Enterprise AI bills out of control: Uber spent its annual budget in just a few months; one company accidentally burned through $500 million in a single month.

Enterprise AI bills out of control: Uber spent its annual budget in just a few months; one company accidentally burned through $500 million in a single month.

```

The AI boom is facing an expensive reality check.

According to Bloomberg, as companies roll out AI tools to all employees on a large scale, uncontrolled bills are starting to surface. Uber Technologies recently exhausted its annual AI budget within a few months and was forced to set limits on employee use of AI programming tools.

Another unnamed company, after introducing Anthropic’s Claude to its employees, burned through $500 million in a single month unexpectedly because there were no usage restrictions. These cases are shaking market confidence in AI investment returns.

Consulting firm Bain pointed out in a report titled "Your AI Budget Is Growing, But the Returns Are Not," that although some companies have achieved 10% to 20% cost reductions through AI, many are making massive additional investments before these savings fully materialize—in essence, they’re making faith-based bets.

Meanwhile, leading AI labs such as OpenAI and Anthropic are preparing for IPOs with valuations approaching $1 trillion. Market enthusiasm for AI remains high, but warning signs about uncontrolled costs are becoming harder to ignore.

Uncontrolled bills: From "unlimited use" to emergency brakes

Cases of runaway corporate AI spending are emerging rapidly, often due to the same oversight—lack of usage restrictions.

Uber is one of the most representative cases. The ride-hailing giant, after employees started using AI programming tools extensively, exhausted its annual AI budget within just a few months, and was compelled to set usage limits. Walmart also tightened employee access to AI tools for generating spreadsheets and presentations, changing from unlimited token supply to a controlled model.

The most extreme case comes from an unnamed company. Reportedly, after launching Anthropic’s Claude to employees without any usage restrictions, it unexpectedly consumed as much as $500 million in token fees in just one month. The financial sector has also voiced numerous complaints about AI costs.

This phenomenon has structural roots. AI service providers have always charged enterprise clients based on usage, but most employees previously accessed AI via flat-rate subscriptions or pilot programs with price caps, keeping token costs off the radar. Now, as companies roll out AI agents, programming tools, and various applications enterprise-wide, the cost from each query accumulates, resulting in bill pressure that becomes unbearable.

ROI debate: Savings unfulfilled, investments doubled

The controversy over AI investment returns is longstanding, and the latest corporate practices add new notes to this debate.

Bain’s report reveals a common dilemma: some companies have indeed achieved 10% to 20% cost reductions via AI, but more have significantly increased AI investments before those savings are fully realized, resulting in a "burn money first, wait later" faith-based bet. Measurable returns on AI spending have yet to form broad consensus at the enterprise level, with each company fighting separately to justify their spending.

Last year, a trend called "tokenmaxxing" was popular among enterprise users—using AI as much as possible to boost productivity and compete on internal leaderboards. But more and more companies are learning painfully: the more tokens used, the higher the AI cost.

Nvidia CEO Jensen Huang this week promised "crazy" returns for investors from AI, and called those who doubt AI’s potential returns "crazy." But critics argue that the imminent wave of AI company IPOs may mark a peak, rather than the start of a new boom.

Optimists: The experimental phase isn’t over, value remains to be unlocked

Despite significant cost pressures, some believe the current predicament is just an inevitable pain of the early stage of AI applications.

Runway AI co-founder Anastasis Germanidis, when asked how AI video-generation tools would shrink Hollywood budgets, offered a different perspective. He said that reducing budgets isn’t the goal: "We’ll see more visual storytelling in the world, creating more content with the same budget." This view represents the core logic of AI supporters: the greatest value of AI is not only in cutting costs but in producing more valuable work.

The core argument from optimists is: companies are still in the experimental stage, many haven’t found the right metrics, and AI’s true value needs time to be validated.

Usage-based pricing: AI is becoming an endless utility bill

The root of cost overruns lies in the deep mismatch between AI pricing models and enterprise budget management.

AI services are essentially metered software—every query, every hallucinatory draft, every bloated programming session incurs a fee. As companies move from limited pilot projects to large-scale company-wide deployment, this metered pricing model is making AI bills more and more like an endless utility bill.

This structural issue is forcing companies to confront a question previously avoided by AI labs: is this technology useful enough to be worth paying for? Bain’s report and the real-life experiences of Uber, Walmart, and others show the answer is far more complex than the scenario Jensen Huang described.

Risk warning and disclaimerThe market has risks, investment needs caution. This article does not constitute personal investment advice and does not consider individual users' specific investment goals, financial situation, or needs. Users should consider whether any opinions, views, or conclusions in this article fit their own situation. Investing based on this is at your own risk. ```