Big technology companies are hitting a wall with AI spending, as tools meant to boost productivity are instead creating major budget headaches.
Uber, for example, pushed its engineers to use Anthropic’s Claude Code tool at the end of 2025. The company added leaderboards and rankings to encourage use, and adoption soared from 32 percent to 84 percent in just a few months. By April 2026, Uber had already spent its entire yearly AI budget. Engineers were burning through US$500 to US$2,000 worth of tokens every month. The chief technology officer later admitted the original budget plan was no longer realistic, while the chief operating officer said it was time for the engineers to go back to the drawing board, not artificial intelligence.
Microsoft faced the same issue. Its product teams loved Anthropic’s Claude Code so much that token costs shot up fast. The company is now canceling most of those licenses before its fiscal year ends in June 2026, and engineers are being “forced” to use GitHub Copilot instead. Microsoft has invested billions in Anthropic, but the high internal usage forced it to cut back on the very tool it helped fund.
At Nvidia, chief exeuctive Jensen Huang is pushing for more “tokenmaxxing.” He believes engineers should be willing to spend up to half their salary on AI tokens to maximize output. Ironically, one vice president revealed that compute costs for his team have already surpassed what the company pays them in salaries.
The issue comes down to a few things. Pay-per-token pricing works reasonably well for small experiments or projects but quickly becomes unsustainable at scale. Many companies also lack proper usage tracking and spending controls, making it difficult to manage costs effectively. Further, the hardware side, including GPUs and power, is also limited and costly.
Firms are trying different fixes. Some are setting stricter rules, mixing in cheaper open-source models, or running more AI locally on their own machines. In my opinion, it is a bit like the early days of mobile phones. Calls used to cost a lot per minute. Once competition and better technology arrived, prices dropped and everyone started using them more. In the long term, AI might do the same once hardware improves and technology advances.
Allen Au is a tech startup founder, AI architect, and YouTuber
