Uber Technologies IncUber exhausted its 2026 AI coding budget in four months and set per-employee caps, indicating cost pressures.
Rising consumption costs could slow enterprise adoption of generative and agentic artificial intelligence, Bernstein said in an analyst report. The phenomenon, dubbed "token shock," occurs when AI consumption grows faster than per-token prices decline, pressuring budgets as software providers shift to usage-based pricing. Agentic AI intensifies the strain, with one EY-reviewed project showing a roughly 30-fold cost increase per interaction when moving from direct large language model prompts to agentic workflows. Companies are already imposing spending controls: Uber exhausted its 2026 AI coding budget in four months and set a $1,500 per-employee cap per agentic coding tool, while a major French insurer reduced use of Anthropic's Claude and Walmart capped access to an internal AI agent. Many clients have postponed, scaled back, or cancelled AI projects, and one provider estimated fewer than one-quarter of pilots yielded positive returns. Inference expenses may account for about 80% of an AI model's lifetime costs, shifting focus from training to long-term operational spending.
Uber Technologies IncUber exhausted its 2026 AI coding budget in four months and set per-employee caps, indicating cost pressures.
Walmart Inc.Walmart capped access to an internal AI agent, reflecting cost-control measures.
HubSpot IncRising token costs slow enterprise AI adoption, reducing demand for AI software platforms like HubSpot's.
A major French insurer reduced use of Anthropic's Claude, directly impacting demand for its AI services.