A 20% token price cut can wipe out a 1GW AI company's $4B operating profit

AccBalanced · x · 2026-11-16

An illustrative model shows how sensitive AI economics are to token pricing. Assume a fully integrated AI company: $50B of capital, 400,000 GPUs, 16 quadrillion billable tokens/year, $20B revenue, just under $4B operating profit. A 10% realized price cut more than halves profit; a 20% cut wipes it out despite $7B EBITDA. Math: a 20% price cut requires 25% more tokens sold to keep revenue flat (1.25 × 0.80 = 1.00). If throughput rises 25% and all extra output sells, profit recovers to $4B even at -20% price; if volume only grows 10%, profit falls to $1.3B. Throughput gains only help when the extra output sells — and if it requires more GPU-hours or hardware, profit must absorb those costs. This is why operators want hedging and bilateral long-term contracts.

Original post →

More from Venture

Venture channel →