Several of the world's largest technology companies are pulling back on the unrestricted AI access they granted employees over the past two years, as the computing costs behind that usage climb faster than executives anticipated. According to reporting from the Financial Times and The Information, Meta, Amazon, Walmart, Cisco and Uber have introduced spending caps, usage monitoring, or restrictions designed to curb what one report described as runaway internal AI consumption. The shift comes only months after many of these same companies actively encouraged staff to adopt AI tools as widely as possible.

Meta's case illustrates the scale of the problem. The company told roughly 6,000 employees in an internal memo that it would impose limits on token usage — the unit by which AI providers typically charge for computing work — after costs tied to internal AI consumption reached into the billions of dollars. Meta is also rolling out an internal monitoring platform, described as an "AI Gateway," that gives teams real-time visibility into token consumption and spending, along with automated alerts for unusual usage spikes. The company is simultaneously encouraging employees to shift toward internally built tools rather than third-party AI products, in an apparent effort to bring both costs and oversight in-house.

The dynamic driving these decisions is rooted in how AI services are priced. Unlike traditional software licensing, most AI providers charge based on token consumption — effectively billing for the amount of computational work a model performs on a given task. As employees moved from simple chatbot queries to AI agents capable of carrying out multi-step tasks autonomously, the computing demand per task increased substantially, in some cases by orders of magnitude. Goldman Sachs analysts have forecast that the rise of agentic AI could drive a roughly 24-fold increase in token consumption by 2030, a trajectory that has prompted companies to scrutinize whether the value generated by each AI-assisted task justifies its cost.

Microsoft offers a parallel example outside Meta. According to reporting cited by Tom's Hardware and The Verge, Microsoft has been steering engineers away from Anthropic's Claude Code toward its own Copilot CLI tool, a move attributed in part to the rising cost of third-party AI usage as adoption expanded across the company. Uber, meanwhile, had encouraged AI adoption through internal leaderboards tracking employee usage — an initiative that reportedly backfired as staff inflated their usage with low-value queries simply to climb the rankings, prompting a rethink of that approach.

The cost pressure is not confined to large technology firms. Smaller companies are reporting similar strain as AI agent adoption scales. Legal AI firm Harvey has said its token usage rose more than twelvefold over the past year, a pattern echoed across industries as companies move from pilot programs to broader deployment. Software firm Workato has also reported a sharp jump in AI-related costs after agent adoption expanded among its workforce. Industry observers note that while the cost of individual AI computations has generally declined over time due to model efficiency gains, aggregate spending continues to rise because adoption is growing even faster than per-task costs are falling.

The broader implication is a maturing phase in corporate AI strategy. Early enthusiasm for blanket AI rollouts is giving way to more deliberate governance, including routing simpler tasks to cheaper models, reserving premium AI systems for higher-value work, and requiring departmental accountability for usage. For AI providers, the trend raises questions about revenue durability if major customers are actively working to reduce consumption rather than expand it. For investors and policymakers, the development is a signal that enterprise AI spending, while still growing in absolute terms, is no longer scaling without scrutiny — a dynamic that could shape capital allocation decisions across the technology sector in the months ahead.

Key Takeaways

Meta, Amazon, Walmart, Cisco and Uber have introduced caps or monitoring on employee AI usage after costs rose sharply, according to Financial Times and The Information reporting.
Meta's internal AI costs reached into the billions of dollars, prompting token limits for roughly 6,000 employees and a new internal usage-monitoring platform.
Token-based pricing means costs scale directly with usage, and the shift from chatbots to autonomous AI agents has sharply increased per-task computing demand.
Goldman Sachs projects token consumption from agentic AI could rise roughly 24-fold by 2030, intensifying pressure on enterprise AI budgets.
Microsoft has reportedly shifted engineers toward its own Copilot CLI tool partly to manage rising third-party AI costs, while Uber's usage-leaderboard initiative backfired.

Note: Some figures in early reporting (e.g., total dollar amounts, specific budget caps) vary by source and have not been independently confirmed by the companies named; this article reflects verified reporting from Financial Times, The Information, Fortune, and Tom's Hardware as of June 2026.