Why did Microsoft decide to cut two data centers?

On Friday, U.S. stocks plummeted due to Microsoft's drastic cuts in data center Channel Check, and I observed that this topic quickly gained traction over the weekend, with various Twitter influencers starting to debate after the market closed.

The rumor is this:

Rumor going around that $MSFT is paying penalties to get out of some data center contracts & power agreements-- changing tone drastically on CAPEX growth. Might be another DeepSeek shockwave to the semi supply chain.

The trending narrative over the weekend is as follows:

In fact, since January, some clients have been asking us whether Microsoft has cut its IDC. We have verified that it indeed cut 2 IDCs, Wisconsin Kenosha and Georgia Atlanta, and confirmed this through research and North American data center experts in early February. Historical minutes can also be checked through the Jiuqian minutes database, with the interview I conducted on February 9th.

Here are some details and key points.

The Wisconsin Kenosha data center covers 240 acres, and this plan is delayed by 3 years, with the reasons for the delay:

• Microsoft is not participating in the Stargate project, and the capacity from Wisconsin is being transferred to Stargate.
• The Wisconsin Kenosha project was originally planned for a $3.3 billion CAPEX investment.
• After this cancellation, there will be no compensation for Stargate's CAPEX through other IDCs.

The Georgia Atlanta data center covers 488 acres, and this plan is delayed by 1.5 years, with the reasons for the delay:

• Microsoft overestimated the demand around Atlanta; the actual demand is currently less than the supply, making it unnecessary to open this data center separately.
• The Atlanta area has historically often experienced oversupply situations, making it easy to overestimate; there are few enterprise clients here.
• No oversupply issues have been observed in other regions.

Because the utilization rates of other Deepseek IDCs have increased:

• No decline in utilization rates has been seen outside of Atlanta; Atlanta is an exception
• Deepseek saw a 15% increase in GPU utilization after Azure, which is expected to last for 6 months before returning to normal utilization.
• The increase in utilization is mainly due to Double Purchase, where customers are testing the new Deepseek model while using existing models.

The expert also discussed the debugging situation at the Phoenix City IDC and Azure's CAPEX expectations.

Due to the complexity of the GB debugging process, it is estimated that the first batch of 10,000 to 20,000 cards at the Phoenix City IDC will go live in March, while the remaining 70,000 to 80,000 cards may not be operational until the end of May.

The expert believes that CAPEX will not see a quarter-on-quarter increase in the next few quarters, and there may even be a slight quarter-on-quarter decline in CAPEX a year later. However, the expert has always been relatively pessimistic, so this may not be a reliable reference for CAPEX.

Risk Warning and Disclaimer

The market has risks, and investment should be cautious. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment based on this is at one's own risk