Newsletter · 2026-07-30 · Ashutosh Agarwal

Record Signings, a Moratorium, and Meta Joins the Landlords - Powering AI: Data Centers, Land & REITs - Week of July 30, 2026

Data centers, land and REITs newsletter for the week of July 30, 2026. Digital Realty signed as much business in six months as it did in all of last year and raised guidance, New York froze new large data center approvals for a year, and Meta hired Amazon's former cloud chief to rent out its own spare compute, turning the biggest tenants into potential competitors.

Powering AI: Data Centers, Land & REITs

Week of July 30, 2026: Record Signings, a Moratorium, and Meta Joins the Landlords

Two things happened this week that pull in opposite directions. The largest data center landlords are having the best year of their lives: Digital Realty signed as much business in six months as it did in all of last year. And the Governor of New York froze new data center approvals for twelve months. Both are true, both matter, and the gap between them is where the money is going to be made or lost.

TL;DR

• Digital Realty's CEO says demand is "well outpacing the supply," with record signings, a $2-billion-plus backlog on 15-year contracts, and raised full-year guidance. He also says the equipment supply chain is "as tight as it's ever been."
• Meta is hiring a landlord. It brought in Dave Brown, who ran Amazon's cloud business, to lead an effort to rent out its own spare computing power. The biggest tenants are starting to become competitors.
• Power, not capital, is the binding constraint, and the timelines got worse. Waiting for a grid connection now runs to 2030 or 2031 in many places, while the cost to build a megawatt of data center has gone from under $1 million (in its previous life as a Bitcoin mine) to $10–15 million and rising.

What's New

Digital Realty had the best six months in its history, and its CEO is not remotely apologetic about it. On Power Lunch, "Tech Earnings Preview, Top Analyst Calls, Digital Realty CEO Interview 7/27/26", Andy Power, President and CEO of Digital Realty (DLR), an operator, not a commentator, laid out a quarter that came with raised full-year guidance, the stock up 25% on the year and coming off its best week in nearly four years:

"We just reported our earnings last Thursday. Our record signings are in many categories. We've signed as much signings in one half of a year, basically, as we did the whole last prior year."

Two details matter more than the headline. First, the shape of the demand: 6,000 customers across three buckets: ordinary corporate technology upgrades, cloud computing, and now AI, spread over "50-plus metropolitan areas, six continents." That is not one bet on one AI boom. Second, the duration: "We are signing long-term 15-plus-year contracts, a $2-plus billion backlog with those type of high investment rate credit customers." Backlog here means leases already signed but not yet generating rent, money in the pipe.

The most interesting moment was when the hosts put Mark Cuban's line to him, that a lot of data centers will end up as pickleball courts once compute gets more efficient. Power didn't dismiss the idea. He redirected it:

"I'm not disagreeing with his thesis, but I don't think that's going to happen in digital realties data centers."

His argument is geography. The workloads he houses have to sit physically close to the people using them, because the speed of light is a real constraint. "When our customers put our workload with us in Northern Virginia, it likely has to stay there, and it can't be going to North Dakota or another state." A speculative AI training warehouse in the middle of nowhere can become a pickleball court. A building that handles a bank's transactions in Northern Virginia cannot easily be replaced.

He was also unusually blunt about the bottleneck, which is the thing that protects him and also limits him: "The supply chain for every element of the data center supply chain, from utility construction components all the way to our four walls, is as tight as it's ever been." If you want generators and switch gear (the electrical guts of a data center), "you have to be locking in components for multiple years in the future."

New York just became the first big state to say no. On Power & Politics, "NY Governor Kathy Hochul on data center moratorium Executive Order, July 24th, 2026", Governor Kathy Hochul explained a one-year freeze on new large data center approvals, and her arithmetic is worth reading slowly. There are 30 applications pending in New York. One county, St. Lawrence, up on the Canadian border, has five of them by itself. If those five alone were all approved and built:

"they would consume as much power as one nuclear reactor as I'm trying to build more nuclear. So they would take all the energy that we're trying to create so I can help drive down utility costs."

