Major corporations are pulling back on artificial intelligence deployments after discovering that running these tools at scale costs far more than anyone budgeted for, with some companies burning through their entire annual AI allocations in just a few months.
Uber reportedly exhausted its entire 2026 AI budget by April, driven largely by engineers making API calls that cost thousands of dollars per person. Amazon, Walmart, Cisco, and Meta are among the companies slashing or curtailing AI tool usage as costs spiral beyond initial projections.
The inference cost problem
AI inference, the process of running trained models to generate outputs, now accounts for roughly 85% of total enterprise AI budgets, according to the FinOps Foundation.
Numerous firms have reported that annual AI budgets are being depleted within one to three months, with costs doubling or tripling year over year. The per-token pricing model that underpins most commercial AI services means every query, every generated response, every automated workflow chips away at the budget.
Rationing the robots
Internal strategies include limiting high-volume token usage, instituting hard spending caps, and prioritizing AI applications that demonstrate a clear return on investment.
Some executives have described the current dynamic as creating “a monster,” with unsustainable per-token cost increases arriving alongside each new generation of AI models.
What this means for investors
The tightening of AI budgets at some of the world’s largest companies carries real implications for the broader technology investment landscape. For crypto markets specifically, tokens tied to AI infrastructure, including projects that promise cheaper inference through decentralized compute, could see renewed interest if enterprises start shopping for alternatives to expensive centralized providers.
The companies most likely to benefit from this environment are those offering cost-efficient inference, model optimization tools, and AI spending management platforms. Investors watching this space should pay close attention to which AI service providers can survive a world where their biggest customers are suddenly counting every token.