Microsoft has canceled multiple data center leases in the United States, affecting “a couple of hundred MWs” of capacity. The company reportedly has also pulled back on its international data center expansion, according to an industry update from investment analysts from TD Cowen, a multinational investment bank.The shift, based on channel checks conducted by their analysts, suggests that Microsoft is recalibrating its AI infrastructure needs amid a possible oversupply of data center capacity.
The report, published on February 21, 2025, states that Microsoft has terminated select leases with at least two private data center operators across multiple U.S. markets.It also highlights Microsoft’s decision to delay or halt the conversion of Statement of Qualifications (SOQs) into full leases, signaling that planned expansions may not materialize as expected. Furthermore, a significant portion of Microsoft’s projected international data center spending is now being redirected to the U.S., suggesting a slowdown in overseas AI infrastructure investment.
Microsoft Reverses Course on AI Data Center Growth
This shift represents a notable change from Microsoft’s prior AI expansion strategy, where the company had been rapidly securing AI computing capacity to support its cloud business.Microsoft had aggressively pursued large-scale data center projects to accommodate OpenAI workloads, positioning Azure as a leading AI compute provider.
However, recent signs point to a more cautious approach. Last year, Microsoft announced large-scale AI infrastructure investments, yet by January, the company had already begun halting major projects. The most notable example was its decision to pause construction on a $3.3 billion AI data center in Wisconsin, which was originally intended to support OpenAI-powered workloads.
A Microsoft spokesperson confirmed about the decision, “We have temporarily paused early construction work for this second phase while evaluating scope and recent changes in technology.”This aligns with TD Cowen’s analysis that Microsoft is adjusting its data center strategy to reflect evolving AI demand and infrastructure priorities.
Why Is Microsoft Slowing Down AI Infrastructure Expansion?
According to TD Cowen, the slowdown may be linked to an oversupply of AI data center capacity. The firm’s channel checks indicate that Microsoft has:Walked away from multiple 100+ MW data center deals that were still in early or mid-stage negotiations.Allowed over 1 GW of Letter of Intent (LOI) agreements on large-scale sites to expire.
Abandoned at least five land purchases for future data center construction in multiple Tier 1 U.S. markets.
These findings suggest that Microsoft may have overestimated near-term AI infrastructure needs, particularly in relation to OpenAI’s compute requirements.
Analysts believe that Microsoft’s earlier demand signals were based on expected OpenAI growth, but recent shifts in AI workload forecasting may have reduced immediate infrastructure needs.
One reason Microsoft has given for the lease terminations is facility and power delays at certain sites, according to the TD Cowen report. This mirrors a strategy previously used by Meta, which in 2023 canceled multiple U.S. data center leases after deciding to scale back its metaverse-related infrastructure investments.
Microsoft has also been shifting toward more sustainable data center operations, including a plan to eliminate water usage in cooling systems by 2026. This shift could introduce further complexity in site selection, potentially contributing to project delays or reconsiderations.
Reallocation of International Investment
Beyond U.S. lease cancellations, TD Cowen’s findings also highlight a pullback from international AI data center expansion. Microsoft is reallocating a considerable portion of its planned international investment toward U.S. markets, signaling a material slowdown in overseas leasing activity.
This move aligns with industry trends, as data center investments are increasingly concentrated in North America, where AI compute demand remains strongest.However, it also suggests that Microsoft may be scaling back AI infrastructure in regions where power availability or regulatory factors pose additional challenges
Facility and Power Constraints Also Play a Role
One reason Microsoft has given for the lease terminations is facility and power delays at certain sites, according to the TD Cowen report. This mirrors a strategy previously used by Meta, which in 2023 canceled multiple U.S. data center leases after deciding to scale back its metaverse-related infrastructure investments.
Microsoft has also been shifting toward more sustainable data center operations, including a plan to eliminate water usage in cooling systems by 2026. This shift could introduce further complexity in site selection, potentially contributing to project delays or reconsiderations.
Competitors Push Forward as Microsoft Slows AI Expansion
While Microsoft is pulling back from some AI data center commitments, Amazon Web Services (AWS) and xAI are continuing to expand aggressively.
AWS in December announced plans for its Ultracluster supercomputer, a groundbreaking infrastructure project designed to challenge Nvidia’s dominance. The Ultracluster, codenamed “Project Rainier,” is built using Amazon’s custom Trainium chips and is set to become one of the world’s most powerful AI supercomputers to date.
At the same time, Elon Musk’s xAI announced to dramatically scale up its Colossus supercomputer to 1 million GPUs, signaling a different approach to AI infrastructure. Unlike Microsoft, which is focusing on cloud-based AI compute via Azure, xAI is investing in in-house AI supercomputing to reduce dependence on external cloud providers.
Google is also expanding its AI compute capacity through custom silicon. Google Cloud continues to refine its tensor processing unit (TPU) strategy, optimizing its cloud AI infrastructure to compete with Microsoft’s Azure AI services.
What This Means for the AI Data Center Market
The divergence in AI infrastructure strategies between Microsoft, AWS, Google, and xAI suggests that there is no consensus on the best way to scale AI compute resources.
Microsoft’s decision to scale back expansion could indicate that the industry is reaching a transition point, where raw compute power is not the only factor driving AI growth.
TD Cowen analysts believe that Microsoft’s pullback is directly tied to OpenAI’s evolving compute requirements. Microsoft’s AI expansion in 2023 and early 2024 was heavily influenced by the expectation that OpenAI workloads would demand continuous infrastructure scaling.
However, if OpenAI has optimized its models or adjusted its cloud strategy, Microsoft may no longer need as much additional capacity in the short term.
Microsoft’s Future AI Infrastructure Strategy
Despite its pullback from some AI data center investments, Microsoft is unlikely to reduce its AI ambitions. The company remains heavily invested in AI, with Azure continuing to serve as a key platform for enterprise AI applications, Copilot+ PCs, and Microsoft 365 integrations.
Going forward, Microsoft’s AI strategy may shift toward infrastructure efficiency and sustainability, rather than sheer scale. This includes investments in liquid cooling, renewable energy integration, and hardware acceleration, which could make its existing data center footprint more efficient without requiring constant physical expansion.
Microsoft’s ability to balance AI growth with infrastructure sustainability will be a key factor in its long-term competitiveness. While AWS and xAI push for rapid compute expansion, Microsoft appears to be taking a more selective, strategic approach—prioritizing optimization over immediate scaling.
