---
title: "AI Agents Drive Token Demand: Companies Spend Millions Without Knowing If It Will Pay Off"
description: "🚀 AI agents are driving token demand — companies are spending millions without knowing if it will pay off. Microsoft, Uber, and other giants are facing rising AI costs that don’t always deliver expected benefits. The future of AI depends on hardware efficiency and companies’ ability to adapt to new realities. 💸🤖"
date: 2026-05-29T11:56:18.000Z
lang: en
url: https://xab.info/en/posts/ai-agents-drive-token-demand-companies-spend-millions-without-knowing-if-it-will-pay-off
tags: []
publisher: "XAB.info"
---

# AI Agents Drive Token Demand: Companies Spend Millions Without Knowing If It Will Pay Off

![Microchip with 'AI' inscription on a printed circuit board, symbolizing the growing demand for tokens and AI-related expenses](https://xab.info/media/2026/05/29/ii-agenty-razgonyayut-spros-na-tokeny-kompanii-tratyat-millyony-ne-znaya-okupitsya-li-eto/ii-chip-na-ploshche.webp)

Artificial intelligence has ceased to be just a trendy buzzword — it has become a real financial challenge for tech giants. Major companies like Microsoft and Uber are facing an unexpected surge in AI system costs, which is already noticeably impacting their budgets.

Uber exhausted its entire AI budget within just a few months. The company’s CTO, Praveen Neppalli Naga, admitted that token consumption does not always correlate with the delivery of useful features for users. COO Andrew McDonald added that many investments are not yet yielding the expected returns.

Microsoft is also not immune. The company has begun disconnecting developers from the corporate subscription to the developer assistant Anthropic Claude Code and plans to fully transition employees to Copilot by June 30. This move appears not only as a consolidation of tools but also as an attempt to control spending.

### Explosive Growth in Token Consumption

According to Goldman Sachs forecasts, active use of AI agents could increase global token consumption by 24 times in the coming years. Agent-based AI consumes resources thousands of times more than standard chatbots, making its use extremely costly.

In March, Nvidia CEO Jensen Huang stated that if an engineer earning $500,000 annually does not consume $250,000 worth of tokens over the same period, it should raise concerns. Such figures have become the norm: company executives boast about the scale of AI adoption but rarely discuss its actual effectiveness.

Airbnb reports that 60% of new code is generated by AI; at Google, this figure stands at 50%, while at Uber, 80% of software engineers use AI, and 60% of code is created with AI assistance. However, there is no guarantee that such expenditures are justified.

### AI Costs Are Rising Faster Than Labor Compensation

Peter Steinberger, creator of OpenClaw and an OpenAI employee, admitted that his three-person team consumed $1.3 million worth of tokens in a single month. This confirms that AI costs are rising faster than the wages of the workers they are meant to replace. Mass layoffs, which company leadership attempts to justify through AI implementation, are becoming increasingly untenable.

There is hope that next-generation hardware will significantly reduce AI usage costs. Nvidia is preparing the Vera Rubin platform, which promises a tenfold increase in performance per watt compared to existing solutions.

### Infrastructure Challenges

More than 50% of data center projects announced with Blackwell equipment in mind have been canceled or frozen. There is uncertainty about how they will operate if launched this year. Google, Oracle, and Microsoft plan to use this equipment for six years before upgrading, which is difficult to reconcile with the progress promised by accelerator manufacturers.

The reality is that even with reduced token costs, the explosive growth in the number of AI agents cannot be offset by improved hardware efficiency. There is no guarantee that sufficient equipment will ever be available to meet AI demand.

If even major players like Microsoft and Uber are adjusting their AI implementation plans, it becomes increasingly difficult to envision how the rest of the business should proceed. If AI usage is reduced to cut costs, AI developers will never have the funds to compensate for the enormous infrastructure expenses.