Microsoft’s AI Infrastructure Shift – A New Balance Between Ownership and Leasing

Microsoft is doubling down on self-built AI data centers while maintaining strategic colocation partnerships. Here’s what that means for the future of cloud infrastructure.

Mar 02, 2025

Dear Reader,

Microsoft is canceling hundreds of megawatts of data center leases across the U.S., shifting more investment into self-built AI campuses.

Yet, it’s not abandoning colocation.

Instead, the company is refining its approach.

Expanding where it has power and land control while maintaining leased capacity to meet immediate and regional needs.

For years, Microsoft has leased a larger share of its data center footprint than some of its hyperscale competitors, with approximately 18% of its cloud infrastructure relying on third-party colocation.

Leasing has allowed Microsoft to scale rapidly without tying up capital in real estate and power development.

However, as AI reshapes cloud economics, Microsoft appears to be adopting a more hybrid approach—expanding self-build investments in U.S. power-secure regions while continuing to lease both domestically and internationally where necessary.

This shift suggests Microsoft is prioritizing long-term ownership for AI compute hubs while maintaining colocation for agility.

If this model proves successful, it could influence how other hyperscalers balance ownership, leasing, and AI infrastructure investments.

Here’s what you need to know.

The Big Move: Microsoft’s AI Data Center Strategy

The AI era is driving unprecedented demand for power, cooling, and compute infrastructure. As Microsoft scales to meet these demands, it’s adjusting how it deploys capital and resources—owning more where strategic, leasing where flexible.

Microsoft’s recent lease cancellations do not indicate a reduction in overall investment. Instead, they reflect a shift toward self-built AI campuses in power-secure U.S. markets like Virginia, Texas, and Arizona, where long-term cost and energy control can provide competitive advantages.

By reallocating resources to regions where infrastructure is more predictable, Microsoft may be positioning itself for more sustainable hyperscale growth.

However, leasing remains an essential component of its strategy. Microsoft continues to rely on colocation in key areas within the U.S. and internationally, particularly in markets where permitting delays, power constraints, or geopolitical uncertainty make self-building more complex. This hybrid approach allows Microsoft to meet immediate capacity needs while maintaining strategic ownership of its most critical AI hubs.

Microsoft isn’t just shifting away from colocation—it’s rebalancing where and how it invests in infrastructure. This nuanced approach could become the blueprint for the next phase of hyperscale expansion.

How Microsoft Stacks Up Against Amazon, Meta, and Google

Microsoft’s evolving strategy comes as every major hyperscaler refines its AI infrastructure model. While its increased self-build investments in the U.S. signal a shift, its continued reliance on leasing in select markets shows it is not abandoning colocation altogether.

Meta, for example, is taking a different approach, reportedly in talks for a $200 billion AI data center initiative that, if executed at scale, would dwarf Microsoft’s current spending.

Meta is focusing heavily on self-built, AI-first infrastructure, betting on a fully owned model for long-term efficiency. Microsoft’s approach, by contrast, maintains greater flexibility, balancing high-density, self-owned AI compute hubs with leased capacity in strategic locations.

Amazon remains committed to a hybrid model, blending self-built campuses with long-term colocation agreements. As the world’s largest cloud provider, AWS has maintained high flexibility through leasing while gradually increasing its owned footprint.

Microsoft’s shift toward selective self-building could put pressure on Amazon to consider whether its balance between owned and leased infrastructure needs adjustment.

Google is the most unpredictable player in this race. While its DeepMind and Gemini AI projects require significant compute, its approach to infrastructure expansion has been more measured than Microsoft’s or Meta’s.

If Microsoft’s hybrid strategy proves effective, it may set a precedent for Google and others to refine their leasing vs. ownership models for AI-scale growth.

Microsoft is refining its infrastructure strategy, but will other hyperscalers adopt the same balance—or will they lean further into fully owned AI infrastructure?

