Newsletter · · Ashutosh Agarwal
Transformers and Local Backlash Become the Real Ceiling for Data Centers - Data Center Landlords & Power - Week of August 27, 2026
Data Center Landlords & Power for the week of August 27, 2026. Podcast synthesis on Vertiv guiding to 14 billion dollars with a 45 percent second half, a Wall Street Journal tally of roughly 3 trillion dollars of AI commitments sitting off balance sheets, hand-built grid transformers running to five years and 14 million dollars each, and a field operative calling 71 percent community opposition the new floor.
Data Center Landlords & Power
Week of August 27, 2026: Transformers and Local Backlash Become the Real Ceiling for Data Centers
Everything around the data center landlords got very loud this week: an equipment maker raising its numbers again, a Wall Street Journal tally of $3 trillion in spending that never shows up on a balance sheet, the biggest data center lease ever signed, and a transformer shortage that could decide who gets to build at all. The question quietly shifted this quarter, from "can they lease the space?" to "can they power it, and will the neighbors let them build it?"
TL;DR
- The picks-and-shovels are still booming. Vertiv's CEO stood by a $14 billion revenue guide for this year and 37% growth, with second-half growth accelerating to 45% and a backlog he called himself "fully confident" in. When the company selling the wiring is that busy, the demand behind the landlords is real.
- The scary number is the one you can't see. A Wall Street Journal analysis says the big tech firms have roughly $3 trillion of AI spending commitments sitting off their balance sheets, leases not yet started and chip purchase orders, on top of the ~$600 billion they already report. Meta's giant Louisiana campus is 80%-owned by a private-credit firm and its $27 billion of debt never touches Meta's books.
- Power, not money, is the ceiling, and now it's specifically transformers. Big grid transformers are hand-built, cost up to $14 million, and can take five years to get. Demand for them could double by next year. Meanwhile community opposition to data centers has gone bipartisan, with one field operative saying 71% opposition is now "the floor," not the ceiling.
What's new
The clearest read on demand this week came from the company that wires the landlords' buildings. On At Barron's, "Vertiv's CEO on Nvidia, AI, Growth, and More", Gio Albertazzi, CEO of Vertiv (VRT), an operator, and about as close to the real order flow as you can get, since 80% of Vertiv's business is the power and cooling gear that goes inside data centers, laid out numbers that only make sense if the buildout is still accelerating:
We have guided a $14 billion top line for this year... we are in this moment at 37% year-on-year growth... 45% year-on-year second half. Fully confident in the backlog and our ability to execute on that.
Backlog here just means orders already booked but not yet delivered, money in the pipe. Two details matter beyond the growth rate. First, who's buying: not just the household-name hyperscalers and the "co-locators" (developers who build data centers and rent them out), but a "rapidly accelerating" new group Albertazzi calls neoclouds, companies like CoreWeave whose entire business is renting out AI computing. That's a widening customer base, not one bet. Second, his framing of the power problem. He refuses to call it a ceiling:
Access to power seems to be like creating a ceiling that the industry would not ever be able to break. But in fact, we see that a lot of power behind the meter is being developed. So again, what seemed to be a constraint has become just a pacing item.
"Behind the meter" means generating your own electricity on site instead of waiting for the utility grid. His new industry mantra, "tokens per second per megawatt," is a tidy way of saying every hard-won unit of power now has to be squeezed for maximum output. When the arms dealer is this confident and this specific, it tells you the tenants behind the REITs are still spending.
The most important number this week is one that doesn't appear on any balance sheet. On The Banker Next Door, "Big Tech AI spending has gone off the rails!", host Dr. Joseph Bergquist, a commentator rather than an insider, walked through a Wall Street Journal analysis that put a jaw-dropping figure on the hidden part of the AI buildout: roughly $3 trillion of commitments across nine top tech companies that don't show up as debt or capital spending, because accounting rules let them sit off the books until a lease begins or a chip is delivered.
The mechanics matter for anyone who owns a landlord. Leases that haven't started yet total about $904 billion (versus $248 billion of leases already on the books), and chip and equipment purchase commitments run to $1.52 trillion (versus $356 billion of reported long-term debt). Alphabet alone has $811 billion of purchase commitments as of June, up from $332 billion just three months earlier. And the single cleanest illustration is Meta's Hyperion campus in Louisiana, "the size of about 1,700 football fields":
Though Meta is the builder, neither Hyperion nor the $27 billion in debt that's financing its construction show up on Meta's balance sheet. Funds managed by the Wall Street firm Blue Owl Capital own the majority of a joint venture that, in turn, owns the campus.
