Microsoft cuts AI usage as chip prices soar to $50,000
AI chip prices have surged up to $50,000 per unit, with demand outpacing supply due to escalating AI model training and inference needs, threatening to make AI operations prohibitively expensive. Rising costs are forcing companies like Microsoft and Uber to cut AI usage, while chip shortages and high production costs further strain budgets.
Big tech is pouring billions into AI chips that now cost as much as a Tesla Model 3— and the bill is about to explode. A single Nvidia Blackwell GPU cluster can run $50,000-plus, and a medium-sized data center now needs thousands of them. Non-AI chips aren’t any cheaper; CPU and memory prices have jumped to historic highs. The squeeze is tightening because AI demand keeps rising while chip factories can’t keep up.
The explosion in demand comes from two fronts. First, companies still train bigger models, but the real surge is coming from inference—the actual use of AI once models are live. Goldman Sachs predicts we’ll burn through 120 quadrillion tokens a month by 2030, a 24-fold jump. Each “agentic” AI task—where AI handles multi-step jobs instead of one-off prompts—can chew through hundreds of times more compute than a simple chat. That’s why Microsoft recently yanked most of its employee AI coding licenses: the compute bill outran the salaries it was meant to replace. Uber blew through its entire 2026 AI coding budget in four months. Even if inference prices fall 90 percent, companies won’t see cheaper AI because the token count per task is skyrocketing and providers won’t fully pass savings on.
Chip factories can’t scale fast enough. A new fab can cost tens of billions and take years to build. Chipmakers are cautious—build too much and they risk being stuck with unsold stock if demand cools. Meanwhile, production lines are being shifted toward AI chips because they’re more profitable, leaving shortages for non-AI chips and pushing prices higher. Newer chips are also harder to make: they require extra fabrication steps, pricier materials, and cutting-edge tech that inflates costs further. Add inflation, trade tensions, and geopolitical shocks, and the math gets uglier.
The knock-on effects are already hitting balance sheets. Cloud providers are raising prices. Startups that bet on AI productivity tools are finding their savings vaporize when compute costs spike. The promise of AI-driven efficiency is colliding with the reality of chip economics: if demand keeps outpacing supply, the AI economy may stall before it scales.
