Update: Rising US Borrowing Costs Won't Slow Massive AI Data-Center Buildout as Potential Profit Outweighs Spending
(Updates with comments from Morgan Stanley starting in 13th paragraph.)Rising interest rates won't stop companies such as Alphabet's (GOOG, GOOGL) Google, Amazon (AMZN) and Microsoft (MSFT) from spending enormous amounts of money to build artificial intelligence data centers because the potential profit far outweighs slightly higher borrowing costs, according to industry analysts.The yield on benchmark 10-year US Treasuries rose to 4.58% on Thursday from 3.96% on Feb. 26 as investors worry that rising inflation could prevent the Federal Reserve from cutting interest rates. Earlier this week, the rate reached its highest level since January 2025. That affects borrowing costs for AI hyperscalers that are on track to spend $800 billion in capital expenditures this year and an additional $1 trillion next year.Rates will rise and inflation will remain a concern as the war in Iran will keep oil above $80 a barrel until February, Peter Tchir, head of macro strategies at Academy Securities, said in an interview with. Still, the expected revenue gain from AI products and services is at this point outweighing concerns that rising rates will dampen the data-center buildout, benefiting companies in and adjacent to the AI space including real estate investment trusts, he said."Right now, the profitability of these data centers and AI, and the perceived profitability, just means that they're not really going to be constrained by 50 or 100 basis points in yield," Tchir said. "These are fairly large bets that this is going to work, and it's going to work in a huge scale, in which case borrowing at 5%, 7% or 9% will turn out kind of trivial."It costs $45 billion to $50 billion to build out 1 gigawatt of data-center capacity, said Mandeep Singh, global head of technology research at Bloomberg Intelligence. SpaceX revealed in its initial public offering prospectus this week that it's renting one of its data centers to Anthropic for $1.25 billion a month, or about $15 billion a year."If it costs $50 billion to build an AI data center, and you're able to generate up to $15 billion in revenue in year one, then it takes three and a half years to get your investment back, and then obviously you'll make returns from year four onward," Singh said in an interview.Analysts agreed that benchmark borrowing costs will continue to rise this year."The bond market is a little bit freaked out, we're seeing inflation and risk in the current environment putting a lot of pressure on longer duration Treasury yields to get to very high levels," Elizabeth Templeton, senior product manager for fixed-income indexes at Morningstar, said in an interview. "Seeing the 30-year yield at 5.1% this week, the highest since 2007, is certainly an indication that there's some worry in the markets right now around inflation. That could certainly continue to impact the 10-year the rest of this year."Smaller AI companies including CoreWeave (CRWV) and Nebius (NBIS) could be affected more by the rise in borrowing costs than hyperscalers Amazon, Google and Microsoft, Bloomberg's Singh said. Those companies and others have already sold $300 billion in debt to fund their AI investments this year, according to Bloomberg News. CoreWeave and Nebius didn't respond to a request for comment.Still, the scale of AI borrowing is so large that it can't be ignored, said Kevin McPartland, an analyst at Crisil Coalition Greenwich. Debt deals that are already underway shouldn't be affected, he said."It doesn't take much of a move when you're talking about billions of dollars of financing to really change the economics," he said. "The devil's advocate would be: These are literally the largest companies in the world that have an incredible amount of free cash flow, and so these are not two- or three-year plans, these are five- and 10-year plans, in which case I'm sure they've modeled out the risk of everything, from interest rates to other geopolitical issues," McPartland said."If you're committed for 10 years to spending tens or hundreds of billions, of course you don't want the cost of financing to go up, but maybe the answer is some short-term slowdowns, but no long-term change in strategic planning."Investors should stay exposed to AI but be more selective, Morgan Stanley analysts said Friday in a note to clients.Increased borrowing costs have led to an uptick in rotation across equities, exposing some weakness in AI-aligned companies, the analysts said. Still, AI earnings were "resilient," volatility is contained, and valuations support staying exposed to the sector. the note said."The recent adjustment does not look like a classic risk-off episode or a wholesale defensive rotation," Morgan Stanley said. "It is better characterized as a selective unwind of crowded AI-led momentum exposure, with higher yields providing an additional tailwind to value."The two main data center REITs -- Equinix (EQIX) and Digital Realty Trust (DLR) -- have been refinancing debt and financing their development at roughly the current level of interest rates for the last couple of years, Jeffrey Langbaum, senior REIT analyst for Bloomberg Intelligence, told.That's dented their earnings growth but hasn't deterred them because the returns they generate from the developments outpace the debt costs, he said. Equinix and Digital Realty didn't respond to requests for comment."The returns they are getting on their developments are well in excess of the costs of capital," he said. "My thesis is that even if overall demand shrinks, they should still be able to get their share because they're keeping the size of their development business at a manageable level and not getting out over their skis and trying to expand too far too fast."Equinix sales in the second quarter that ends on June 30 are pegged at $2.58 billion and adjusted funds from operations are estimated at $11.24 a share, according to estimates compiled by FactSet. If realized, that would be up from $2.26 billion and $9.91 a share, respectively, in Q2 2025.Digital Realty Trust revenue in the second quarter is projected by analysts in a FactSet survey at $1.65 billion, while adjusted funds from operations are seen at $1.80 a share. Sales in Q2 last year were reported at $1.49 billion and AFFO was $1.68 per share.Data-center REITs are seeing a tailwind from momentum behind artificial intelligence expansion, Wells Fargo Investment Institute analysts John Sheehan and Amanda Martinez said in a note to clients earlier this month.REITs have a diverse range of offerings including colocation, which allows for multiple users, from hyperscalers to smaller companies, at a single location and interconnection, which means lower-latency connections and better tenant retention, as "particularly notable features" of some data-center buildouts, the analysts said."We are favorable on the data-center REITs subsector as we believe it possesses durable growth prospects, attractive margins, and solid pricing power," Sheehan and Martinez said in their note. "We also view the sub-sector as an attractive route for gaining exposure to the AI theme within the real estate sector, particularly as AI use cases continue to expand and support sustained demand and pricing power."Academy's Tchir said he expects the 10-year Treasury yield to rise to 5% in the next few months, and that investors are rewarding AI capital spending."We're almost in what I call free money stage, where if you announce $10 billion to spend, your stock goes up $20 billion, so why wouldn't you announce spending?" he said. "We are so underinvested in data centers and AI that even if your project turns out not to be as good as you thought, it's still going to do well, because someone needs that compute right now, and for the foreseeable future."Matthew Leising and Tim WeatherheadPrice: $383.20, Change: $-4.46, Percent Change: -1.15%
