Microsoft downshifts on data centers - Oracle and CoreWeave reap benefits

Microsoft has significantly reduced its data center expansion plans, scaling back more than previously reported. The decision follows major changes in its relationship with OpenAI and underperformance in its own AI business.

Microsoft had originally planned an unprecedented expansion of its data center infrastructure in 2023 and 2024. According to SemiAnalysis, the company signed binding preliminary agreements for over 5 gigawatts of new capacity, which was intended to go online between 2025 and 2028. Alongside these commitments, Microsoft acquired tens of thousands of acres of land worldwide, accelerated construction at existing locations, and secured gigawatts of power capacity.

Financial figures reflect the scale of these ambitions. In fiscal year 2024, the book value of Microsoft’s property, plant, and equipment rose by $26 billion, while its land holdings increased 44 percent year-over-year. Construction commitments tripled to more than $35 billion. At its peak, Microsoft accounted for over 60 percent of all new leasing contracts in the hyperscale segment, SemiAnalysis said.

Despite this expansion wave, the company halted all new leasing activity in mid-2024. The widely reported 2 gigawatts in canceled leasing contracts pertained only to non-binding letters of intent.

According to SemiAnalysis, the more consequential move was Microsoft’s decision to freeze 1.5 gigawatts of self-developed capacity scheduled for 2025 and 2026. Many large-scale projects have since seen little progress. While land development activities continue, construction of buildings and orders for cooling and electrical infrastructure have been stopped or delayed. Microsoft is now pursuing a more flexible approach to future capacity expansion.

The end of exclusivity with OpenAI and internal product challenges

Microsoft’s original expansion was driven by OpenAI’s rapid scaling. Following the success of ChatGPT, OpenAI planned to multiply its computing capacity, with Microsoft initially playing a central role.

This situation changed fundamentally in 2025, when the exclusive partnership between Microsoft and OpenAI ended. OpenAI shifted much of its training infrastructure to Oracle and Crusoe, which are building a 1.2-gigawatt data center at the first "Stargate" campus in Texas. OpenAI also signed a five-year, $11.9 billion contract with CoreWeave and invested $350 million in the provider. In the future, CoreWeave GPUs could also be used for OpenAI inference tasks—a function previously handled exclusively by Microsoft.

In addition to these external developments, Microsoft is also facing weaker-than-expected demand for its own AI products, according to SemiAnalysis. Progress in developing large language models internally has lagged behind competitors such as Amazon.

Anecdotal evidence and analysis of current order volumes indicate that Microsoft’s enterprise AI demand is also falling short of expectations.

Implications for capital expenditures and suppliers

Despite these cutbacks, Microsoft is still planning to expand total capacity in 2025 and 2026. However, the structure of its investments is changing: spending on GPUs is increasing significantly, while construction of new company-owned data centers is expected to stagnate or decline slightly. The company is shifting more toward leasing models.

For suppliers such as Vertiv, the impact is less dramatic than often assumed, according to SemiAnalysis. Much of the leased data center space is already under construction, so the pipeline of orders remains stable. However, the balance of power among hyperscalers is shifting, with providers like CoreWeave and Oracle gaining greater prominence.