Meta owns 20% of that venture, Blue Owl 80%, and Meta's lease payments (initial commitment ~$12.3 billion, a four-year term starting in 2029) are what pay the bondholders. This is the "credit-driven real estate cycle" argument the bears have been making, except now it's a Wall Street Journal footnote with a dollar figure, not a theory.
The biggest data center deal ever announced this week is also the clearest picture of how circular the financing has become. On Factor This, "This Week in Cleantech (08/21/2026) - Data center backlash is cross-partisan", the hosts flagged that OpenAI signed a 20-year, 10-gigawatt lease in Southern Ohio with SoftBank's SB Energy, "the largest data center project announced to date." The eye-opener is the credit support: NVIDIA is backstopping the completed facility to the tune of up to $105 billion, so the developer can borrow against it without lenders sweating whether OpenAI walks away. NVIDIA also becomes the exclusive chip supplier for the first half of the campus and is taking a $1.5 billion equity stake in the developer. As co-host Mike Casey of TigerComm put it, dryly:
NVIDIA, it's not just de-risking someone else's bet. It's also lining up hundreds of billions in chip sales into the same campus and has talked about financing OpenAI's purchase of that hardware too.
For context, that $105 billion backstop is actually the smaller version: NVIDIA had reportedly floated a ~$250 billion all-up guarantee earlier in the summer before investors balked and the stock fell 5%. The chip vendor financing the customer that buys its chips is the definition of a circular loop, and it's now the structure behind the marquee deal.
Power is the binding constraint, and it has a very specific bottleneck: transformers. The same Factor This episode surfaced a New York Times piece with a genuinely alarming supply-chain fact. The big grid transformers, the truck-sized units that step high-voltage electricity down for local use, are "mostly bespoke and hand-built and can take up to five years and cost as much as $14 million each," and the US manufactures only 20% of what it buys. Replacing the aging fleet and powering new data centers "could double the demand for large power transformers by 2027." Batteries are stuck in the same jam: roughly 750 gigawatts of storage projects are waiting in the interconnection queue (the line to plug into the grid), and the median wait has stretched from 1.5 years in 2015 to 5 years in 2025, mostly because the grid gear to connect them isn't available. The developer's view was even starker. On Motley Fool Hidden Gems Investing, "America Is 10x Behind China in AI Infrastructure", the CEO of ExoWatt, an operator building solar-plus-storage power for data centers, gave the scale of the gap:
In the US, we're adding about 50 gigawatts of power to the grid per year... they need to cover a gap that's in the order of 70 to 100 gigawatts at a minimum... China adds 540 gigawatts of power to the grid per year. So more than 10 times our capacity.
He also captured how desperate tenants have become for any power at all: where customers once wouldn't bother with anything under "300, 500 megawatts a gigawatt," now "any power you can give me, I'm going to take it, even if it's in small quantities," because even 10 to 100 megawatts is worth "hundreds of millions of dollars." For a landlord that already controls powered, connected square footage, that scarcity is the whole moat.
Community opposition to data centers stopped being a local zoning nuisance and became a structural, bipartisan movement. This is the read-through that could reprice development pipelines, and it came from the same Factor This episode. Molly Taft of WIRED, discussing her piece "How Data Centers Broke American Politics," described opposition that cuts across left and right, populists like Marjorie Taylor Greene and Thomas Massie on one side, climate activists on the other, uniting against the same projects. And Mike Casey, whose firm is literally on the ground doing community engagement in a county that voted 67% for Trump, put a number on it that should worry anyone with a speculative build pipeline:
Your average community anywhere in America starts at 71% opposition... 71% is not the ceiling. It's the new floor.
His warning to developers: the old playbook of "sneaking this past with a bunch of public meeting notices attended by three half-asleep senior citizens" is dead. One recurring theme from the right, he noted, is that people increasingly see data centers as "surveillance machines." Permitting risk just got a lot more expensive to underwrite.
