April 27, 2026
The Token Cycle
The same providers who built the cultural argument for expanding AI consumption are now balancing infrastructure against revenue. For product people, adoption creates lock-in before any conscious decision about dependency.
Jensen Huang said he would be "deeply alarmed" if a $500K engineer wasn't consuming $250K in tokens per year. The market took that seriously.
Uber gave Claude Code access to 5,000 engineers in December 2025. Adoption went from 32% to 63% in three months, with 11% of production backend being generated by AI and 1,800 code changes per week without human authorship. In four months, the annual AI budget was exhausted. CTO Praveen Neppalli Naga was direct: back to the drawing board.
The adjustment came from multiple directions. GitHub froze new Copilot Pro and Pro+ subscriptions, with per-subscriber costs exceeding plan values. Anthropic tested removing Claude Code from the Pro plan for new users, reversed course within 48 hours under public pressure, but signaled it would try again.
The cycle became readable: the same providers who built the cultural argument for expanding consumption are now balancing infrastructure against revenue. Behind the scenes, companies like Zapier are already monitoring individual token consumption to separate efficiency from waste.
For those working in product, the implication is concrete. Adoption creates lock-in before any conscious decision about dependency. When the vendor recalibrates mid-cycle, room to maneuver is already scarce.
The market responded exactly to the incentive it received. The bill arrived for everyone at the same time.