What customers are paying for AI.

Revenue has exploded. It still has further to go than it has come. Customer-paid AI revenue reaches roughly $919B by 2030 on the Ledger’s own forward path — about 40 times what it was in 2025. But the infrastructure being built to serve it has to be paid for. Written off over the 5.6 years its owners actually book, and covered out of gross profit rather than revenue, the build-out needs revenue of about $1.5T a year by 2030 — still 1.7× what it is likely to be.

One number in that calculation is a judgement, not a measurement. Nobody discloses a clean gross margin for serving frontier models. The published figures treat training costs, free usage and volume discounts differently at every provider, and the credible range runs from about 30% to 50%. The Ledger holds it at 40% as an editorial position and publishes the swing: at the low end the bar is $2.0T, at the high end $1.2T. Revenue falls short of all of them.

The gap is closing, and quickly: it was 15.1× in 2025. Revenue compounds faster than the bar rises. But the bar rises every year the spending continues, and revenue does not reach a trillion dollars until 2031, which is the part that takes longer than the growth rate alone suggests.