The debate
The bull case: scarcity is the moat, and this week it hardened into physical objects. The bull argument has always been that whoever already owns powered, leased, connected space owns something that literally cannot be reproduced fast, and this week the constraints got very concrete. Five-year lead times on hand-built transformers. A 750-gigawatt battery queue with a five-year median wait. A US grid adding 50 gigawatts a year against a 70-to-100-gigawatt shortfall, per the ExoWatt CEO. And the demand behind it isn't softening: Vertiv raised into a 45% second half, and Liz Ann Sonders, Chief Investment Strategist at Charles Schwab, argued on Excess Returns, "The Rally is Broadening. The Earnings Growth Isn't." that hyperscaler capex is headed to "something like $900 billion next year," and, crucially, that unlike the late-1990s telecom bubble, this time "the build is associated with current demand, and it's undershooting current demand." If demand is running ahead of supply and supply is physically capped, the incumbent landlord wins.
The bear case: it's a debt-financed property cycle, and the financing just got a lot more visible. The bear argument sharpened this week precisely because the numbers came out from behind the curtain. $3 trillion of off-balance-sheet commitments. Meta's Louisiana campus 80%-owned by a private-credit firm with $27 billion of debt Meta doesn't report. Alphabet, Amazon, and Meta all running negative free cash flow, meaning they're now spending more than their businesses bring in and funding the gap with borrowing. And the counterparty concentration is glaring: on Barron's Streetwise, "Oracle and Oil Refiners. Plus, Pyramid Scheme or nah?", Adam Levine, senior technology writer at Barron's, pointed out that Oracle's headline ~$640 billion backlog rests on a very narrow base:
Almost half of that is one contract with OpenAI. And so the short answer to the question is, the company has become in some very large degree a proxy for OpenAI... people worry about whether OpenAI will fulfill their end of a $300 billion contract, because they don't have $300 billion right now.
The most memorable bear framing came from the freight side. On Freightvine, "Zac Rogers, CSU and Dale Rogers, ASU | 10 Years of the Logistics Manager's Index", Zac Rogers, the Colorado State economist who co-authors the Logistics Managers' Index, reached back to the dot-com fiber glut:
We will overbuild. I mean, the incentives to be the first or be one of the remainders is just too strong. It reminds me of 2000, 2001, the global crossing... So someone will lose a lot of money, but the data centers, they're not going to, they're going to be used. And so... there will be a reckoning. It's going way too fast.
Note the nuance even in the bear case: overbuilding doesn't mean the buildings sit empty forever, "demand will always catch up with overcapacity, it just takes time." The pain is financial, borne by whoever financed the wrong building at the wrong moment, not necessarily physical vacancy.
Where I come out. This was a genuinely two-sided week, which is rare. The bulls had the operating evidence: Vertiv's raised guide, the transformer and power scarcity, Sonders saying demand is undershooting supply. The bears had the financing evidence: $3 trillion off the books, negative free cash flow, Oracle-as-OpenAI-proxy, and a respected economist openly invoking Global Crossing. The honest read is that both are describing the same fact from opposite ends: a colossal, increasingly debt-financed construction boom whose demand is real today and whose financing structure is fragile tomorrow. And one caveat worth stating plainly: the true operator with skin in the game this week (Vertiv) was bullish, while most of the bears were commentators reading balance sheets. Weigh accordingly, in both directions.
Names in play
Two names had genuine, thesis-moving guest commentary this week.
