Uber exhausted its AI budget, and companies' computing expenses are already outpacing salaries

Artificial intelligence is rapidly transitioning from experimental spending to one of the heaviest lines in corporate budgets. At some companies, costs for models, tokens, and computing are now competing not with software licenses, but with payroll. A telling signal came from Uber: the company's CTO reported that the annual AI budget for 2026 is already exhausted, with high token costs being the primary reason.

What was recently perceived as a flexible tool to accelerate team productivity is now becoming a full-fledged operating expense line that cannot be ignored. The problem is that AI economics don't scale the way traditional corporate software does. If a subscription to a typical service is usually predictable, expenses for generative models grow with every request, every new user, and every additional use case.

The more a company embeds AI in search, support, analytics, code, internal assistants, and customer-facing products, the higher the token bill, cloud GPU costs, data storage, monitoring, and security expenses become. At the pilot stage, these sums may seem manageable, but once in production, the budget begins to shift dramatically, especially when thousands of employees or customers start using the service. Meanwhile, a significant portion of costs is not hidden in a single contract but spread across multiple line items: APIs, cloud services, vector databases, logging, data protection, fine-tuning, integration, and maintenance.

This is precisely why AI in the corporate environment often turns out to be more expensive than it appears from the first presentation or invoice. Against this backdrop, the forecast for global IT spending seems logical: in 2026 it should reach $6.31 trillion, which is 13.

5% more than 2025 levels. A substantial portion of this growth comes from AI infrastructure, software, and cloud services. Money goes not only toward access to models but also to the entire ecosystem around them: server capacity, integrations, observability tools, response quality control mechanisms, and data protection measures.

For business, this means that AI implementation can no longer be viewed as a local initiative of a single team — it is about a systemic review of the IT budget. If companies used to debate whether to hire another analyst, developer, or manager, now a different question increasingly arises: does a particular AI scenario deliver better ROI than a living specialist or standard automation? This also changes the management approach.

Companies can no longer simply allow employees to use neural networks or connect an API in a couple of products. They need financial limits, cost control per scenario, choosing between expensive and more compact models, caching repeated requests, and rigorous assessment of where AI truly creates value. Increasingly, companies are routing requests by complexity level, limiting context length, using smaller models for routine tasks, and employing local solutions where confidentiality and predictable costs matter.

In many cases, it's cheaper to keep some processes with people or automate them using traditional methods than to blindly pay for each generation. Uber's example is important precisely because of this: even a large technology business with a strong engineering culture is not facing the problem of AI access, but the problem of its profitability. The next stage for the market is not a race for maximum AI features, but a shift toward disciplined usage.

The winners will not be companies that connect more models, but those who learn to calculate the cost of an answer, correlate it with revenue or time savings, and promptly turn off expensive scenarios without clear returns. AI finally stops being a trendy overlay on business and becomes infrastructure. And infrastructure, as Uber's example demonstrates, requires not only ambition but also very strict budget control.