Microsoft Plots 38 Gigawatts of Data Center Capacity

Microsoft is planning for 38 gigawatts of total data center capacity, including 26 gigawatts of new AI-focused compute, after shortages forced it to turn away some AI and cloud business, Bloomberg Technology reported on Sept. 10, 2026.

Microsoft Corp. (MSFT) is laying out a build-out of its computing estate on a scale normally discussed by national grid operators rather than software companies. According to a report from Bloomberg Technology, the company's artificial-intelligence-focused data center plan will add 26 gigawatts of compute, part of a broader target of 38 gigawatts of total data center capacity.

The number that matters commercially, though, is not the gigawatts. It is the admission behind them: shortages of capacity forced Microsoft to turn away some AI and cloud business. For a hyperscale provider, that is the most expensive sentence in the industry — demand arrived, and the company could not serve it.

What 26 gigawatts of new compute actually implies

Gigawatts have quietly replaced square footage as the unit of account in the data center business, because power, not floor space, is the binding constraint. A gigawatt is the output of a large nuclear reactor. Twenty-six of them, dedicated to AI-focused compute, is a commitment to secure generation, transmission and cooling on a timeline that utilities measure in years and interconnection queues, not quarters.

Set against the 38-gigawatt total capacity figure, the 26 gigawatts of AI-focused additions imply that the remaining 12 gigawatts covers everything else in the fleet — the general-purpose cloud, storage, enterprise workloads and the legacy footprint. That 12-gigawatt figure is illustrative arithmetic on the two numbers Microsoft's plan puts forward, not a separately reported disclosure, but it frames the shift: on this plan, the majority of Microsoft's electrical footprint is being pointed at AI.

Three practical constraints sit between the plan and the plug:

• Power procurement. Capacity at this scale requires long-dated contracts for firm generation, and increasingly for new generation that does not yet exist.
• Grid interconnection. Securing megawatts on paper is not the same as energizing them; queue positions and substation build-outs govern the calendar.
• Accelerator supply. Racks need silicon. A power envelope only converts into sellable compute when the chips, networking and cooling arrive together.

Turning away business is the tell on Azure demand

Cloud providers rarely acknowledge refusing revenue. When they do, it reframes how investors should read growth. If reported cloud growth has been running below what customers wanted to buy, then the constraint has been on the supply side, and the reported figure understates underlying demand rather than describing it.

That cuts both ways. Supply-constrained growth is a high-quality problem, and it argues that the capital going into the ground has identified customers waiting at the other end. But it also means the growth rate is, for now, a function of how fast Microsoft can energize buildings — an engineering and permitting variable, not a sales one. Investors trying to model Azure will increasingly find themselves modeling construction schedules, turbine deliveries and utility interconnection dates.

The competitive dimension is unforgiving. Capacity turned away does not sit in a queue politely; AI workloads are portable enough that a customer denied capacity in one region will test a rival's. Every quarter of shortage is a quarter in which someone else builds the relationship.

How the market took it

The disclosure landed on a soft session for large-cap technology. Microsoft last traded at 492.44, up 0.16% on the day, having closed the prior session at 491.65 and ranged between 486.00 and 494.52, as of the last trade at 20:00 GMT on Sept. 10, 2026.

That modest gain stands out against the tape. The Nasdaq 100, tracked by the QQQ exchange-traded fund, closed at $708.69, down 1.06%, its weakest showing among the major benchmarks. The S&P 500 proxy SPY finished at $757.83, down 0.60%, and the Dow 30 proxy DIA at $520.75, down 0.63%. A modest advance in a name of Microsoft's weight, on a day when the tech-heavy index fell more than a percent, suggests the market read the build-out as evidence of demand rather than as an unwelcome escalation of spending.

That reading has not been automatic in this cycle. Investors have oscillated between rewarding hyperscalers for capturing AI demand and penalizing them for the depreciation and free-cash-flow drag that follows a capital-intensive build. Framing the spend as a response to business already turned away is the strongest available answer to the second concern.

