Why it matters
The median-versus-average gap is the whole story: a heavy-user tail drags the average ($140,842) to roughly 60x the median ($2,246). That skew is why blanket AI budgets miss and why per-user guardrails beat top-line caps.
Tokenmaxxing read
Two levers hide in Ramp's data. Output tokens cost 3–5x input, so verbose agents bleed money on the expensive side of the ledger—trim output, not just calls. And with a $0.07-to-$1.42 spread per million, model routing alone can cut a bill 20x before you touch usage.
Source takeaway
Ramp's headline: from January 2025 to April 2026, token usage jumped 1,001% while dollars spent rose 497%—falling per-token prices kept spend growth well below the usage explosion, but the bills still nearly sextupled.
