The AI Luddite Premium
Everyone has an opinion on whether the data center buildout is a bubble. This week somebody checked what the bond market actually charges for it.
Trepp tracked four data center CMBS deals, three of them priced in 22 days between June 9 and July 1, carrying roughly $3.1 billion of mortgage debt. Two of the three came wide of initial price talk. That does not usually happen. An issuer wants interest heavy enough to tighten the spread. Getting pushed the other way is the market saying something.
So: is the CMBS market finally discounting AI infrastructure risk, or is old-fashioned credit underwriting still driving the hierarchy?
Old-fashioned credit underwriting, by a wide margin. The Blue Owl deal priced tightest at 120 over Treasuries, right at talk, on 13.8 years of weighted average lease term and 100% investment-grade rent from a AA tenant. Blackstone went at 140, ten wide, on 19.5 years of lease term and 100% IG rent, carrying the highest leverage in the group at 74.9% LTV. The widest priced at 175, also ten wide, on 4.7 years of lease term and 48.7% IG rent, despite low leverage and the strongest debt yield of the four.
Fifty-five basis points from tightest to widest, and it is the rent roll that explains it. Lease term and tenant credit did the work. A 10.5 debt yield at 58% LTV is normally a clean deal. Under five years of remaining lease term on a data center is what made investors ask for more.
Trepp still hedged, and the hedge is the good part. Given the run of headlines about moratoriums, power pricing, water use, and tax breaks for hyperscalers, they suspect a small sentiment premium is creeping in. They named it the AI Luddite premium, after the 19th century textile workers who smashed the looms. The irony they flagged: community pushback constrains new supply, which is a gift to the buildings already standing. Nobody has watched a full lease cycle roll on a data center yet. How many re-lease, and how future-proof they turn out to be, sits underneath every one of these spreads.
The cycle read came from Marty Allen of Grandbridge, who is building out master and special servicing capacity now. His forecast: "CRE is not monolithic, and this wasn't a single cycle, but maybe several overlapping ones." He is also candid that extend-and-modify worked far better than anyone would have guessed in 2020. The new wrinkle is duration. The market's shift from ten-year to five-year paper pulls the next maturity peak closer than people are used to.
On the equity side, CVC's Rob Lucas described a carry structure built for this kind of market. Portfolio company executives share the upside and, unusually, the downside. They "will also be hit financially, substantially, if that business fails." Set that against last week's read from Hunt Club, where executives negotiate money off the top of a deal, before the waterfall, just to stay. Same broken waterfall, opposite fix.
Commercial real estate shows discuss the economy in 88% of episodes against 30% across our full corpus, and at a sourer sentiment than any other cohort. Data center mentions corpus-wide are running 9.7% this month, down from a 14.1% peak in February. The noise is fading. The paper is still pricing.
Sources: Lonnie Hendry (Trepp), The TreppWire Podcast, "410. The Plot Thickens: Inflation Rerun, AMC's Comeback, Data Center Spreads and Corporate Real Estate Shifts," Jul 24, 2026. Marty Allen (Grandbridge Real Estate Capital), The TreppWire Podcast, "409. The Next Phase of CRE Capital Markets," Jul 21, 2026. Rob Lucas (CVC), Dry Powder, "Himalayan Heights w/ CVC's Rob Lucas," Jul 22, 2026.
Four recent data center CMBS deals show investors discounting weighted average lease term and investment-grade tenancy rather than AI infrastructure sentiment, though Trepp suspects a small sentiment premium is creeping in.