After months of encouraging employees to use AI tools, many firms are now limiting access as the cost of compute — measured in tokens — has risen faster than expected and depleted budgets in a matter of months.
What changed and why it matters
Reports in Business Insider and The Economist describe a rapid reversal from the early ‘‘tokenmaxxing’’ phase, when companies urged staff to consume as much AI capacity as possible in hopes of productivity gains. Token pricing, which reflects the compute needed to run AI models, turned out to be substantially more expensive than many had anticipated.
Arvind Jain, CEO of Glean, a company focused on enterprise-level AI, told CNBC that some organizations have seen their AI budgets exhausted in one or two months despite those being planned as annual allocations. That has forced finance chiefs to make difficult trade-offs between spending on tokens or on people.
Examples of limits and reactions
-
Marty Kausas, CEO of AI software company Pylon, said that if usage trends continued, his Anthropic subscription could cost the company as much as $1.4 million per month. Pylon responded by imposing token limits for certain non-technical roles.
-
Uber announced that it spent its full annual AI budget within four months and later imposed a monthly token cap of $1,500 per employee on certain coding platforms.
-
Large organizations such as Walmart and cryptocurrency exchange Coinbase have introduced restrictions or shelved internal contests and incentive programs that encouraged heavy AI use.
-
OpenAI CEO Sam Altman publicly acknowledged that customer bills have become a significant issue, after many organizations previously did not worry about such expenses.
From prestige to problem
Earlier this year token consumption became a prestige metric in tech companies. According to the Ramp AI Index, technology and media firms spent an average of $66 per employee on AI in May, up from $59 in April. Token-based incentives — for example OpenAI’s earlier promise of $15,000 worth of compute to its first AI security fellows — illustrate how companies used compute as rewards or perks.
Nvidia CEO Jensen Huang also highlighted the expectation that high-paid engineers should be heavy AI users, underscoring how quickly token use became embedded in corporate culture.
Workplace tensions and allocation choices
Token limits have produced internal tensions. Developers complain that caps hinder their ability to experiment and grow professionally, potentially leaving them less competitive in the job market. At the same time, many firms prioritize giving most tokens to units that directly generate core value, meaning departments like marketing or finance may face stricter limits.
Rachel Laycock of consulting firm Thoughtworks told The Economist that token allocation typically follows where companies expect the most value, which often disadvantages non-core functions.
How companies are responding
Firms are testing several approaches: hard token caps, request-and-approval workflows for larger allocations, and monitoring systems to curb excessive use without fully cutting access. Some companies prefer usage controls that limit waste while preserving the ability for teams that can extract real value from AI to request more capacity.
Executives are not only focused on cost reduction but are also scrutinizing whether AI spending actually improves productivity. Decisions about who receives tokens are increasingly driven by measured value rather than broad encouragement to consume compute.
Bottom line
After an initial phase of promoting heavy AI usage, many companies are now reining in access as token costs balloon and budgets are strained. Organizations are experimenting with allocation rules and controls to balance encouraging useful AI adoption against the need for financially sustainable AI spending.