Her conclusion on the full pipeline: "we would not have enough to make the power of 30 right now without depleting what we need for New Yorkers and our New York businesses." Projects under 50 megawatts are exempt, as are data centers serving hospitals, bank back offices, and universities. So this is aimed squarely at the big AI campuses.

The detail that will travel furthest is about subsidies. Hochul says she sat down with developers to learn what they get elsewhere in the country:

"And they basically admitted we don't need the tax breaks. We don't need them. We're making enough money. So I found that interesting."

She also said New York intends to "create a template for the rest of the nation," which is either a promise or a threat depending on which side of the trade you're on. For existing landlords, a moratorium on new supply in a constrained market is not obviously bad news. It is a moat made of paperwork. For anyone with a development pipeline in a blue state, it is a real problem.

Meta wants to become your competitor. This is the week's most important development for anyone who owns a data center landlord. Meta reported Wednesday and the stock fell 9%. On Halftime Report, "Fed and Tech Earnings in Focus 7/29/26", the panel flagged the organizational move behind the headline:

"We brought on Dave Brown. He led Amazon's cloud business. He's an executive here. Now he's going to lead our attempt to lease out our compute."

Hiring the person who built Amazon's cloud business to run your compute-rental effort is not an exploratory memo. On Bloomberg Tech, "Earnings Roundup: Meta, Microsoft & Qualcomm", Ed Ludlow of Bloomberg drew the distinction precisely: Meta "operates data centers at hyperscale for its own business, for its internal workloads. That's very different to renting out compute capacity to third parties." Zuckerberg's own line in the release, that AI is "opening the door to entirely new enterprise opportunities," is doing a lot of work.

Why is Meta suddenly interested in being a landlord? Because the spending has gotten genuinely uncomfortable. Per Morning Brew Daily, "Investors Question Kevin Warsh's Credibility & People Are Talking Less?", total costs and expenses rose 55% from a year ago, and free cash flow, the cash left after running the business and paying for buildings, came in at $784 million, down from $8.5 billion a year earlier. That is a 91% collapse. Advertising was fine, at $59.3 billion and ahead of expectations. The buildings are the problem: a $9 billion project in Canada, a $14 billion venture with BlackRock in El Paso, and a single site in rural Louisiana that will cost more than $50 billion. Brent Thill, the Jefferies technology analyst, put the market's complaint in one line: "why don't we have more of a concrete plan?" Zuckerberg's answer on the call was that Meta expects to "have some left over" and has "a few customers in the pipeline," with no names and no numbers.

On Bloomberg Tech, Mandeep Singh, Global Head of Technology Research at Bloomberg Intelligence, was appropriately unimpressed with Meta's chances: "They are late to that cloud game. They are late to that API game... who those anchor customers are going to be? That's a million dollar question." He also flagged the number that actually explains the 9% drop, Meta's operating margin "went from 43% to 31%. It's a massive drop," driven not by hiring but by depreciation on buildings it already put up. Notably, Meta raised its total expense guidance to $165–169 billion but did not raise the midpoint of its capital spending, which Singh reads as forced discipline: "the market has no tolerance for higher Meta capex."

Meanwhile Google admitted, on the record, that it is renting space from other people. This is the single best piece of evidence for the colocation landlords that surfaced all week, and it came from a tenant. On Squawk on the Street, "10AM Hour: Google Cloud CEO, IBM CEO, Tesla's Worst Day in More Than a Year 7/23/26", Thomas Kurian, CEO of Google Cloud, told Jim Cramer:

"So for us, when we look at the short term, we're going to rent some capacity for a few quarters. It allows us to bring customers in, bridge them over to when we have sufficient capacity available... And that's why, because of this unprecedented demand, we are not just offering our own systems, but also renting capacity from some partners."

