Microsoft Is Sitting on a Pile of Unused GPUs -- Redmondmag.com

Generation AI

Microsoft Is Sitting on a Pile of Unused GPUs

We may now know why Microsoft tapped the brakes on datacenter construction earlier this year. It's not that it can't get equipment, it can't get power.

Speaking on the All Things AI podcast, Microsoft CEO Satya Nadella made a stunning admission: he has the GPU hardware in his hands, he just can't get any power to run them.

"Quite frankly, the biggest issue we are now having is not a compute glut, but it's power and it's sort of the ability to get the builds done fast enough close to power," he told the show's hosts. "So if you can't do that, you may actually have a bunch of chips sitting in inventory that I can't plug in. In fact, that is my problem today. It's not a supply issue of chips. It's actually the fact that I don't have warm shells to plug into."

It's a stunning admission that Microsoft, a near $4 trillion company, can't power its own datacenters. He didn't say which ones, but they are most likely the ones in Wisconsin and Ohio that Microsoft paused construction.

Datacenters have been drawing their power from the commercial power grid, to the ire of consumers who are seeing their electric bills go up as datacenters consume more power. In an attempt to be more nimble and responsive, not to mention better neighbors, Microsoft and other hyperscalers like AWS and Google have begun going into business as power companies.

Microsoft's power initiatives are indeed complex, and are mostly centered around sustainability goals. One of those is the use of nuclear power. Earlier this year, Microsoft signed a 20-year power purchase agreement (PPA) with Constellation Energy to buy power from the Three Mile Island Nuclear Generating Station Unit 1, which has been shut down since 2019.

Once operational in the 2027–2028 time frame, the reactor will provide roughly 835 MW in capacity, supplying not only a nearby Microsoft datacenter but the nearby regional grid as well.

What's really scary is that 835 megawatts is not enough. There is talk of multi-gigawatt data centers in the pipeline.

Nadella also said that it sometimes can't meet location demands. "Sometimes you may have to say no to some of the demand, including some of the OpenAI demand," he said. "Because sometimes [OpenAI CEO] Sam [Altman] may say, 'Hey, give me build me a dedicated, multi gigawatt datacenter in one location for training.' That makes sense from an OpenAI perspective, but doesn't make sense from a long-term infrastructure build out for Azure."

The implications are far reaching, well beyond just Microsoft.

If Microsoft's data centers can't get power, then there's a pretty good chance that the other datacenters from Amazon, Google, etcetera are having power problems as well. This will not be a problem solved by switching providers and if it is, that's bad news for Microsoft.
If Microsoft is sitting on a bunch of unused products, why should it order anymore? The same goes for the competition. If many of them are having trouble powering their datacenters with their existing deployments, should they be in a rush to purchase and deploy more?
How severely will this impact Microsoft's growth plans and projections? Nadella did not go into detail on how many datacenters are impacted. If it's only one data center, that's not too bad period but if it's across multiple data centers, then that could have an impact on Microsoft's AI rollout. And by extension, it will impact OpenAI as well which is dependent on Microsoft data centers for its growth.

I have a query in to Microsoft, I will let you know if it comments.

Posted by Andy Patrizio on 12/02/2025

Featured

Where Air Gapped Backups Actually Fail, Part 2

Air gapped backups can still fail when configuration drift, lost encryption keys and routine human mistakes go unnoticed until recovery is needed.
Nadella Says Enterprise AI's Future May Depend Less on Frontier Models Than Learning Systems

Satya Nadella is making the case that the next phase of enterprise AI will not be won by the company with the highest-scoring model on a public leaderboard.
AI Agents Could Put Enterprise GenAI Budgets Under Pressure

Enterprises moving generative AI from pilots into production may be underestimating the cost of running those systems at scale, particularly as inference workloads and AI agents become more common, according to a new Gartner report.
Banks Targeted by Fileless Phantom Stealer Phishing Campaign

A new phishing campaign is targeting banks and other high-value organizations with Phantom Stealer, a commercially available infostealer that runs in memory to avoid traditional detection, according to new research from Fortra.
Where Air Gapped Backups Actually Fail, Part 1

Air gapped backups can provide critical ransomware resilience, but restore testing gaps, documentation drift and media rotation mistakes can leave organizations unsure whether they can actually recover.