Corporate America Begins Rationing AI Tools as Costs Surge

Large corporations are beginning to pump the brakes on artificial intelligence adoption as the soaring cost of computing tokens causes expenses to skyrocket [1]. After a period of unrestricted experimentation, many enterprises have seen their AI spending double or triple, with some exhausting their entire annual budgets in just three months [2].

Key takeaways

• Companies including Uber, Meta, Microsoft, Salesforce, and DoorDash are implementing new measures to track or limit AI usage [2].
• Token usage has grown significantly, with Google reporting it now processes over 3.2 quadrillion tokens monthly, seven times more than a year ago [1].
• Data from EntelligenceAI suggests that only 18% of spending on advanced AI coding tools currently translates into shipped products for users [1].
• Some employees previously engaged in "tokenmaxxing," a practice of using maximum computing power to appear AI-forward, even as providers shifted to usage-based pricing [2].

From Unrestricted Experimentation to Cost Control

The initial corporate strategy of encouraging widespread AI integration was intended to signal to Wall Street that firms would not be left behind by technological disruption [2]. However, as model providers moved away from "all-you-can-eat" subscriptions toward usage-based pricing, the financial reality of these tools became clear [2]. Executives now warn that "all motion is not progress," with Meta’s Chief Technology Officer Andrew Bosworth noting in an internal memo that token usage alone is not a valid measure of business impact [1].

To manage these costs, companies are adopting various strategies. Uber reported that it had exceeded its annual budget for autonomous, or "agentic," AI by March [1]. Microsoft has limited access to certain programs, such as Anthropic’s Claude Code, for some employees, though a company spokesperson stated this was to standardize internal tools rather than a cost-cutting measure [1]. Meanwhile, Salesforce has introduced a tracking system to ensure token consumption is directly linked to positive business outcomes [2]. Some industry leaders, such as the CEO of coding automator Factory, Matan Grinberg, have pointed out that employees often use expensive, premium-tier models for simple tasks that could be handled by cheaper alternatives [2].

Why it matters

The shift toward rationing AI reflects a broader transition from the "early innings" of adoption to a phase where companies are demanding measurable returns on investment [2]. While some critics view these spending constraints as a sign that the rapid growth of the AI sector may slow, others argue that corporate usage is still climbing faster than initial forecasts [2]. As businesses continue to refine their strategies, the industry remains focused on increasing computing efficiency and steering tasks toward more cost-effective models [1]. Despite the current friction, proponents emphasize that teams are still in the process of discovering where the highest productivity gains lie and how best to quantify them [1].