When Uber caps its AI budget, the meter is real
Global AI spending is on track for roughly $2.5 trillion in 2026, up about 44% year over year, and the bills are landing on real budgets. The mechanism is structural: as models get cheaper per token, teams use far more tokens, so total spend keeps climbing anyway.
The tell came when even Uber reportedly burned through its 2026 AI coding budget in four months and capped AI tools at around $1,500 per employee per month. When a company that size has to ration AI, the per-token model is doing exactly what it is built to do, which is tax your own adoption.
The Stavryn take
- Metered cloud AI gets more expensive the more useful it becomes. On hardware you own, the marginal cost of a query is basically electricity.
- A flat plan plus owned hardware turns an open-ended operating expense into a fixed cost you control.
Put your own figures into the cost calculator, see flat pricing, or read why you probably do not need a frontier model.