Microsoft to triple Azure capacity to 38GW by 2032 after turning away customers

The disclosure is a rare admission of scarcity from a hyperscaler: Microsoft says demand outran supply to the point where it turned customers away and routed traffic to rivals. The fix is a capacity expansion from roughly 12GW today to more than 38GW by 2032 — a threefold-plus buildout that signals both the scale of AI compute demand and how far behind supply has fallen.

The move mirrors the week's cross-vendor theme of compute scarcity — OpenAI paused Pro sign-ups the same days, and Google committed €13B to Finnish data centers with a 22-year nuclear supply deal. Microsoft's decision to break out Azure revenue in dollar figures starting fiscal 2027 is a transparency shift investors have demanded, letting the market size the AI-cloud business directly.

The strategic subtext is competitive positioning: turning customers away is a growth risk, and the 38GW target is Microsoft signaling to enterprises that the capacity crunch is temporary. Whether it can execute a buildout of that magnitude — power, land, chips, and cooling — on schedule is the open question, especially as every hyperscaler chases the same GPUs and grid interconnects simultaneously.