Cramer pushed: paying for third-party capacity hurts your margins, but you have so much demand you don't want it going elsewhere. Kurian: "Exactly, Jim." Google Cloud grew 82% to nearly $25 billion in the quarter, operating income more than tripled to $8.8 billion, margins hit a record 35.6%, and the backlog reached $514 billion. Existing customers are spending "roughly 50% more than the commitment" they originally signed.

The caveat, and Kurian was careful about it, is the word bridge. Renting is explicitly a stopgap "for a few quarters" until Google's own buildings are ready. Cramer asked the right follow-up, whether Google will renew those third-party leases. Nobody knows. That question is the whole colocation thesis in miniature.

The cost of building a megawatt has gone up roughly tenfold, and an operator put a number on it. On Galaxy Brains, "Datacenter Expansion and Strategy with Simrit Dhinsa", Simrit Dhinsa of Galaxy Digital, who is actually converting a former Bitcoin mining site into an AI data center for the cloud provider CoreWeave, gave the cleanest cost comparison anyone has offered:

"the Bitcoin mining side was probably sub a million per megawatt to build out a full Bitcoin mining data center. Now we're looking at 10 to 15 and growing millions per megawatt as infrastructure, as supply chains get tighter. So the scale of it has basically increased by multiple 10."

That is $10–15 million per megawatt, or $10–15 per watt, and rising. The reason isn't just fancier equipment. A Bitcoin mine could switch off whenever power got expensive; an AI data center serving enterprise customers cannot. As Dhinsa put it: "there was no redundancy with Bitcoin mining... Now we have to be online 100% of the time." Redundancy is expensive.

What made his project possible at all was inherited electricity. The Helios site he bought "came with 800 megawatts of power approved plus all the transformers basically installed are on PO," already ordered. Texas's grid operator has since approved another 800 megawatts, taking the site to 1.6 gigawatts. In this market, a site with approved power and transformers on order is the asset. The concrete is almost incidental.

The Debate

The bull case: scarcity is the moat, and it got deeper this week.

Start with the queue. On Catalyst with Shayle Kann, "The commercial battery comeback", Tim Haid, Senior Vice President at Voltus, gave hard numbers on how long it takes to plug anything new into the grid, the "interconnection queue," essentially the waiting list to connect:

"If you want to build a utility scale battery in PJM today and you submit your interconnection application to get in the queue, on average, you're waiting about six years right now. In Kaiso, you're waiting maybe nine years. In ERCOT, you're waiting maybe four years."

PJM is the grid covering the mid-Atlantic and much of the Midwest, including Northern Virginia, the densest data center market on earth. Six years. And on Switched On, "Gas Is Letting Data Centers Grow Beyond the Grid", a BloombergNEF analyst confirmed the same picture from the developer's side: "a lot of the conversations we're having is you can't get grid connected until 2030, 2031."

The workaround is to generate your own power on site, "behind the meter," in the jargon, and that's faster, at "a year, two years or even three years" for a gas turbine, engine, or fuel cell. But that door is closing too, and the detail here is excellent. Turbine manufacturers are now rationing. They'll ask you to "put down like 20% of your order in prepayment," or prove "you have permits, you have land, you have financing for the data center before the actual manufacturer is going to consider even partnering with you to manufacture the turbine for delivery by 2030." And they won't jump their existing customers: "these businesses are relationship businesses... you don't want to prioritize data centers and undercut all of the existing relationships you have with IPPs." The framing that stuck with me is that the industry is chewing through its options in order, grid connection, then on-site gas turbines, then lighter aeroderivative turbines, then fuel cells, and each one gets used up in turn.

Then there's the new normal in how deals get done. Haid described what's now standard: hyperscalers must "bring their own capacity" to get a data center approved. Since "hyperscalers think in gigawatts, not megawatts," and "when hyperscalers think about where do I get net new capacity, especially in really constrained regions like PJM, there's not really anywhere to go in the short term." Put all of this together and you get the bull case in one sentence: whoever already owns powered, leased square footage owns something that cannot be reproduced at any speed. That is exactly what Power's record signings, 15-year contracts and rising backlog look like from the inside.

The bear case: this is a property cycle wearing a technology costume.

