Morningstar DBRS' Takeaways From 2025 Credit Outlook New York: Artificial Intelligence Fuels Demand for Data Centers, an Emerging Asset Class

As part of its takeaways series, Morningstar DBRS is publishing several write-ups about pertinent topics discussed at 2025 Credit Outlook New York, an industry conference on our expectations for structured finance in 2025 as well as highlights on emerging asset classes. One such emerging asset class is data centers, which is particularly timely given DeepSeek's recent arrival on the world stage. The Chinese artificial intelligence (AI) start-up claims that training costs for its model were only $6 million over two months, miniscule compared with the $100+ million that Microsoft, Amazon, Alphabet, and others have spent. And yet, independent benchmarking suggests that its R1 model is able to meet or beat models from big Silicon Valley names like OpenAI, despite using less powerful Nvidia processors. (American export controls blocked Chinese companies from accessing high-end Nvidia processors).

Demand for data centers has skyrocketed with the increased attention on AI. DeepSeek's emergence does not remove this growth in demand, but it may change the magnitude if less computing power is needed, per Stacey Mawson, Senior Vice President, Asset Finance, at Morningstar DBRS. She noted competition in the market was always a possibility. For more information, please see our commentary titled "Could DeepSeek Deep-Six Data Centre Credit Ratings?"

Depending on the structure of the transaction, we have two approaches to analyzing data centers: the project finance approach and the structured finance approach. The project finance method focuses more on cash flow. Is there enough stable cash flow to cover debt payments during the life of the loan? In addition, Mawson explained that for greenfield projects with construction risk we analyze both the construction phase and the operating phase separately, with the lower of the two generally determining the ultimate issuer and debt credit rating. For the construction phase, we start with the contractor, looking at its size, reputation, and track record, as the project's success depends on the contractor's ability to carry it to completion. We also analyze the construction's complexity, available credit enhancements, and the project's schedule and budget. Please see "Global Methodology for Rating Essential Digital Infrastructure" for more detail.

Our structured finance method also incorporates cash flow performance in addition to a real estate-centric valuation approach, per Michael Vidmar, Senior Vice President, North American Real Estate Adjacent Ratings, at Morningstar DBRS. Is the value of the property enough to cover the loan amount? We also look at the quality of the tenancy and roll in reviewing the ability to cover debt service as well as the physical and operational qualitative aspects such as power capacity and scalability, critical infrastructure redundancy, connectivity, and renewable energy, among others, per Vidmar; our "Rating and Monitoring Data Center Transactions" methodology further outlines our considerations under the structured finance approach that has successfully rated transactions across both asset-backed securities and commercial mortgage-backed securities issuer platforms in the U.S. and Europe.

No matter the type of data center, from large hyperscale data centers to smaller colocation facilities, a stable tenant roster is important. According to Mawson, in our analysis of hyperscale data centers, we review the length of the initial lease period, the number of renewal options available, the sponsor's ability to change the lease rate at renewal, the current lease rate, and how competitive the data center is to market in the area. If the debt term is shorter, then we will analyze a full cash sweep scenario at maturity, per Mawson. This helps us to see how many renewal periods are needed for full repayment to better understand the lease renewal risk.

On a related note, Victor Leung, Senior Vice President, Project Finance, at Morningstar DBRS outlined churn risk for colocation facilities. The churn rate¿the rate at which tenants move in and out of data center space¿is a key factor in analyzing colocation facilities, which have hundreds of tenants. We start with the historical industry average of the facility type and then take into account any features specific to the property. According to Leung, the industry's average churn rate has hovered around 10 basis points per month, or about 1% to 2% annually. However, in his experience, the actual churn rate has been higher than that, with some months having spikes well above the industry average. In general, we can stress churn rates to be as high as 10% annually in our analysis to see if the debt can handle these spikes. Vidmar added that the source of most of the churn is mergers and acquisitions activity.

Beyond AI, consumer demand for 5G internet and cloud services will also prop up the need for data centers around the world.

Written by Caitlin Veno

Notes:
All figures are in U.S. dollars unless otherwise noted.

For more information on data centers, visit https://dbrs.morningstar.com or contact us at info-DBRS@morningstar.com.

DBRS, Inc.
140 Broadway, 43rd Floor
New York, NY 10005 USA
Tel. +1 212 806-3277