Microsoft reportedly targets 38 GW of data-center capacity by 2032

Why it matters

Microsoft's reported target would add about 26 GW of capacity and increase its AI-focused fleet more than sixfold. For Azure-dependent companies, the pace and composition of that buildout will shape access to AI compute as Microsoft directs an expected $175 billion toward calendar 2026 capital expenditures.

Microsoft, the Redmond, Washington-based software and cloud company led by longtime cloud executive Satya Nadella, reportedly plans to expand data-center capacity from about 12 gigawatts today to roughly 38 gigawatts by 2032, with AI-specific capacity accounting for about one-third of the future fleet.

Reuters reported on September 10th that Bloomberg obtained the target from people familiar with the plan. The reported figure is a planning target from Bloomberg's sources, not a formally announced construction schedule. The locations, ownership mix, financing and timetable for the additional capacity remain undisclosed.

Bill Gates and Paul Allen started Microsoft in 1975 to write software for the Altair 8800, a computer sold to hobbyists as a box of parts. Microsoft's executive biography of Nadella says he joined the company in 1992 and served as executive vice president of its Cloud and Enterprise group before becoming CEO in 2014. Under his leadership, the company's growth has become increasingly tied to data centers, accelerators and long-term infrastructure commitments.

The scale of the reported roadmap marks a sharp change from the economics of Microsoft's early software business. The company now supports Azure, Microsoft 365, Copilot, its own AI work and customers including OpenAI with an infrastructure fleet measured in gigawatts. Microsoft employed about 223,000 people as of June 30th, according to its 2026 annual filing, while its cloud expansion also depends on leased facilities, equipment suppliers and power arrangements outside that workforce.

The plan tilts hard toward AI

Microsoft's reported capacity would grow about 3.2 times by 2032. The mix inside that footprint would change faster.

Only about 2 gigawatts of Microsoft's current 12-gigawatt fleet is centered on AI-specific chips, according to Reuters' account of the Bloomberg report. AI infrastructure is expected to occupy roughly one-third of the proposed 38 gigawatts. That implies about 12.7 gigawatts of AI-focused capacity, more than six times the current figure.

Microsoft disclosed another measure of prospective demand in its fiscal 2026 fourth-quarter earnings call: commercial remaining performance obligations reached $678 billion, up 84% year over year, with roughly 30% expected to be recognized as revenue within the next 12 months. Microsoft 365 Copilot exceeded 30 million paid seats, and net paid-seat additions more than doubled from the preceding quarter.

Those figures provide evidence of present demand. They cannot establish how much capacity customers will consume in 2032 or how much of the reported fleet Microsoft will ultimately build, lease, equip and place into service.

Microsoft's recent history makes that distinction material. In March 2025, Reuters reported that Microsoft had abandoned data-center projects representing about 2 gigawatts of electricity capacity in the United States and Europe during the preceding six months. Reuters cited TD Cowen analysts, who attributed the pullback to oversupply relative to Microsoft's demand forecast.

The earlier pullback shows that capacity plans can change before projects become operating infrastructure. A site under consideration, a signed lease, an equipped building and a data center serving paying customers represent different levels of commitment. Bloomberg's reported 38-gigawatt target does not disclose how much capacity has reached any of those stages.

Microsoft has provided more detail about one large project. Its planned campus in Pecos, Texas, announced on June 22nd, is expected to provide approximately 2 gigawatts. In its announcement, Microsoft described Pecos as one of the largest single capacity additions in its history and said the multibillion-dollar investment would span five to seven years. A 38-gigawatt fleet would require several additions on that scale, plus leased sites and smaller campuses, if the reported target is realized.

Nadella's cloud bet becomes a capacity bet

During Microsoft's July 29th earnings call, Nadella said the company had added 88 data centers during fiscal 2026, including 31 in the final quarter. He also said Microsoft had cut GPU "dock-to-live" times in its largest regions by nearly 50% over the fiscal year, reducing the interval between delivery and operational use.

The earnings materials also show that Azure and other cloud-services revenue grew 43% year over year in the June quarter. CFO Amy Hood said customer demand continued to exceed available capacity and that additional computing resources brought online during the quarter were quickly sold. Microsoft Cloud revenue reached $214.4 billion for the fiscal year, while annual Azure revenue surpassed $100 billion for the first time.

Those results support Microsoft's case for adding near-term capacity. The 2032 target still rests on a much longer demand forecast and gives no project-level schedule for turning planned gigawatts into revenue-generating infrastructure.

Microsoft also faces competitors making large commitments of their own. Amazon says Project Rainier will use nearly 500,000 Trainium2 chips for Anthropic workloads, while Google is promoting its TPU 8t training and TPU 8i inference accelerators. Oracle has announced a deployment beginning with 50,000 AMD MI450 GPUs in 2026.

The physical commitments extend beyond chips. Amazon has announced a $15 billion Indiana expansion adding 2.4 gigawatts, while Google disclosed a 13 billion euro investment in Finnish data centers, energy and grid infrastructure. Microsoft is therefore planning against both customer demand and rival cloud providers' efforts to secure capacity.

The accounting horizon is getting longer

Reuters reported that Microsoft expects capital expenditures of $50 billion in its fiscal first quarter of 2027 and $175 billion during calendar 2026. The $175 billion figure is a calendar-year expectation rather than a recurring annual run rate.

Effective in fiscal 2027, Microsoft extended the estimated useful lives of data centers and office buildings from 15 years to 25 years. The company said during its earnings call that the change would move more future data-center leases from finance leases to operating leases. Microsoft also said finance leases appear in its capital-expenditure calculation, while operating leases do not.

Microsoft's annual filing records operating leases as right-of-use assets and lease liabilities, while finance leases are recorded in property and equipment and lease liabilities. Reuters reported that extending the lease period from 15 years to 25 years lowers Microsoft's annual reported capital expenditures. Cash payments and lease commitments remain relevant measures of the company's infrastructure spending.

Microsoft reported in its fourth-quarter earnings materials that it spent $41 billion on capital expenditures in the June quarter. Roughly two-thirds went toward shorter-lived assets, primarily CPUs and GPUs. The company separately reported $5.6 billion in finance leases and $35.8 billion in cash paid for property and equipment.

Because CPUs and GPUs account for a large portion of spending, that structure could let Microsoft slow purchases of shorter-lived hardware if demand weakens. Construction, leasing and hardware installation can also leave the company with different levels of financial exposure at different sites. Microsoft has not provided that breakdown for the reported 38-gigawatt fleet.

For founders building on Azure, execution will shape access to AI compute, regional availability and the cost base beneath their products. Microsoft has shown that it can sell newly added capacity under current conditions. Its own 2025 pullback also demonstrates that forecasts can be revised before planned infrastructure reaches customers.

The reported 38-gigawatt target expresses Nadella's cloud strategy in physical units. Reaching it would require Microsoft to add about 26 gigawatts while increasing AI-focused capacity more than sixfold. Bloomberg's sources have supplied the destination; Microsoft has yet to publish the route.