Microsoft Diverts GPU Capacity to Its Own AI, Azure Waits

Microsoft's stock is down more than 24% from twelve months ago, worse than the rest of the Magnificent 7, and inside the company employees are questioning a record $190 billion capex on AI infrastructure this year. That is the setup for Ashley Stewart's Business Insider piece framing this as the hardest stretch of Satya Nadella's tenure.

The specific tension worth watching is how Microsoft is allocating scarce GPUs. CFO Amy Hood reportedly indicated that if the GPUs that came online during the first half of the fiscal year had been given to Azure instead of Microsoft's own AI products, Azure would have grown more than 40 percent instead of 39. In other words, Copilot and internal AI workloads are being fed first, and paying Azure customers get whatever is left. Executives told the reporter they are 'shopping for capacity everywhere,' including evaluating Amazon and Google, and Microsoft has already leaned on Amazon to help bail out GitHub after a series of outages.

Why this matters if you are not a Microsoft shareholder: the story Microsoft has told for a decade is that Azure is where you go for scalable compute. If the hyperscaler with the biggest capex line in the industry is now renting from its two closest rivals, the mental model of 'just buy more Azure' quietly changes. Nadella conceded supply is expected to remain constrained at least through 2026, with commercial RPO of $627 billion showing demand the company simply cannot serve fast enough. Copilot, the product all of this is meant to justify, still reportedly lags ChatGPT and Claude. GitHub, the crown jewel of the developer stack, is dealing with repeated outages, executive turnover, and pressure from newer tools like Cursor and Anthropic's Claude Code.

The honest caveat is that this is a single Business Insider account sourced through executives and employees who spoke to Stewart, and Microsoft is still growing an AI business at real scale in absolute terms. What the reporting does not give you is a clean quantitative split of how much internal-versus-external GPU allocation has actually shifted, or how large the leases from Amazon and Google really are.

For AWS and Google Cloud, the opportunity is obvious: when a rival is capacity-short and openly renting from you, that is a story to tell your own enterprise buyers. For everyone else, the takeaway is that even the best-funded AI operator in the world is capacity-constrained, and the next year of cloud negotiations will look different because of it.