Microsoft has terminated leases in the U.S. with at least two private data center operators, TD Cowen analysts wrote in a report published Feb. 21.

After the report sent jitters through the market, Microsoft representatives, appearing at a conference in Australia, “strongly refuted” any speculation that the company had changed its $80 billion construction strategy.

Microsoft, which famously considered reopening the nuclear facility at 3-Mile Island to power its future AI facilities, could be in an oversupply position, TD Cowen said.

“Our recent checks indicate that, in some situations, Microsoft is using facility power delays as a justification for the (lease) termination,” they said.

The cancelations total “a couple of hundred” megawatts. Microsoft has also pulled back on its conversion of statement of qualifications (SOQs) to leases, and re-allocated a portion of its international spending to its domestic budget, according to the report.

In an SOQ-to-lease agreement, the company would give data center developers a list of qualifications that could then be fulfilled by data center developers. Microsoft would then take the lease.

TD Cowen speculated that the move could mean that Microsoft feared it was in an “oversupply” situation, meaning the company fears it has bought more AI computing power than it believes it needs.

Microsoft is the primary benefactor for OpenAI, one of the primary competitors in the emerging market.

After the report came out, Microsoft said the numbers were interpreted incorrectly. The company typically invests according to a 10-year cycle, and the recent moves were following a typical pattern, Microsoft said.

“Thanks to the significant investments we have made up to this point, we are well positioned to meet our current and increasing customer demand,” a company spokesperson said, market tracking website Seeking Alpha reported.

The TD Cowen analysis did not comment on the report as an indicator of a wider movement.

China-based DeepSeek AI debuted in January, offering an application that could match U.S. programs’ abilities but required much less electricity. Most U.S. tech companies said they remained dedicated to developing their systems.