Investor skepticism is rising regarding the substantial investments U.S. tech companies are making in AI infrastructure, particularly in light of slow returns and advancements from the Chinese startup DeepSeek, which has demonstrated AI technology that rivals or surpasses Western competitors at a significantly lower cost.
Based in Redmond, Washington, Microsoft has canceled leases amounting to "a couple of hundred megawatts" with at least two private data center operators, as reported by analysts led by Michael Elias, who referenced supply chain investigations.
Additionally, Microsoft has halted the conversion of statements of qualifications, which are preliminary steps to formal leases, the analysts noted, mentioning that other tech companies, including Meta Platforms, have previously taken similar actions to reduce capital expenditures.
The report was released late Friday but gained traction on social media over the weekend, leading to coverage by various media outlets on Monday.
A company spokesperson confirmed that Microsoft’s plan to invest over $80 billion in AI and cloud capacity for this fiscal year remains on schedule. "While we may strategically pace or adjust our infrastructure in some areas, we will continue to grow strongly in all regions," the spokesperson stated.
Microsoft's stock, which lagged behind most major tech companies last year, saw a decline of approximately 1% in late morning trading.
The lease cancellations represent a significant change for a company that has invested billions in data centers to alleviate supply constraints that have hindered its ability to meet AI demand.
According to Bernstein analyst Mark Moelder, this news could suggest a decrease in demand, particularly following disappointing quarterly results from leading cloud providers, but it also reflects the capacity expansion Microsoft has undertaken in recent years.
Microsoft faced challenges in meeting demand due to difficulties in securing sufficient capacity. As a result, management might have opted to rent data centers and GPU resources, potentially at a significant premium, and pursued additional agreements for future capacity beyond their actual requirements, according to Moelder.