The sharpest bear argument of the week wasn't about AI demand at all. On Know Your Risk Podcast, "What If AI Is Really a Real Estate Cycle?", the host read out a passage, sourced to the Groundbreaker newsletter, and also cited by Luke Gromen on MacroVoices, that reframes the whole thing:

"the market is pricing AI as a technology cycle when its actual anatomy is that of a credit driven real estate cycle, which is precisely why the 08 mechanics apply."

The reasoning is uncomfortably good. Technology cycles fail slowly, through obsolescence and competition. Property cycles fail mechanically, because of construction lag: "long construction lags that guarantee supply arrives after demand has turned." Walk down the AI buildout, the argument goes, and every feature is property development in disguise, a building on entitled land, financed with debt against the structure, leased to tenants on long take-or-pay terms. Hence:

"This is not a software business that happens to own servers. It is a real estate business that happens to own compute."

And the kill shot on timing: "Real estate cycles break the same way every single time, not when demand collapses. It rarely does. But when the rate of demand growth decelerates against the fixed supply, the boom has just finished building." Note what that implies. You don't need AI to disappoint. You only need the rate of growth to slow while the buildings keep completing.

The concentration risk is the other half: "The hyperscaler has in economic substance extended a concentrated infrastructure credit facility to tenants with no independent operating income. If those tenants default, the backlog evaporates into non-cash impairments." Every landlord in this complex is exposed to a handful of counterparties.

The arithmetic supports the worry. On The Financial Exchange Show, "AI Spending Puts Microsoft and Meta on the Spot", Chuck Zotta of Armstrong Advisory Group estimated the hyperscalers will spend $2.5–3 trillion on capital projects between 2027 and 2029 while generating around $400 billion of annual profit, and will carry $400–500 billion a year in depreciation charges, enough to wipe out current profitability "unless they generate an additional trillion dollars in annual AI sales." His colleague Paul Lane noted Google is spending roughly $490 million per day on AI against $39 billion of quarterly cash coming in.

You can see the strain in the financing. Alphabet raised its 2026 capital spending guidance to $195–205 billion from $180–190 billion and went free-cash-flow negative (by $5.9 billion) for the first time in two decades as a public company, per The Rundown, "Tesla Tanks As AI Gamble Gets Pricier, Alphabet's CapEx Spending Rattles Investors". Meta went to the bond market for $12.5 billion at a 7.53% yield to fund Texas construction, per Wall Street Unplugged, "BlackRock and Meta's $12B bond deal changes the AI landscape". And on Prof G Markets, "How Big Tech Offloaded The Risk Of AI", the discussion was about how much of this is being pushed off balance sheets entirely, through special purpose vehicles and private credit funds. When 7.5% bonds and off-balance-sheet vehicles are funding the buildings, the "credit-driven" part of the bear thesis stops being a metaphor.

Where I come out.

Both sides are arguing about the same fact, an enormous, debt-financed construction boom, and disagreeing only about whether the buildings are scarce or abundant. This week the operators had the better evidence. Power's signings are booked, his contracts run 15 years, and his tenants are investment-grade. Kurian confirmed a hyperscaler is paying market rent today because it can't build fast enough. And the queue data says new competing supply physically cannot arrive quickly. The bear case is a statement about 2028 and 2029, when the Louisiana and El Paso campuses finish, when depreciation lands, and when Meta's rental business either has anchor customers or doesn't. It is a very good argument about the back half of this cycle. It is not this quarter's argument.

One honest caveat: the bulls this week were mostly operators with something to sell, and the bears were mostly commentators with no operating exposure. Weigh accordingly in both directions.

