Major tech companies are sharply cutting their internal AI spending after discovering that aggressive experimentation with artificial intelligence costs far more than expected. Uber spent its entire 2026 AI budget in just four months, while OpenAI's CEO recently called AI costs a "huge issue" for customers. This spending crackdown contradicts a bold prediction made by Nvidia CEO Jensen Huang just a few months ago.

What Changed Since Jensen Huang's Token Prediction?

In March 2026, Huang made a striking statement about AI spending. He said he would be "deeply alarmed" if a software engineer earning $500,000 annually wasn't spending $250,000 per year on tokens, the units of data that power AI interactions. Tokens represent the computational cost of sending prompts to AI systems and receiving responses. Huang's comment reflected the prevailing tech industry mindset at the time: more AI use equals more productivity and innovation.

But that philosophy has collided with financial reality. Companies encouraged employees to compete in "tokenmaxxing" contests, burning through as many tokens as possible as a measure of productivity. Meta and Amazon employees even used internal leaderboards to track who consumed the most tokens. The assumption was that intensive AI use would drive breakthroughs and competitive advantage.

Now, facing actual bills, companies are reversing course. Uber recently imposed a $1,500 monthly cap per employee per coding tool. This represents a dramatic shift from the "go all-in" mentality that dominated just months earlier.

Why Are Token Costs Spiraling Out of Control?

The problem lies in how companies are using AI. While the cost per token has generally fallen, companies are deploying AI for complex tasks that require exponentially more tokens than simple queries. These include "agentic" applications like autonomous coding and chain-of-thought reasoning, where AI systems must perform multiple internal steps to solve a problem.

"Those agents internally pose many, many queries in the process of getting you to your answer. I don't think there's any precise statistic here, but sometimes it takes 500 times as much, a thousand times as many tokens," said Gary Marcus, a cognitive scientist and AI researcher.

Gary Marcus, Cognitive Scientist and AI Researcher

In practical terms, asking an AI to write code or reason through a complex problem can consume 500 to 1,000 times more tokens than a simple question like "what should I make for dinner?" This explains why companies that embraced aggressive internal AI use suddenly found themselves hemorrhaging budget.

How Are Companies Rethinking Their AI Strategy?

The new approach is called "tokenomics": understanding token costs and using AI strategically rather than experimentally. Instead of encouraging unlimited AI use, companies are now evaluating where AI actually delivers value and at what cost.

• Micro-Experiments: Focus on small, targeted tests to identify where AI genuinely improves speed or quality compared to human work, rather than deploying it broadly across all functions.
• Cost-Benefit Analysis: Evaluate whether AI solutions perform tasks faster than humans and calculate the actual cost per task to determine if the investment makes financial sense.
• Function-Specific Approaches: Recognize that different departments have different AI needs; how human resources uses AI differs significantly from how engineering or legal teams should deploy it.

"Companies have moved away from very kind of naive experimentation. Instead, they are dealing with the realities of integrating AI in a broad-based way into their organizations," explained Nestor Maslej, CEO of a consulting firm advising companies on AI adoption and former editor-in-chief of Stanford University's AI Index Report.

Nestor Maslej, CEO of AI Consulting Firm and Former Editor-in-Chief of Stanford AI Index Report

This shift represents a maturation of how enterprises view AI. The honeymoon phase of "experiment with everything" is ending, replaced by disciplined evaluation of return on investment.

What Does This Mean for AI Companies' Bottom Line?

The spending pullback creates a dilemma for AI providers. If companies stop "tokenmaxxing" and reduce their token consumption, AI companies may not generate the revenue they projected. Yet they also face intense competition and pressure to keep prices competitive to maintain market share.

OpenAI is reportedly considering lowering token costs to attract users away from competitors like Anthropic. Meanwhile, Chinese startup DeepSeek recently announced a 75 percent discount on its primary model. Anthropic has shifted its pricing model to include both flat fees and usage-based charges, while Microsoft-owned GitHub Copilot changed its pricing structure in early June to tie costs directly to token consumption.

"If tokenmaxing is not sustained, then these AI companies are probably not going to make the revenue that they thought," noted Gary Marcus.

Gary Marcus, Cognitive Scientist and AI Researcher

The AI sector is entering a critical phase where the technology's value must be proven through tangible business outcomes, not just consumption metrics. Companies are no longer willing to treat token spending as a proxy for innovation. Instead, they're demanding evidence that AI investments actually improve efficiency, reduce costs, or create new revenue streams.

Huang's March prediction about $250,000 annual token spending per engineer may have been technically accurate for companies pursuing aggressive AI integration. But it underestimated how quickly those same companies would question whether such spending actually delivers results. As the AI industry matures, the conversation is shifting from "how much can we spend?" to "how much should we spend to see real returns?"