A data center the size of a power plant — Microsoft’s $17.4B AI engine roars to life

by

Following a recent announcement of letting go of 42 employees from its Redmond base, Microsoft revealed a significant $17.4 billion infrastructure partnership with Nebius – underscoring their commitment to advancing AI technologies.

As an analyst, I’ve just learned about a significant development following Microsoft’s disclosure of a standalone agreement expected to save the U.S. government $6 billion over a three-year period. This announcement, along with previous moves, clearly emphasizes their ambitious $80 billion AI infrastructure strategy. The new deal with Nebius could potentially be a pivotal component of this strategy, so let’s delve into its implications and understand how it aligns with Microsoft’s broader AI initiatives.

Why Microsoft turned to Nebius

Microsoft is rapidly growing its artificial intelligence network, with the quick introduction of Copilot, Azure, and business AI tools causing a strain on their current data centers. By teaming up with Nebius, they can access the Vineland, New Jersey facility earlier, which began construction in early 2025, rather than waiting several months for Microsoft’s own facilities to become operational.

By making this strategic decision, Microsoft aims to spread its infrastructure across various collaborators. This move helps prevent potential disruptions or bottlenecks. We’ve witnessed how delicate digital networks can be, as evidenced by recent cuts to the Red Sea cables causing delays and calls for alternative routes. In essence, Microsoft is following a similar approach-extending its reach as it delves further into AI and strives to become a prominent figure in this domain. Armed with ample resources, it can comfortably take on significant initiatives.

Through handling the basic infrastructure requirements, Microsoft is able to divert funds and expertise towards enhancing sophisticated system designs, energy conservation, and developing sovereign cloud services. Additionally, this agreement provides them with flexibility, allowing for additional capacity expansion as the demand increases.

Arkady Volozh, the founder and CEO of Nebius, said:

Nebius’s primary AI cloud service, catering to AI startups and large corporations alike, is thriving remarkably. Beyond our main operations, we anticipate securing substantial long-term partnerships with prominent AI research institutes and tech giants. I am thrilled to announce the first of these agreements, and I’m optimistic about more such collaborations in the future.

What the deal means for Nebius

For Nebius, it seems clear that a period of steady growth awaits them. Sealing a deal worth $17.4 billion ensures a continuous income stream, providing the organization with a solid and dependable financial base.

In response to the announcement, the market demonstrated swift action, with Nebius’ shares experiencing a significant jump over 50%. This rise underscores robust investor trust in the company.

This agreement additionally confirms the credibility of Nebius. Having Microsoft as a client makes Nebius seem like a genuine contender, which may attract attention from more hyperscalers and AI research centers in search of reliable infrastructure partners.

Similarly significant, it expedites the growth of Nebius. Construction on its colossal data center spanning 2.6 million square feet in Vineland, New Jersey began in early 2025. The first phase of 300 MW dedicated to AI capabilities is set to be activated this summer, with further stages continuing into 2026. With Microsoft’s support, Nebius now boasts the necessary assurance and funding to advance that project, as well as initiate new constructions in Europe and the U.S.

Microsoft’s $17.4 billion deal with Nebius guarantees a 300 MegaWatt data center capacity focused on AI, which is equivalent to the electrical power needed for over 220,000 American households.

300 MW of AI-centric power capacity could be a bit puzzling, just as it was to me initially. To help clarify, let’s break it down: One megawatt (MW) is equal to 1,000 kilowatts, which is roughly the energy requirement for around 750 typical U.S. households. Therefore, when you multiply that by 300 MW, it amounts to providing electricity for more than 220,000 homes simultaneously.

In simpler terms, this data center Microsoft is using is significantly larger than what most businesses typically have. Most enterprise facilities operate within the 10-50 Megawatt (MW) capacity, but this particular site surpasses that scale by a considerable margin. Essentially, it’s like renting a warehouse that’s as large and powerful as a power plant, primarily designed for running artificial intelligence models.

The new reality of AI demand and outsourcing?

Microsoft isn’t unique in offloading data center space to external providers, but its $17.4 billion agreement with Nebius represents one of the most substantial third-party deals ever, particularly for artificial intelligence computation capabilities.

Larger technology companies such as Microsoft, Amazon, and Google typically construct their own infrastructure to accommodate expansion. On the other hand, the rise of generative AI has triggered an immense need for computational resources, which often results in scarcity and compels businesses to seek external providers instead.

This agreement represents one of the most significant third-party AI infrastructure deals yet, expediting Microsoft’s $80 billion AI growth plan at an accelerated pace.

Microsoft has collaborated with external data centers in the past, but generally on a significantly smaller level. The deal with Nebius demonstrates the substantial increase in the level of stakes involved.

Presently, it seems as though Microsoft is unveiling a new billion-dollar deal every week, fueled by the growing AI wave. This prompts us to ponder: Is simply pouring vast amounts of capital into AI and its associated domains a guaranteed route to dominance within this sphere?

If Microsoft appears determined to invest significantly, it’s challenging to identify any potential competitor who might reasonably match those resources. I’m curious to hear other people’s thoughts on this.

Read More

2025-09-10 01:10