Cloud giants take $35 of every $100 in AI lab revenue

For every $100 in AI model revenue, $35 to $40 flows to AWS, Azure, and GCP as inference fees, yielding $10 to $20 in operating profit for cloud providers, Barclays research shows. The findings, published Aug. 28, map the profit distribution of an industry where AI labs' paid inference margins have surged from low double digits in 2025 to 50 to 65 percent or higher in 2026.

"Actual margins may be even higher than our estimates," Ross Sandler, analyst at Barclays, said. He expects them to gradually decline as frontier competition intensifies and compute supply continues to expand.

Enterprise customers and agentic workflows have become must-buy products, driving adjusted gross margins up 30 to 50 percentage points year over year. Barclays constructed two hypothetical frontier lab models to break down the profit differences: Lab A, deriving 70 percent of revenue from APIs and 30 percent from subscriptions, shows adjusted gross margins near 55 percent. Lab B, the inverse at 80 percent subscription and 20 percent API, sits at roughly 38 percent — a 17-percentage-point gap driven by API's inherently higher inference margins, training cost allocation, and partner revenue-sharing.

The industry's revenue trajectory is steep. Barclays estimates total AI lab revenue will grow from $7 billion in 2024 to $137 billion in 2026 and $690 billion by 2028, with training costs falling from 96 percent of the total to 30 percent. Cloud providers' AI revenue as a share of AI lab revenue is also declining — from 153 percent in 2024 to 90 percent in 2026 and an estimated 73 percent in 2028.

Margin Analysis by Product Line

Subscription products such as Claude Code and Codex carry estimated inference margins around 70 percent — the lowest of the three product lines — because AI labs subsidize token costs to retain users. Direct API, the earliest business model, is the most profitable at over 80 percent inference margins, driven by model token efficiency gains, nominal API price increases, and infrastructure-level optimizations including quantization and speculative decoding. Indirect API offers the same end-user experience but bills through cloud providers, widening the comparability gap in financial reporting as its revenue share rises.

Barclays draws an analogy to Uber and Lyft: the same core business produces dramatically different reported figures depending on accounting treatment. Lab A recognizes indirect API revenue on a gross basis, while Lab B uses net-basis accounting or does not recognize strategic partner-operated indirect API revenue at all. As AI labs begin disclosing GAAP financial statements, investors must unpack these recognition differences when comparing companies.

Cloud Providers' Profit Structure

For every $100 in AI lab revenue, Lab A corresponds to $35 in cloud provider revenue, contributing approximately $11.80 in profit after infrastructure costs at a 34 percent operating margin. Lab B, due to strategic partner revenue sharing (20 percent of revenue, subject to a cumulative cap), generates more for cloud providers — $41 in revenue and $19.10 in profit at a 47 percent operating margin. Barclays emphasizes the sharing mechanism inflates apparent margins; stripping it out, actual profit per token is identical. The arrangement is expected to phase out to zero after 2028.

Agentic subscription products create additional value for cloud providers. These stateful runtime products often require access to upper-layer software resources such as databases, making each dollar of revenue more valuable. In some scenarios, revenue-sharing arrangements also exist between cloud providers and AI labs.

AWS, Azure, and GCP are expected to maintain their respective shares of AI lab compute spending over the next two years. But starting in 2028, secured AI infrastructure projects will come online and become the preferred choice for AI labs, potentially causing the three major cloud providers to gradually lose market share in both training and inference.

For investors, the report's implications cut both ways. Cloud providers' AI inference businesses are highly profitable today, supporting the bull case for Amazon, Microsoft, and Alphabet. But the 2028 inflection point — when self-built infrastructure erodes cloud share — suggests the current margin profile may not persist. Meanwhile, AI labs' improving profitability as inference revenue outpaces training costs supports valuations for private frontier labs and public AI names, though the accounting recognition gaps Barclays flags mean reported figures warrant scrutiny.

This article is for informational purposes only and does not constitute investment advice.