Microsoft Plans to Triple Data Center Capacity to 38 Gigawatts by 2032 After Turning Away Cloud Clients

Microsoft is planning to more than triple its global data center footprint, from about 12 gigawatts today to over 38 gigawatts by 2032, according to people familiar with the company's internal planning cited by Bloomberg on September 10. The company is moving to address a shortage that already cost it real customers.

E-commerce platform Temu signed a cloud deal with Oracle after Microsoft couldn't supply the capacity it needed in preferred regions, Bloomberg reported, a detail confirmed across multiple outlets including Yahoo Finance and BigGo Finance. Microsoft's own gaming division capped cloud streaming time for Xbox subscribers when hardware strain hit consumer products. And GitHub, which Microsoft owns, had to lean on rival Amazon Web Services to handle traffic Azure couldn't absorb.

On the GitHub outages, sourcing varies across reports. Tech Times reported GitHub suffered nine outages in May 2026 alone. Yahoo Finance, also citing Bloomberg, described a single eight-hour outage in August that forced GitHub to reroute U.S. traffic through overseas facilities. These are two different claims about two different incidents, not the same event described twice, and neither report reconciles the other.

Chips Sitting Idle, No Room to Plug Them In

CEO Satya Nadella said in a 2025 podcast interview that the constraint was never about chip supply. "It's not a supply issue of chips," Nadella said. "It's the fact that I don't have warm shells to plug into." Microsoft had GPUs sitting in inventory with nowhere to install them.

CFO Amy Hood told investors on a 2025 earnings call the company expected to catch up and didn't. "We had hoped to be in balance by the end of Q4 but we did see some increased demand," Hood said. "So we are going to be a little short, a little tight as we exit the year."

Only about 2 gigawatts of Microsoft's current 12-gigawatt footprint is dedicated to AI-specific chips, according to Bloomberg's reporting as relayed by Ground News and Value Add VC. That share is expected to grow to roughly a third of the 38-gigawatt target by 2032. The rest of the buildout, including the new "East US 3" cluster near Atlanta, is built around conventional CPU workloads rather than the GPUs that power AI models.

The Money and the Accounting

Microsoft's capital expenditures hit $145 billion in the most recent fiscal year, a company record according to Value Add VC. The company has guided for $175 billion in calendar year 2026 and $50 billion in the first fiscal quarter of 2027 alone, per BigGo Finance and Kao Hoon International. Future lease obligations for data centers stand at $329 billion.

Microsoft is also stretching data center lease terms from 15 years to 25 years. That move lowers the annual capital expense the company reports each year, spreading the same cost over a longer runway. It's a legal accounting choice, but it means the headline capex figures understate the real annual commitment Microsoft is locking in.

Not Just Microsoft's Problem

Microsoft isn't alone in this race. Amazon, Google parent Alphabet, and Meta are collectively on track to deploy close to $2.4 trillion on data center equipment and leases in the coming years, according to reporting cited by Value Add VC and NAI 500. A 38-gigawatt Microsoft footprint alone would draw more electricity at peak than the entire state of New York consumes at its highest demand.

That scale is exactly why states are pushing back. Massachusetts, New York, and Oklahoma have each moved this year to require data-center developers to bring their own clean power generation or pay into ratepayer-protection funds, according to Value Add VC, rather than let hyperscaler demand quietly drive up electricity bills for regular households. If Microsoft's buildout strains regional grids, ordinary ratepayers who never signed up for the AI boom shouldn't be the ones covering the bill through higher utility rates. Whether these state measures actually keep costs off consumers, or just shift the fight to permitting delays, is still playing out.

Microsoft has not publicly confirmed the 38-gigawatt figure, and the company did not respond to a request for comment cited in Ground News's reporting. A stated internal planning target is also a long way from megawatts actually flowing through a substation. Grid interconnection queues in many U.S. regions already stretch for years, turbine and transformer lead times remain multi-year bottlenecks industrywide, and community opposition to new server farms has intensified enough to already stall projects in several states. Whether Microsoft actually hits 38 gigawatts by 2032, or revises the number the way Bloomberg's sources say the roadmap could shift, will depend on how fast those physical and political constraints break in the company's favor.