Names in Play

Digital Realty (DLR): the clear winner of the week, and the only data center landlord with a genuine operator voice out there. Record signings equal to a full prior year, $2-billion-plus backlog on 15-year contracts with investment-grade tenants, raised full-year guidance, stock up 25% year to date. The bull point is the latency argument: the workloads in his buildings can't relocate to cheaper land. The bear point is the one he conceded himself: the equipment supply chain is "as tight as it's ever been," which means his own development pipeline is slow and getting more expensive. Watch whether development yields hold as the cost per megawatt keeps climbing. (Power Lunch)

Meta (META): now on both sides of the ledger, which is new. As a tenant it is enormous and still growing. As a would-be competitor it just hired Amazon's former cloud chief to rent out spare capacity. As a credit it looks worse than it did a month ago: free cash flow down 91% to $784 million, operating margin down from 43% to 31%, $12.5 billion of 7.53% paper issued. Next catalyst: whether the compute-rental business names an anchor customer. If it does, every colocation landlord has a new low-cost competitor with 50-billion-dollar campuses already under construction. If it doesn't, this was a distraction from a capital spending problem. (Halftime Report, Bloomberg Tech, Morning Brew Daily)

Alphabet (GOOGL): the tenant that matters most to the landlords right now, because it is the one openly writing rent cheques. Cloud growing 82% with a $514 billion backlog and record 35.6% margins is the demand case. Capital spending at $195–205 billion with negative free cash flow is the reason the stock fell 7% anyway. The specific thing to watch is Cramer's unanswered question: does Google renew the third-party leases when its own capacity comes online, or hand the space back? (Squawk on the Street, The Rundown)

Southern Company (SO): the most interesting utility structure of the week, discussed below.

Not in play, and worth saying out loud: the tower REITs, American Tower, Crown Castle, SBA Communications, generated no operator or analyst commentary at all this week. Neither did Prologis, Rexford, First Industrial or EastGroup individually, nor Equinix. The wireless carriers were effectively silent on capital spending. We are not going to manufacture views on those names from nothing.

Read-Throughs

Utilities: Southern Company has written the playbook everyone else will copy. On Squawk on the Street, "10AM Hour: New Coca-Cola CEO's First Broadcast Interview, UPS CEO on Earnings, Southern Company CEO on Data Center Buildout 7/28/26", CEO Chris Womack, an operator, with more than 9 million electric and gas customers and a brand-new 25-year power supply agreement for an OpenAI project outside Savannah, described how he defused the political problem before it started. Household rates in his territory are "frozen now through the latter part of this decade," and the data centers are paying for that privilege: "they're paying even a premium that allows us to hold these rates."

The contract mechanics are the part worth stealing. Womack listed them: "making sure you're not passing costs on to existing customers, making sure you're getting collateral payments, making sure there are minimum bills." Collateral payments and minimum bills mean the data center pays whether or not it uses the electricity, which is precisely how a utility protects itself against the scenario Womack himself acknowledged: "We know there are going to be efficiencies at some point in the future." He expects "double-digit growth" into the next decade. Compare his position to New York's and you can see why capital is moving south. Womack's own take on the backlash was that "it shouldn't be political," and that operators need to talk more about "closed-loop systems, liquid cooling," the technologies that address the water complaint.

Electrical equipment and turbines: the shortage is now the business model. Womack confirmed the input cost squeeze from the buyer's side: "there's a great demand for turbines. There's a great demand for wire. There's also a great demand for labor," and Switched On documented how manufacturers are exploiting it: 20% prepayments, proof of permits and financing, and delivery slots stretching to 2030. On pricing, fuel cells run "3k, 4k per kilowatt of power installed" against roughly "$2,200, $2,400 per kilowatt" for gas engines. Vertiv came up twice as the way to play data center electrical and cooling equipment, on Schwab Network, "The Big 3: GOOGL, AMZN, VRT" and Stock Club, "Two Overlooked AI Stocks Worth Knowing About", though neither discussion produced hard order or backlog figures.

Batteries and "bring your own capacity": a new market forming in plain sight. Haid's point is that if a hyperscaler must supply its own power to get approved, and utility-scale batteries are six to nine years away in the constrained regions, the only fast option is thousands of small commercial batteries bundled together. "One, one-megawatt CNI site doesn't matter to a hyperscaler. But 1,000 one-megawatt CNI sites matters a lot." He expects this to be "a very, very, very quickly growing sector" over the next 36 to 48 months. Worth the honest counterpoint from host Shayle Kann of Energy Impact Partners, who pushed back that the queue is long precisely because it is full: "there are so many storage projects in front of you in the queue. So we're going to get a ton of that stuff online, too." Scarcity today is not scarcity forever.

