Overview
- Companies that encouraged heavy internal AI use have reversed course and imposed limits after large bills surfaced in mid June, with firms such as Amazon, Meta and Uber rolling out dashboards, caps or stopped incentive programs.
- Executives and CIOs warned that per-employee token spending can scale into the hundreds of millions per year, a risk highlighted by AMD’s Hasmukh Ranjan who estimated exposure up to $900 million for very large workforces.
- Reports of extreme bills—single employees charging more than $150,000 a month and companies spending hundreds of millions in a month—helped trigger internal controls and immediate policy changes.
- Engineers are cutting costs with technical fixes such as lightweight proxy routing, aggressive caching, multi-model routing and self-hosting, with practitioners reporting 70% or greater savings by routing routine work to cheaper models.
- The vendor side is under pressure too as lower-priced Chinese models and disclosed provider losses push industry talks toward more explicit token-based billing and a likely re-pricing of AI services while many firms remain early in AI maturity.