Major technology companies are discovering that throwing money at artificial intelligence does not automatically translate into better products or measurable business value. Microsoft and Uber have both recently restructured their AI spending, signaling a broader crisis in the industry: the cost of running advanced AI systems is rising faster than the tangible benefits they deliver.

What's Driving the Sudden AI Cost Crisis?

The problem began with explosive growth in agentic AI, a category of AI systems that can autonomously perform complex tasks by breaking them into smaller steps. These agents consume tokens, the basic units of AI billing, at a staggering rate. A single agentic AI agent can use more than 1,000 times the tokens of a basic AI chatbot. This means that as companies deploy more autonomous AI systems, their bills skyrocket without a clear correlation to improved products or services.

Uber's Chief Technology Officer recently revealed that the company had exhausted its entire 2026 AI budget in just a few months, prompting a public reckoning about whether the spending made sense. Andrew Macdonald, Uber's Operations Chief, stated that token usage simply did not correlate with useful consumer features. "There was no link between higher token usage and a proportional increase in consumer features with real benefits for their customers," Macdonald explained. Although more code was being shipped, he noted it "was very hard to draw a line" between increased token spending and actual improvements in software.

Andrew Macdonald, Uber's Operations Chief

Microsoft faced a similar moment when it began revoking developers' access to Claude Code, an AI programming assistant made by Anthropic, earlier this month. The company plans to move those developers to its own internal Copilot CLI tool by June 30. While Microsoft framed this as consolidating teams onto tools it is developing, the timing suggests cost-cutting as well, occurring right at the end of Microsoft's fiscal year.

How Are Companies Responding to Runaway AI Costs?

• Consolidating Tools: Microsoft is moving developers away from third-party AI assistants and toward internally built alternatives to reduce token-based billing expenses.
• Questioning ROI: Uber and other companies are demanding clearer evidence that AI spending translates into measurable consumer benefits before continuing large-scale deployments.
• Restructuring Budgets: Both Microsoft and Uber are actively restructuring their AI usage patterns to continue leveraging the technology at scale without exhausting annual budgets in months.

The irony is sharp: many executives have publicly bragged about their AI adoption rates as if high usage alone signals success. Airbnb's Chief Executive Officer told investors that 60 percent of the company's code was now AI-generated. Chime claimed it was shipping 84 percent AI-generated code earlier this year, and Google reported that 50 percent of its code is AI-generated, though always reviewed by human engineers. Yet Uber, which reported that over 80 percent of its software engineers were using agentic AI and over 60 percent of code was AI-generated, found that this level of adoption was not worth the cost.

The scale of token spending can be staggering. Peter Steinberger, creator of OpenClaw and now an OpenAI employee, recently announced that his team of three people spent over $1.3 million in tokens in a single month running agentic AI tools. This reinforces a troubling reality: the cost of AI is now rising above the cost of the workers it is supposed to replace.

What Does Goldman Sachs Say About the Future?

Goldman Sachs estimates that agentic AI could increase token demand by over 24 times in just the next few years. This projection suggests the crisis will worsen unless something changes dramatically. The investment bank argues that massive efficiency gains from next-generation inference chips, such as Nvidia's Vera Rubin platform, will make AI use so much cheaper that companies can continue investing without destroying their budgets.

Nvidia's Vera Rubin platform, which the company will discuss at Computex and officially launch later this year, promises significant improvements. The platform reportedly offers as much as 10 times the performance per watt compared to its predecessors, meaning it can deliver the same computing power while consuming far less electricity. Such efficiency gains would give companies that deploy these new chips an enormous competitive advantage over those still running older Blackwell or Hopper hardware.

However, there is a catch. Over 50 percent of data center projects announced with Blackwell hardware in mind have been cancelled or delayed. Additionally, in late 2025, Google, Oracle, and Microsoft all adjusted their plans to run hardware for six years before replacing it, a timeline that seems incompatible with annual hardware leaps and ambitious AI expansion plans.

Why the Hardware Efficiency Argument May Not Solve the Problem

The fundamental issue is timing. Even if next-generation chips deliver the promised efficiency gains, those improvements are years away from reaching the scale needed to offset the explosion in agentic AI demand happening right now. Token costs are falling in some cases, but the sheer volume of tokens consumed by autonomous AI agents is growing so rapidly that hardware efficiency alone cannot catch up.

This creates a vicious cycle. If major companies like Microsoft and Uber cannot figure out how to afford AI at scale without destroying their budgets, it becomes increasingly difficult to imagine how smaller organizations will manage. And if usage drops because costs become prohibitive, AI companies will struggle to find the short-term profits needed to justify the enormous infrastructure spending they are still trying to defend.

The disconnect between AI hype and AI economics is becoming impossible to ignore. Companies invested heavily in AI infrastructure and staffing based on promises of transformative productivity gains. Instead, they are discovering that the technology, as currently deployed, consumes resources at a rate that outpaces the value it delivers. Until that equation changes, expect more companies to follow Uber and Microsoft in pulling back, restructuring, and demanding proof that AI spending actually improves their bottom line.