Newsletter · · Ashutosh Agarwal

Digital Realty Pays Up for Powered Land as Prologis Builds on Spec - Powering AI: Data Centers, Land & REITs - Week of July 2, 2026

Powering-AI newsletter for the week of July 2, 2026. Digital Realty wrote a $3.5B check for occupied hyperscaler NOI in Northern Virginia and the stock fell 5%, Prologis started building on spec into a starved Chicago market, and Meta's excess-compute pivot cut against colo landlords, with the whole complex now hanging on late-July capex guides.

Powering AI: Data Centers, Land & REITs

Week of July 2, 2026: Digital Realty Pays Up for Powered Land as Prologis Builds on Spec

Happy almost-Fourth. The bankers cleared their desks this week and left us a genuinely useful data center print: Digital Realty writing a $3.5 billion check, a Prologis spec build in Chicago, and a Meta headline that cuts in an uncomfortable direction for landlords. The through-line, as always this cycle, is power. Let's get into it.

TL;DR

  • Digital Realty bought its way to more hyperscaler NOI: $3.5B for Blackstone's stake in three fully-leased Northern Virginia campuses, at a fair-not-cheap 6.5% cap rate, and the stock fell 5%.
  • Industrial supply is still starving: Prologis is building 454,000 sq ft on spec in a Chicago market where new completions fell 33% last year. That's the bull case for logistics rents in one datapoint.
  • The whole complex hinges on late-July capex guides. Hyperscalers are spending ~90% of operating cash flow on capex; any flattening of the FY27 guide is great for them and bad for the infrastructure names that trade off it.

What's new

Digital Realty paid up for occupied hyperscaler NOI, and the market shrugged. On Motley Fool Hidden Gems Investing: "Picking the Winners of the Honeywell Breakup", the Fool's Matt Frankel walked the DLR-Blackstone deal: DLR buys out Blackstone's roughly two-thirds interest in three Northern Virginia data centers for ~$3.5B ($1.2B cash, $2.3B new stock, plus assumed debt and ~$1.4B more capex to finish them). All three are "100% leased to hyperscalers on 15-year deals with 3.6% annual rent escalators," two stabilizing in H1 2027 and the third in H1 2028. Why'd the stock drop 5%? It's dilutive now (DLR and Blackstone each dumping $2.3B of stock), FFO-accretive only once stabilized, and at a 6.5% cap rate Frankel called it "a fair price… not a bargain for top quality assets." Long-duration, escalating, hyperscaler-backed cash flow, but you're paying a full price for it.

A Digital Realty insider spelled out why supply is so hard to add. The best operator color of the week came from Open Circuit: "The new reality for data centers: no easy answers", where Ian Black, SVP and global head of energy at Digital Realty, laid out the new math. Data-center power costs "$10 to $13 a watt" versus "$2 a watt" for renewables, so a project that would be a $1B renewable spend is a $15B data-center spend once you bring your own generation. Land options run just "60 to 90 days" and you commit regardless of your queue position; he's been waiting ~2 years on an Oncor study in Texas. His verdict on the resource stack was blunt:

"There is no future of data centers without gas. And maybe SMRs 10 years from now."

For landlords, that's the moat and the risk in one sentence: the same power scarcity that protects incumbent, already-powered campuses is what makes new supply brutally slow and expensive to add.

Prologis flipped the supply spigot back on, carefully. Per Crain's Daily Gist: "Fight over renter protections brewing in City Hall" (reporting Danny Ecker's Crain's Chicago Business story), a Prologis venture paid ~$29.3M for a 25-acre Glendale Heights parcel and will build twin 454,000-sq-ft warehouses on spec as part of a ~$100M project, completing late 2027. The context is the whole story: Chicago-area industrial availability has sat below 5% for two straight years, near its pandemic-era low, and new local warehouse completions fell 33% last year to the lowest level in over a decade. When the largest owner in the country starts building without tenants into a starved market, that's a landlord voting with its balance sheet on where rents go next.

Meta says it has compute to spare, and that's a double-edged headline. On Squawk on the Street: "Meta's Cloud Ambitions 07/01/26", Evercore's Mark Mahaney (Buy, $930 PT) framed Meta's plan to sell excess AI compute as potentially $15-20B/yr of high-margin revenue, and, more importantly, a tell: "Meta is saying that they've got excess capacity, i.e. maybe they don't need to keep ramping up CapEx next year." He was careful to note this looks idiosyncratic: "Amazon doesn't have this kind of excess capacity… and Google doesn't have it either." The stock jumped 9%. For colo landlords, "the biggest AI spenders have room to spare" is not an unambiguously good message.

The debate

Bull frame, scarcity is the moat. The pieces the bulls want are all on this week's tape. Power is the binding constraint: on Climate One: "When Your New Neighbor Is… a Data Center", Camus Energy CEO Astrid Atkinson pegged hyperscaler grid-connection waits at 3-7 years, with an opportunity cost of "between about five and ten billion dollars a year" for every gigawatt they can't turn on, against a grid that's grown load ~0.5-1%/yr for two decades. On Limitless: "The AI Energy Stack", the hosts put US data-center power demand roughly doubling from ~80 GW in 2026 to ~150 GW by 2028, with full-power lead times near five years and gas turbines sold out to ~2029. Layer on falling industrial supply (see Prologis), and you get the classic re-rate: whoever already owns powered, leased square footage is sitting on a scarce, escalating annuity.

Bear frame, the tenants hold the cards. The counter is just as well-voiced. On Avory: "Six Questions on Meta", the analyst reminded everyone that "most people buy 60, 70, 80% of their compute internally, and then they rent out some of the capacity," self-build is the default, colocation the margin. Meta's excess-capacity pivot pushes the same way. And on Avory: "Five Questions for the Second Half of 2026", the punchline for our complex: hyperscalers are spending ~90% of operating cash flow on capex, and "any flattening of those guides is… probably a good thing for them, bad for the infrastructure plays." The nuclear-restart fantasy also took a cold shower on Limitless: SMRs "not possible" at scale before ~2030, leaving gas and 90-day fuel cells (Oracle's 2.4 GW Bloom deployment) as the real near-term answer. If demand cools even at the margin, the landlords with development pipelines feel it first through compressed yields.

Names in play

  • DLR: buying stabilized, escalating hyperscaler NOI, but paying a full 6.5% cap and diluting to do it; near-term FFO drag, long-term quality. Next catalyst: Q2 print and any read on development yields. (Motley Fool, Open Circuit)
  • PLD: spec-building into a sub-5% availability market with completions down 33%; supply discipline is doing the heavy lifting for the rent thesis. (Crain's)
  • AMT: no fresh operating datapoints, but two wide-moat deep-dives (We Study Billionaires TIP826, Intrinsic Value TIVP079) reiterated the toll-road case. Crown Castle's fiber/small-cell review and SBAC were quiet this week, no operator or analyst voices worth citing.

Read-throughs

  • Power is the whole ballgame. Interconnection queues (3-7 yr), $5-10B/GW/yr opportunity cost, and behind-the-meter gas are now baked into every development decision. Watch it in DLR and EQIX development yields, not just utility multiples.
  • Gas and fuel cells > nuclear, for now. Bloom's 90-day deployments and turbine backlogs to 2029 are the near-term power answer; SMRs are a 2030+ story.
  • Neoclouds are eating marginal colo demand. Nebius, CoreWeave and Iron are arbitraging permitting to sign multi-billion-dollar deals with Meta and Google, a reminder that "who leases the incremental gigawatt" isn't only the traditional REITs.