump substantially by 2028, while greater use of in-house computing and a declining share of training costs could eventually force a renegotiation of cloud providers’ economics.