There isn’t a scarcity of large-scale information middle buildouts deliberate for 2025, however the tempo and scale may go away some motive for concern.
In 2024, Moody’s Scores detailed the fast enlargement of information middle development. This development continues in 2025, however the accelerating funding comes with vital credit score dangers, based on a brand new sector research by the monetary providers big.
The report, launched Might 5, highlights how AI’s evolution is reshaping the info middle panorama whereas elevating issues about potential overbuilding, technical obsolescence, and disruptions from commerce tensions.
Key findings from the Moody’s report embody:
-
AI information facilities are more and more being constructed as huge campus-style “AI factories” with 1-5 GW of energy capability.
-
Information middle rack density has doubled since 2016, with new AI installations exceeding 200 kW per rack.
-
Hyperscalers are regularly adjusting capability plans to match unsure future computing wants.
-
Capital investments required for AI computing signify considerably increased monetary dangers than earlier tech cycles.
-
Rising commerce tensions and tariffs will disrupt provide chains and enhance prices, probably delaying new tasks.
Information Facilities Are Now an Infrastructure Asset Class
Information facilities have advanced from being seen primarily as actual property property to sitting firmly within the vital infrastructure class, attracting vital funding.
This evolution comes with dramatic will increase in venture scale – a pattern that’s accelerating quicker than many anticipated.
John Medina, Moody’s senior vp, informed DCN that information middle scale has advanced dramatically, with 100 MW amenities as soon as thought of massive now giving option to gigawatt-scale tasks.
“What was hyperscale 5 years in the past? It’s very completely different from two years in the past,” Medina mentioned.
The report additionally highlights how information middle densities have elevated considerably, with common rack density rising to about 12 kW in 2024 from 6 kW in 2016. Hyperscale amenities now vary from 40-60 kW per rack, whereas AI information facilities are probably exceeding 200 kW.
AI Factories: Distant Places, Gigawatt Scale
The Moody’s report highlights a number of huge AI information middle tasks, together with the 5.6 GW Marvel Valley venture in Alberta, Meta’s 2 GW campus in Louisiana, and OpenAI’s Stargate AI three way partnership.
This shift towards distant, power-rich areas represents a strategic bifurcation within the business.
These AI factories are altering basic assumptions about information middle improvement, although the report notes that cloud infrastructure nonetheless dominates current capability.
“AI was the motive force, however the capability is coming on-line,” Medina mentioned.
Energy Constraints Reshaping Website Choice
Energy is a big concern total for information middle buildouts, particularly as AI drives energy calls for increased than ever earlier than.
At this yr’s Information Middle World occasion, AFCOM program chair Invoice Kleyman pointedly said that the primary constraint for AI information facilities is energy.
This problem is echoed within the Moody’s report, which identifies energy availability as a key constraint for information middle development past 2028.
“One might say your complete world doesn’t have sufficient energy for what they [AI data centers] want,” Medina mentioned. “The difficulty is when you will get that energy.”
That actuality is resulting in information middle builders and hyperscalers working straight with utilities. There are additionally conditions the place new energy era gasoline vegetation are being constructed on-site with information facilities.
There are additionally conditions like Homer Metropolis, the place a former coal energy plant is being repurposed to fulfill the facility era wants of AI information facilities.
One other space the place information facilities can broaden is to areas the place there’s “stranded gasoline” – that’s, pure gasoline that isn’t at the moment getting used.
“You’re beginning to see these items being cited as a result of they know that the facility plant is definitely cheaper to construct than the servers that I am placing into this facility,” Medina mentioned.
Evolving Monetary Fashions to Handle Danger
With AI information facilities requiring unprecedented capital investments, monetary buildings are evolving to distribute danger, with pre-leasing changing into customary observe.
“I feel that’s the distinction between now versus the dot-com increase,” Medina defined. “Now persons are not comfy placing in billions, not realizing if they’re going to have a tenant.”
Bigger builders are more and more utilizing partnerships to share publicity.
“I feel now what we’ve seen for hyperscale, just about, like 90 plus %, is pre-leased, and the identical factor for the AI campuses,” Medina mentioned. “We’d anticipate most of them to have pre-leasing, however not all of them. It’s additionally why you are seeing them being inbuilt phases.”