Industrial and logistics: the sector is splitting in two, by building size. The one substantive read on warehouses came from Xander Snyder, a real estate economist, on America's Commercial Real Estate Show, "CRE Mid-Year Outlook: The New 10+ Year Cycle, Flat Yield Curves & NOI Strategies". He is "pretty optimistic about industrial" long term, but describes right now as "a stabilization period," with construction that peaked at record levels in 2022 still being delivered and creating "a bit of oversupply in certain regions."

The useful part is the bifurcation. Demand has shifted "away from very large leasing of very large properties, 250,000, 300,000 square foot and above," partly because trade policy uncertainty makes companies unsure where to put their logistics hubs, and partly because "there are just fewer tenants that can lease very, very large spaces like that." His Midwest work found big-box vacancy around 10% while "smaller properties under 50,000 square feet have very low vacancy rates, closer to like 3-4%." That is a meaningful distinction for anyone comparing the big-box landlords against the small-bay infill owners. Long term he still likes it: e-commerce is only "something like 20% now of all retail sales," and growing.

Freight: one genuinely encouraging data point. On the same Squawk on the Street episode, UPS CEO Carol Tomé, an operator, flagged a turn in the trade lane that matters most to her margins: "What we were most encouraged by is the growth that we saw in our China to U.S. trade lane in the second quarter. We hadn't seen that growth for a year. That's our most profitable trade lane. So we're seeing that business come back." Import volume recovering on the China–US lane is an early signal for warehouse demand near the ports, and it is the first positive freight read we've had in a while.

What Changed

Digital Realty's story reversed completely in four weeks. At the start of July, DLR was paying a full price (a 6.5% capitalisation rate, meaning roughly 6.5 cents of annual income per dollar of purchase price) for occupied Northern Virginia campuses, diluting shareholders to do it, and the stock fell 5% on the news. This week the same company reported record signings, raised guidance, and is coming off its best week in nearly four years with the stock up 25% on the year. Nothing about the assets changed. What changed is that the market decided owning already-powered, already-leased space is worth paying for.

Meta's "spare capacity" pivot got the opposite reaction to last time. In early July, the idea that Meta might sell excess AI compute was read as bullish, a sign of discipline and a new high-margin revenue line, and the stock rose 9%. This week, with the same strategy now formalised under a senior hire, the stock fell 9%. The market has stopped reading spare capacity as prudence and started reading it as evidence of overbuilding. For landlords, the important shift is that this is no longer a slide in a deck. There is now an executive whose job it is to compete with them.

Political risk stopped being a talking point and became an executive order. A month ago the data center backlash was a matter of community meetings and local zoning friction. Now the fourth-largest state in the country has frozen approvals for a year and says it is building a template for others. Ben Reitzes of Melius Research, on Squawk on the Street, described the industry response bluntly: "people are racing to the red states with less red tape." He also cited a Morgan Stanley observation that "social backlash to data center development and political uncertainty into the 28 presidential election are causing hyperscalers to start building sooner," meaning the political risk is accelerating construction, not slowing it, as developers rush to get permits before the window narrows. That is a genuinely new dynamic, and it argues for more supply arriving sooner, which is the bears' point.

The cost to build keeps climbing, and it is now confirmed from three independent directions. Dhinsa put it at $10–15 million per megawatt "and growing." Womack, buying the inputs, confirmed upward pressure on turbines, wire and labour. Power, building the shells, called the supply chain "as tight as it's ever been." A developer, a utility and a landlord all describing the same squeeze is about as good as corroboration gets. When the cost to replace an asset rises, the value of an existing one rises with it. That is the strongest structural argument for the incumbent landlords, and it got stronger this week.