The power question moves to the center of the story

A 38-gigawatt plan makes Microsoft an energy company in everything but name. Securing that much firm capacity pulls in nuclear restarts and life extensions, gas turbines with multi-year order books, large-scale renewables paired with storage, and behind-the-meter arrangements that bypass congested grids entirely. Each route carries a different cost curve and a different regulatory clock.

It also puts Microsoft in direct competition with other hyperscalers and with ordinary ratepayers for the same electrons. Utility commissions in several states have already begun scrutinizing how the cost of serving very large data center loads is allocated. A plan of this size guarantees that scrutiny intensifies, and rate-case outcomes are one of the few variables in this build that Microsoft does not control.

What to watch from here

The plan is a target, and targets in this industry are revised. The measurable checkpoints over the coming quarters are straightforward:

• Capital expenditure disclosures. The pace at which spending converts the 26-gigawatt AI plan into energized capacity, and whether management continues to describe itself as capacity-constrained.
• Any change in the shortage language. The moment Microsoft stops saying it is turning away business is the moment supply has caught up with demand — a milestone for margins and for competitive positioning alike.
• Power announcements. Long-dated generation deals, nuclear agreements and interconnection wins are the leading indicators of whether the gigawatt figures are deliverable.
• Cloud growth relative to capacity added. If new capacity fills as fast as it opens, the demand story holds. If utilization lags, the depreciation question returns quickly.

For now, the plan reads as a bet that AI demand is durable enough to justify committing to a power footprint the size of a mid-sized country's generation fleet. The market's mild approval on a down day for technology shares suggests it is inclined to agree — while reserving the right to change its mind the first time capacity gets ahead of customers.

Key facts

• New AI compute planned: 26 gigawatts
• Total data center capacity target: 38 gigawatts
• MSFT last trade: 492.44, +0.16%, as of 20:00 GMT Sept. 10, 2026
• Nasdaq 100 (QQQ) close: $708.69, -1.06%

Frequently asked questions

How much capacity is Microsoft planning to add?

Microsoft's artificial-intelligence-focused data center plan calls for adding 26 gigawatts of compute. That sits within a broader target of 38 gigawatts of total data center capacity, a figure that covers the company's full fleet including general-purpose cloud and enterprise workloads, according to a Bloomberg Technology report published on Sept. 10, 2026.

Why did Microsoft turn away AI and cloud business?

Shortages of available data center capacity forced the company to decline some AI and cloud customers. In hyperscale computing, the binding constraint is typically electrical power and the accelerators to fill it, rather than sales demand, so a provider can face more orders than it has energized capacity to serve.

Why are data centers now measured in gigawatts?

Power has replaced floor space as the scarce input in data center construction. A gigawatt is roughly the output of a large nuclear reactor, and AI accelerators consume far more electricity per rack than traditional servers. Capacity plans are therefore expressed as an electrical envelope, because that is what determines how much compute can be installed.

How did Microsoft shares perform on the day of the report?

Microsoft last traded at 492.44, up 0.16%, after a prior close of 491.65 and a day range of 486.00 to 494.52, as of the final trade at 20:00 GMT on Sept. 10, 2026. That small gain came on a weak session for technology overall, with the Nasdaq 100 proxy QQQ closing down 1.06%.

What does supply-constrained growth mean for Azure?

If Microsoft has been refusing business for lack of capacity, then reported cloud growth reflects how much it could deliver rather than how much customers wanted. Underlying demand would be higher than the headline rate. It also means growth becomes tied to construction, permitting and power interconnection timelines rather than to sales execution.

What are the main risks to a 38-gigawatt plan?

Securing firm generation, obtaining grid interconnection and receiving accelerator hardware all have to line up. Utility regulators are also examining how the cost of serving very large data center loads is allocated among customers. Any of these can slow the conversion of announced gigawatts into revenue-producing capacity, and targets of this size are frequently revised.

Sources

• Microsoft AI Focused Data Center Plan to Add 26 Gigawatts of Compute — Bloomberg Technology

Photo: Hoài Nam · Pexels Licence — source