Companies Curb AI Use as Inference Eats Up 85% of Budgets

Companies initially gave employees broad access to generative AI to boost productivity. But the shift from chatbots to multi-step AI agents has rapidly increased token consumption and inference costs. AnalyticsWeek’s 2026 inference economics report estimates that inference now accounts for 85% of total corporate AI budgets, making the link between spending and actual output a central issue in financial governance.

The Financial Times reported on June 18 that Amazon, Walmart, Cisco, Uber and Meta have introduced quotas, discouraged nonessential use or shifted workloads to cheaper models. Uber exhausted its full-year AI budget by the end of April 2026 and subsequently capped monthly spending at $1,500 per employee for each tool. Workato said its spending jumped sevenfold on the first day after Anthropic switched to token-based pricing in May.