Digging Into Microsoft's Pause on Datacenter Expansion

Microsoft tapped the brakes and people hit the panic button. There's no reason to.

When news broke in April then Microsoft was planning on "slowing or pausing" some of its datacenter construction, including a $1 billion project in Ohio and a $3.3 billion project in Wisconsin, people try to read all sorts of doom and gloom into the news.

Like so many news items, the devil is in the details and it is nuanced. The one thing it is not is a sign that the demand for artificial intelligence technology is slowing.

Multiple companies in the datacenter space have stated for the record that datacenter sales and construction remains strong and there is no sign of slowdown. Vertiv, Amazon and Nvidia all said as much in this CNBC article.

In Microsoft's case, the company confirmed that it is halting early-stage projects on rural land it owns in central Ohio's Licking County, outside of Columbus and one in Mt. Pleasant, Wisconsin. The Wisconsin facility in particular was not insignificant; $3.3 billion and 1.5 megawatts spanning 245,000 square feet on a 200 acre plot of land.

In a statement to Wisconsin Public Radio (WPR), a Microsoft spokesperson explained, "We have paused early construction work for this second phase while we evaluate scope and recent changes in technology and consider how this might impact the design of our facilities."

The statement went on to say that the company plans to meet with state officials once it has conducted its evaluation. "We anticipate that this process will last months," the statement said. "In the meantime, our commitment to and construction of our planned datacenter campus continues with high priority for our business."

The item to note is the statement about recent changes in technology. AI equipment is advancing at a ridiculous pace. What's on the market is already obsolete compared to what is coming (and here is where NVIDIA and the rest run the risk of being bitten by The Osbourne Effect), and if you're going to sink billions of dollars into a datacenter, the last thing you want is for it to be obsolete before the lights even go on.

You have to remember some of the Wisconsin facility was not expected to go online until at least 2027 or later. That's two years, and two years have a habit of creeping up on you. But look at how the landscape has changed in just the two years since ChatGPT exploded on the scene. The fact is it is very hard to look out ahead more than a few months, let alone a few years.

IT customers demand long road maps from their vendors so that they can plan accordingly. Microsoft, even though it is an IT vendor itself, is also a customer since it is buying billions of dollars' worth of equipment to put into its datacenters. They want to know where NVIDIA and the rest of the firms will be in a few years.

This is reflected in a change from both NVIDIA and AMD, providing longer road maps than previously. For the longest time, NVIDIA would only publicly discuss. its upcoming generation of process to technology. Now it is talking about two and three generations out over the next few years, something that is new to them. AMD is in a similar situation with its competitive product. For the longest time all they would talk about was the next generation, Now they're giving previews of up to two generations ahead.

In addition to future plans, there's also the fact that Microsoft is already committed to a massive investment in datacenter infrastructure. In 2020 Microsoft announced an expansion of Azure to include availability zones (AZ). An AZ is a distinct, isolated location within a cloud provider's region that has its own power, networking, and cooling infrastructure.

A region can have multiple AZs. It is designed to be independent of other zones in the same region, so that a failure in one AZ does not affect the others. Investing in AZ means investing in an awful lot of equipment because an AZ means redundant resources.
Combined this with its relationship with OpenAI, and you have catalysts for significant increasing datacenter spending. In 2023 their total capex was $35 billion which jumped to $75 billion in 2024 and Satya Nadella has said the firm will spend upwards of $85 billion in 2025.

So while the oversupply argument is legitimate, the capex spend suggests Microsoft is continuing with an aggressive datacenter expansion plan which inherently comes with some resetting of priorities here and there.

This happened at the same time as Microsoft significantly ramped up its self-build efforts, acquiring tens of thousands of acres around the US and the globe, accelerating construction of existing sites and securing gigawatts of power for future sites. Taken together, these moves signaled Microsoft's preparation for what was the most ambitious infrastructure buildout in history.

Another consideration is workload redistribution. OpenAI has reportedly started shifting workloads to Oracle, expanding from just being on Microsoft's Azure cloud. This Is being wrongly interpreted as OpenAI moving away from Microsoft.

Such a move makes no sense. Oracle, for all of its investment, remains an also ran in the cloud space with just 3 percent market share according to the Synergy Research Group. Microsoft is second only to Amazon. Why would you trade a near leader for a laggard?

Because it's not trading, it's adding Oracle. OpenAI's move is a part of the $500 billion Stargate project recently announced where the initial equity funders are SoftBank, OpenAI, Oracle and MGX, so it seems clear OpenAI will be expanding its technology to the Oracle Cloud infrastructure, not leaving Microsoft.

Don't take my word for it, look at OpenAI's statement in their Stargate announcement:     

This also builds on the existing OpenAI partnership with Microsoft. OpenAI will continue to increase its consumption of Azure as OpenAI continues its work with Microsoft with this additional compute to train leading models and deliver great products and services.

There are other elements that could come into play, such as just the strategic reallocation of assets. Microsoft made a lot of plans that looked good initially but reality changed their perspective on things and they realized that they didn't need to build the Wisconsin datacenter as initially planned.

There's also the potential for supply chain issues. Shortages have plagued the industry since even before the COVID lockdown and they have persisted due to extremely high demand. There's also the potential impact of tariffs on semiconductors and that scenario seems to change daily.

Of course, there is always the case of oversupply, that Microsoft may have overestimated demand for AI capacity and it's simply pulling back. But if that were the case it would probably pull back on multiple fronts not just one.

Microsoft has signed leases for more than 20 gigawatts of power and has postponed leases for a few hundred megawatts. In the grand scheme of things, this is a tweak not a slamming of the brakes.

Posted by Andy Patrizio on 05/30/2025