According to several foreign media reports, Barclays Bank of the United Kingdom recently pointed out in a research report on the AI industry that for every 100 US dollars of revenue received by an AI model company, about 35 to 40 US dollars will go to the three major cloud giants in the form of inferred computing power costs, namely Amazon Web Technology, Microsoft's Azure cloud service platform, and Google Cloud platform. Cloud service providers can obtain an operating profit of about $10 to $20, corresponding to an operating profit margin of up to 35% to 45%. At the same time, the profit margin of AI Lab's paid inference business has risen sharply, from a low double-digit level in 2025 to 50% to 65% or even higher in 2026, and adjusted gross margin increased by 30 to 50 percentage points year-on-year. Barclays analysts believe that actual profit margins may be higher than estimated in the report, but profit margins are expected to gradually decline as competition in cutting-edge models intensifies and the supply of computing power continues to increase.

According to several foreign media reports, Barclays Bank of the United Kingdom recently pointed out in a research report on the AI industry that for every 100 US dollars of revenue received by an AI model company, about 35 to 40 US dollars will go to the three major cloud giants in the form of inferred computing power costs, namely Amazon Web Technology, Microsoft's Azure cloud service platform, and Google Cloud platform. Cloud service providers can obtain an operating profit of about $10 to $20, corresponding to an operating profit margin of up to 35% to 45%. At the same time, the profit margin of AI Lab's paid inference business has risen sharply, from a low double-digit level in 2025 to 50% to 65% or even higher in 2026, and adjusted gross margin increased by 30 to 50 percentage points year-on-year. Barclays analysts believe that actual profit margins may be higher than estimated in the report, but profit margins are expected to gradually decline as competition in cutting-edge models intensifies and the supply of computing power continues to increase.