Money
What the AI business actually costs and earns.
-
Why AI companies lose money on inference
Gross margins in enterprise AI were 45 per cent in 2025 and are expected to reach 53 per cent in 2026. Inference is the largest single line behind the gap.
-
What an inference provider actually sells
Not tokens. A place in a queue, a promise about where your data goes, and a rate that holds until the provider decides otherwise.
-
Rent or serve: what inference actually costs
Token APIs win on utilisation, not on price. You pay per token when idle; you pay per hour when the GPU is idle. The crossover is how full you keep it.
-
Provider prices compared: one model, five bills
DeepSeek V4.1 Flash is quoted by five companies at five different rates. The gap is what each host thinks its queue, its region and its reliability are worth.
-
What a million tokens costs across the frontier
Input prices run from $0.10 to $10 per million tokens, a hundredfold spread across models that all call themselves frontier. A full window is a bill, not a budget.