Microsoft Plans 38-Gigawatt Data Center Expansion Amid Compute Shortage
Bloomberg’s Brody Ford reports that Microsoft is considering a multiyear data-center expansion that could more than triple capacity to 38 gigawatts, as computing shortages force it to turn away AI and cloud customers. The plan, which Microsoft has called inaccurate in its reported details, reflects both the risk of customers shifting workloads to rivals and rising demand for CPU capacity alongside GPUs as AI agents interact with enterprise systems.
Microsoft’s reported capacity plan would more than triple its fleet
Brody Ford reported that Microsoft is planning a multiyear expansion that would more than triple its data-center fleet, reaching as much as 38 gigawatts. The company currently has 12 gigawatts of capacity worldwide, Ford said, including two gigawatts specifically for AI.
Ford characterized the figure as a plan rather than a fixed outcome, subject to delivery and demand conditions. Microsoft initially did not respond to Bloomberg’s request for comment, he said; it later said the numbers were inaccurate without offering much more detail.
Company-filings data shown by Bloomberg put Microsoft’s projected data-center lease commitments that had not yet commenced at $329 billion by the first quarter of 2026. Those leases are not a measure of installed or online capacity. They do, however, indicate the scale of commitments associated with a buildout that depends on facilities being delivered and brought online.
A capacity shortfall can hand customers to competitors
The commercial pressure is immediate: capacity a cloud provider cannot supply may be bought from someone else. Ed Ludlow said compute constraints had caused Microsoft to turn away some AI business. Ford said Microsoft, Amazon, and Google all face the same constraint: with more data-center supply, they could generate more cloud revenue.
Temu illustrates the risk. Ford described the Chinese e-commerce company as a major Microsoft Azure customer that could not obtain more capacity from Microsoft and instead went to Oracle. For Microsoft, the issue is not simply foregone usage in the moment. A customer seeking capacity elsewhere can begin placing workloads with a competing cloud provider.
This is the kind of situation that Microsoft is really trying to avoid.
The reported expansion is therefore directed at a competitive constraint as much as a technical one: adding capacity fast enough to serve demand while customers are still looking for it.
Agent workloads can raise demand for CPU capacity
Microsoft already has a substantial infrastructure base. Brody Ford said its 12-gigawatt global fleet is likely larger than that of any provider except Amazon Web Services, while noting that precise estimates across all rivals are unavailable.
The workload mix cannot be reduced to GPUs. Ludlow noted that cloud infrastructure long supported conventional software, hosting, and storage. Ford said Microsoft has been emphasizing that, in the agentic era, much of what people consider AI work happens on CPUs rather than solely on GPU infrastructure.
Inference agents do not only run models. They may access databases and create records, Ford said—work that draws on CPU capacity. The infrastructure requirement is therefore not confined to accelerators used for model computation. It also includes the general-purpose compute supporting the databases, enterprise systems, and transactions agents interact with.
In this era, you need maybe more CPUs than you expected even a year or two ago.
That makes the reported 38-gigawatt trajectory broader than a bet on GPU-intensive model workloads. Microsoft is expanding against current cloud shortages while preparing for an AI workload mix that can require both accelerators and CPU-backed operations.