Vertiv (VRT), the operator voice, and it was bullish. A $14 billion revenue guide, 37% growth, a second half accelerating to 45%, and management "fully confident in the backlog." The bull point is that Vertiv sits upstream of every landlord and hyperscaler, so its order book is a leading read on data center demand, and it's still climbing, with neoclouds a new growth leg. The bear point is valuation and volatility: the stock has swung from $70 to $370 and back toward $270 in a year, and any plateau in the buildout hits the picks-and-shovels names first and hardest. Next thing to watch: whether the "timing disruptions" management flagged really do get absorbed in the second half, as promised. (At Barron's)
Oracle (ORCL), a cautionary read-through for every landlord exposed to a single mega-tenant. Down ~25% year-to-date after being up ~30%, with negative free cash flow and a backlog that's roughly half one OpenAI contract. Adam Levine's line, that the company is now "a proxy for OpenAI," is the concentration risk the whole colocation complex quietly shares. Next catalyst: any sign of whether OpenAI can actually fund a ~$300 billion commitment. (Barron's Streetwise)
Read-throughs
Power and grid equipment: the shortage moved from megawatts to hardware. The binding constraint this week wasn't "is there enough electricity" in the abstract, it was the physical gear to move it. Transformers at up to five years and $14 million each, with demand possibly doubling by 2027; a 750-gigawatt battery backlog; PG&E quoting nearly four years just to get circuit breakers. In New York, Con Edison now requires battery developers to help pay for grid upgrades, which added ~$21 million to the average project and has already triggered at least 25 cancellations. One genuinely investable angle surfaced: solid-state transformers, which swap hand-wound copper for modular electronics. Heron Power (founded by ex-Tesla executive Drew Baglino) just raised $140 million to build a US factory capable of 40 gigawatts a year, and hyperscalers are lining up as first customers because NVIDIA has already redesigned its data center blueprints around them. (Factor This)
Behind-the-meter power: a useful contrarian voice. Most of the coverage treats on-site generation as the industry's escape hatch. On Catalyst with Shayle Kann, "Can AI revolutionize grid operations?", Josh Wong, CEO of Think Labs, an operator on the grid-software side, pushed back. He agrees "tens of gigawatts of data centers now... are being planned with behind-the-meter generation" as a bridge to a grid connection, but he doesn't think the grid is actually full:
I believe the grid has enough existing latent capacity to connect the majority, if not all, of the data centers today. We just need to find where they are and improve the overall processes of planning and operations.
If he's right, some of the scarcity that underpins the bull case is a planning failure, not a physical wall, and planning failures eventually get fixed. Worth holding in mind before treating today's power shortage as permanent.
Industrial and logistics: the warehouse is quietly the safe asset again. The best read on warehouse demand came from the Logistics Managers' Index. Warehousing utilization hit 69.4 in June, the highest in about four years, up from 42.9 last December, a 26-point jump in five months. Zac Rogers offered a telling reason that ties straight back to this newsletter's theme: commercial real estate owners can sell offices, malls, warehouses, or data centers, and with offices and malls weak and data centers facing backlash, "warehousing of all their products is kind of the safest one." Capacity is especially tight for smaller operators (a 42.9 reading versus 49 for large firms), suggesting the squeeze is real at the small-bay, infill end of the market. (Freightvine)
Freight: data centers are now propping up an otherwise soft trucking market. On Freight 360, "Freight Market Reality Check (Dean Croke of DAT)", Dean Croke of DAT, the freight-data analyst, described a market kept alive by the buildout: he was "shocked at the number of carriers... who are servicing data centers. It's propping up. It seems half the industry right now." The flatbed and open-deck spot market that hauls transformers, racks, and steel to construction sites has been "underpinning the spot market for almost two years." His caution is the mirror image of the bull case: "if the data centers truly do stop and demand isn't replaced with demand somewhere else, that could have a significant impact." Separately, he flagged a structural shift in trucking costs: new English-language and drug-testing rules are permanently shrinking the driver pool, so "this is no longer a cycle... a structural movement in the floor price to run a truck." Read that as sticky cost inflation for anyone shipping goods, including the 3PLs that lease the warehouses.
What changed
The backlash graduated from one governor's executive order to a nationwide, bipartisan floor. A month ago, the freshest political-risk data point was New York's one-year data center moratorium, a single blue state drawing a line. This week the framing is bigger and more durable: opposition is cross-partisan and populist, and a field operative doing on-the-ground community work in deep-red counties says 71% opposition is now the floor, not the ceiling. This is no longer a story about one governor; it's a structural headwind to every speculative development pipeline, red state or blue.
The off-balance-sheet financing went from a vague worry to a $3 trillion Wall Street Journal number. Last month the concern about hidden leverage (special-purpose vehicles, private credit, risk pushed off balance sheets) was largely conceptual. This week it's quantified: ~$3 trillion of commitments off the books, Meta's Louisiana campus 80%-owned by Blue Owl with $27 billion of debt Meta doesn't report, Alphabet's purchase commitments up from $332 billion to $811 billion in a single quarter. The bear thesis didn't change; the evidence got a lot harder.
The demand signal came from the arms dealers, and it's still green. A month ago the bullish demand read came from a REIT operator (a record-signings quarter). This week the read-through is intact and arguably stronger: Vertiv raised into a 45% second half with full confidence in its backlog, and a leading strategist argued capex is heading to $900 billion next year while still undershooting demand. When the companies feeding the landlords keep raising numbers, the safer inference is that the demand is real.