More businesses are buying AI than ever, and spending less per head on it. Both are true. Two things are true in Ramp's AI Index right now, and together they are a pricing lesson. More businesses than ever

…see more

are paying for AI. And the money per head is falling. Average token costs have dropped to around $0.68 per million, down from a 2026 peak near $1.15 in March. Spend per employee at the heaviest AI-using firms fell about 10% in August. Adoption is still growing, but the rate is slowing. And the volume is landing on cheaper, older models rather than the newest frontier releases. Ramp's data covers spend on its own corporate cards, so read it as a large sample rather than the whole economy. Here is the part that should concern anyone selling a premium tier. Buyers did not decide the frontier models were bad. They decided the difference had stopped being felt, for the jobs they actually run. Summarise this. Draft that. Pull the numbers out of these documents. Once a cheaper option clears the bar for the work in front of someone, the better option becomes a preference rather than a purchase. Most of us sell something with a ceiling like that somewhere in the price list. The mistake is defending a premium the buyer can no longer perceive, using benchmarks they never run and quality they cannot see. The commercial move is to find where perception flattens, then either price under it or shift the argument onto something still felt. Risk carried. Evidence supplied. Time returned. Somebody accountable when it goes wrong. Performance a buyer cannot perceive is not performance. It is cost. At what point does your buyer stop being able to feel the difference between your best tier and your middle one? #Pricing #GTM #AIAdoption