Microsoft reportedly plans to triple data center capacity to 38 gigawatts by 2032

Microsoft is planning to expand its global data center footprint to more than 38 gigawatts by 2032, up from roughly 12 gigawatts today, according to people familiar with the plan cited by Bloomberg. Microsoft has not confirmed the figure in an official release or regulatory filing, and the report is based on unnamed sources rather than a company announcement — but the scale and specificity of the reporting, including a named target year and current baseline, points to an internal planning figure rather than speculation.

The driver behind the expansion is capacity scarcity, not just AI ambition. According to the report, server shortages have already forced Microsoft to turn away some cloud and AI customers — a direct admission, if accurate, that demand for Azure compute has outpaced the company's ability to build and provision infrastructure fast enough.

The scale in context

A footprint of 38 gigawatts would exceed New York state's peak electricity consumption — a useful reference point for grasping just how large a single company's compute infrastructure has become. Of that total, AI-dedicated capacity is expected to represent roughly one-third, growing from about 2 gigawatts today. The build-out figure covers company-owned and leased facilities but excludes compute rented from neocloud providers like CoreWeave, meaning Microsoft's actual controllable AI capacity — including third-party capacity it leases — would be larger still.

Why "turning away customers" is the more important detail

Data center expansion announcements are common in the current AI infrastructure buildout cycle — what makes this one notable is the reported admission that Microsoft has been capacity-constrained enough to decline business. For a company the size of Microsoft, turning away paying cloud customers is a significant signal: it means the bottleneck in AI infrastructure right now isn't primarily model capability or chip availability at the frontier lab level, but the more mundane constraint of physical data center construction, power procurement, and cooling capacity keeping pace with demand.

That bottleneck has been visible elsewhere this week: Google's €13 billion Finland investment, backed by a 22-year nuclear power deal to secure guaranteed electricity, and reports of a Pentagon-brokered loan to shore up US data center supply chains both point to the same underlying pressure — compute demand from AI is currently outrunning the physical infrastructure needed to deliver it, and power procurement specifically has become as much a strategic constraint as chip supply.

What to watch next

If the 38-gigawatt figure holds, the more consequential story over the next several years won't be whether Microsoft can secure enough GPUs, but whether it — and its utility and construction partners — can build physical capacity and secure power contracts fast enough to keep pace with a roughly threefold expansion target. Microsoft has not officially confirmed the plan, and a figure of this scale, sourced from unnamed people familiar with internal planning, should be read as a strong signal of direction rather than a locked commitment.

Source: Bloomberg: Microsoft AI-Focused Data Center Plan to Add 26 Gigawatts of Compute