The Global Context: AI Infrastructure is Evolving

Microsoft’s data center repositioning is part of a broader realignment within the AI infrastructure sector. Across global markets, hyperscalers and regional operators are adapting to shifting regulatory, technological, and economic conditions.

North America

Nvidia Reports Record AI Revenue – The company has announced unprecedented earnings, fueled by surging demand for its Blackwell AI chips.
Meta Explores $200 Billion AI Data Center Investment – The company is reportedly in discussions to build a massive AI infrastructure network.
Apple Expands Data Center in Waukee – As part of its $500 billion U.S. investment initiative, Apple is increasing capacity at its Iowa-based data center.
CoreWeave Prepares for $4 Billion IPO – The specialized AI cloud provider is moving toward a potential $4 billion initial public offering, positioning itself as a key player in high-performance AI compute infrastructure.

Europe

Spain Approves 438MW Data Center Projects – The country’s grid operator has given Solaria the go-ahead for two large-scale data center developments.
Italy’s 1GW Data Center Initiative Moves Forward – Eni is partnering with MGX and G42 to develop a large-scale AI data center network across Italy.
UK Nears Decision on £10 Billion Data Center Investment – Authorities are reviewing plans for a major hyperscale project in Cambois, Northumberland.

Asia-Pacific

Alibaba Invests $53 Billion in AI & Cloud – The Chinese tech giant has announced plans to allocate $53 billion over three years to strengthen AI-driven cloud infrastructure.
Indonesia-Singapore Data Center Bridge Announced – Gaw Capital and Sinar Primera are collaborating on a cross-border digital infrastructure project.
TikTok Commits $8.8 Billion to Thai Data Centers – ByteDance is significantly expanding its cloud and AI compute capacity in Thailand.

Middle East & Africa

e& Sells 40% Stake in Khazna Data Centers – The UAE telecom giant has finalized a $2.2 billion deal, marking one of the region’s largest hyperscale transactions.
Riyadh to Host New AI Data Center – DataVolt has secured land from MODON to construct a cutting-edge AI and cloud facility in Saudi Arabia.
Kenya’s iXAfrica Launches First Phase of NBOX1 – The new data center facility now includes a Kenya Internet Exchange Point (KIXP) PoP, improving regional connectivity.

South America

Mexico’s Largest Data Center Campus Energized – ODATA has successfully brought its hyperscale data center online, reinforcing the country’s role as a regional cloud hub.
Entel Bolivia Expands Data Center Operations – The telecom provider has launched a new data center and is advancing fiber expansion efforts.

The Bigger Picture – AI Infrastructure is Being Rewritten

Microsoft’s evolving balance between self-build AI hubs and strategic leasing suggests that hyperscalers are entering a new phase of infrastructure planning. The traditional model—leasing for flexibility and self-building for control—is being refined to account for AI-driven power and capacity demands.

If this approach proves effective, it could lead to:

A redefined role for colocation providers, as hyperscalers prioritize ownership in high-power regions but lease elsewhere.
More strategic site selection for self-builds, favoring power-secured locations with regulatory stability.
Greater diversification in leasing agreements, with hyperscalers maintaining agility while securing long-term owned AI campuses.

For investors, policymakers, and data center operators, the key takeaway is clear: AI cloud infrastructure isn’t just about ownership or leasing—it’s about balancing both in the right markets.

Will Amazon, Meta, and Google adjust their strategies in response?

What’s your take on Microsoft’s strategy? Reply to share your insights.

Until next week,

Obinna

Tell us what you thought of today’s email.

Good?
Ok?
Bad?

Hit reply and let us know why.

PS... If you're enjoying this newsletter, share it with a colleague.

And whenever you’re ready, here’s how I can help you:

1. Subscribe to Global Data Center Hub’s weekly newsletter for expert insights.

2. Have a question? Hit reply, and I’ll answer it in the next edition.

3. Explore our archive for more in-depth analysis on AI’s impact on data centers.

Let’s tackle the future of data center infrastructure together.

Global Data Center Hub