By Aditya Soni
(Reuters) – Microsoft has scrapped leases for sizeable data center capacity in the U.S., suggesting a potential oversupply at the tech giant as it builds out artificial intelligence infrastructure to meet a possible surge in demand, TD Cowen said.
Skepticism has been growing among investors over the billions of dollars U.S. tech firms have been channeling into AI infrastructure due to slow payoffs and breakthroughs at Chinese startup DeepSeek, which has showcased AI tech on par with or even better than its Western rivals at a fraction of the cost.
Redmond, Washington-based Microsoft has canceled leases totaling “a couple of hundred megawatts” of capacity with at least two private data-center operators, analysts led by Michael Elias said, citing supply chain checks.
Microsoft has also paused converting statement of qualifications, or precursors to formal leases, the analysts added, saying other tech firms including Meta Platforms had previously made similar moves to lower capital spending.
The note was published late on Friday but it picked up traction on social media platforms over the weekend, with several media outlets covering the development on Monday.
Microsoft’s plan to invest over $80 billion in AI and cloud capacity this fiscal year remains on track, a company spokesperson said. “While we may strategically pace or adjust our infrastructure in some areas, we will continue to grow strongly in all regions,” the spokesperson said.
The company’s shares, which underperformed most Big Tech stocks last year, fell about 1% in late morning trading.
Any lease cancellations would mark a sharp shift for a company that has been spending billions of dollars on data centers to overcome supply bottlenecks that have limited its ability to meet AI demand.
The news could possibly indicate lower demand, Bernstein analyst Mark Moelder said, especially after lackluster quarterly results from major cloud companies, but it was also reflective of the capacity build-up at Microsoft in the past years.
“Microsoft needed to meet demand and had a great deal of difficulty finding capacity. Management may, therefore, have rented, even at a meaningful premium, data centers and GPU capacity and negotiated more deals for additional future capacity than they needed,” Moelder said.
(Reporting by Aditya Soni in Bengaluru; Editing by Sriraj Kalluvila)
About the Author
