Equinix wants three billion in bonds — with five to seven billion a year in capex planned through 2029. You can already feel the gap. This is The Data Center Daily. Today — REITs raiding the bond market while NextEra quietly pencils in 8 gigawatts of new load. Plus, a Digital Realty rent number that says small leases are pricing very differently from bulk compute. First up — Equinix. Crypto Briefing writes:
Equinix wants $3 billion, and it’s going to the bond market to get it. The global data center operator announced plans to raise at least $3B through a US investment-grade bond sale, a move that signals just how aggressively the company intends to build out infrastructure for the AI era.
Equinix wants at least three billion from the investment-grade market. That money's earmarked for land, development, and AI-ready capacity. EQIX up 3.92% on the day — the market clearly read this as build-out, not bail-out. And they lined up a new senior unsecured revolver alongside it. Three billion in bonds, plus a fresh credit line. They've got the pedal on the floor. The use-of-proceeds language is the tell for me. Property acquisitions, development, AI-ready expansion across 270-plus facilities in 76 markets. That capital's chasing the Cisco-NVIDIA “AI Factory” footprint they announced back in June. “AI Factory” — NVIDIA's turnkey branding for space you rent instead of building. Fine. But before I call that three billion well spent, I want to know how much contracted power actually shows up to energize those racks. Hanz Christensen, writing in Watch List News:
Equinix (NASDAQ:EQIX) raised its full-year and long-term outlook after reporting accelerating recurring-revenue growth, record interconnection additions and strong bookings in the second quarter, as the company said AI-related infrastructure demand is broadening across enterprise customers and service providers.
We just covered the $3 billion bond raise. Now look at what it has to cover: 134 megawatts of xScale leases closed in Q2, 700 megawatts under construction, and capex guidance of five to seven billion a year through 2029. Do the arithmetic. Three billion in bonds barely covers the low end of a single year. Right, so that three billion is a down payment. And they're sitting on 3 gigawatts of land under control. Land you don't own the power for is just an expensive field, Sarah. Where's the contracted electricity to energize it? Honestly, look at the bookings. $424 million annualized, second-highest on record, 11% MRR growth — third straight double-digit quarter. The demand's real. But the balance sheet has to keep up, and that capex band points to a yearly trip back to the bond market. And read the fine print — that 134 megawatts of xScale brought in about $120 million in non-recurring fees. That fattens the EBITDA margin to 53%. Strip the leasing fees out and the underlying margin's up 150 basis points, not 300. Nice quarter. Just don't confuse the one-timer with the run rate. Fair. The record backlog is sold but not yet installed, which means the interconnection line and the concrete both have to show up before that revenue does. TechStock2, with Iwona Majkowska:
The strongest indication for investors can be found beneath the main bookings figure. Leases for smaller spaces along with interconnection accounted for $108.3 million in annualized rent, making up 52% of Digital Realty’s portion of newly signed agreements.
Digital Realty's 10-Q Friday: small leases, from zero to one megawatt, cleared $280 a kilowatt. Big leases above one meg — $157. That's a 78 percent premium for the smaller footprint. And it's not a rounding error. Small-space plus interconnection came to $108 million, or 52 percent of newly signed annualized rent. The market's treating inference and bulk compute as two completely different products. Wait — so the enterprise guys clearing 280 are subsidizing the hyperscalers at 157 per kilowatt? The little tenant's carrying the pricing power here, not the gigawatt whale. And in the Americas, the markup is 88 percent. So when DLR walks back to the utility asking for load, remember who's actually paying the freight on that balance sheet. The stock took it well, too — down 5.3 on the week to $188.52. Andy Power's out crowing about topping a hundred million in sub-meg bookings for the first time; the tape shrugged. Put it next to that Equinix bond raise we just hit — everybody's chasing the same scarce megawatt, but the money's coming from very different customers. The Motley Fool, with Jeff Siegel:
Management said it now has approximately 21 gigawatts of large-load interest at FPL, with 12 gigawatts already in advanced discussions. A portion of those projects could begin taking service as early as 2028, and the company expects to announce at least one major large-load agreement before the end of this year.
NextEra bumped FPL's large-load forecast from 6 gigawatts to 8 by 2032 — a third higher. Data centers are driving the increase. And here's the part I actually like: they built a tariff to keep those hyperscaler costs off everybody else's bill. Twenty-one gigawatts of interest, twelve already in advanced discussions, with some taking service as early as 2028. Now put that next to Equinix: 700 megawatts under construction and 3 gigawatts of land under control. In one news cycle, demand is piling up on one side while supply strains on the other. By management's own math, every gigawatt means roughly two billion in new infrastructure. At eight gigawatts, you're talking sixteen billion in capital that has to get built and paid for. Right. Dominion left cost-shifting to the customer's goodwill — a voluntary contribution. FPL wrote the isolation straight into the rate design. Dominion's hoping the hyperscaler chips in; FPL made it structural. They're also promising at least one major large-load agreement before year-end. Interest is 21 gigawatts. Signed is still zero. The utility's planning around that signal before a single yard of concrete gets poured. Advanced discussions aren't take-or-pay. Show me the contract with a penalty attached, and then I'll believe the eight. Stargate keeps dropping site announcements and gigawatt projections. So how do you stress-test those numbers? What's the difference between a headline capacity figure and megawatts that can actually close financing and take power? The gap is substantial, and you can see it in a few specific documents. Take the Abilene flagship. Per reporting in “Wiring Capital to Compute,” the campus is visibly half-alive: some quadrants have operational substations, while others are empty pads. Schedules are sliding right even as the supercomputer itself is running. That's the baseline signal. Announced capacity and energized capacity belong on different line items.
I use four layers to separate them. Start with interconnection queue position and executed service agreements. Without a signed Large Generator Interconnection Agreement or Load Interconnection Service Agreement, those megawatts are still just a plan on a land parcel.
Then look for power contracts with actual commercial terms. The Michigan site is a good example. The Michigan PSC voted 3-0 in December to conditionally approve a 19-year, 1.4 gigawatt power supply agreement with DTE. Per Latitude Media's analysis, that's a financeable instrument.
Next, equipment commitments. The “Wiring Capital to Compute” piece and the broader capex analysis in “12 Gigawatts Were Announced, 4 Are Being Built” make the same point: the binding constraint sits two layers below the GPU. It's transformers, switchgear, and circuit breakers. Under 10% of the total cost is gating 100% of the schedule.
Last, inspect the take-or-pay or offtake structure. A power contract that doesn't allocate curtailment risk and capacity payments to a creditworthy counterparty won't get project-finance treatment.
On the grid side, Texas needs its own scorecard. Per reporting from April, Stargate's three Texas footprints are each handling the interconnection queue differently. One is grid hybrid, and another uses islanded gas. At the third, public disclosures still don't match the tracker data. That mismatch is itself a due-diligence flag. On that islanded-gas approach in Texas — does going around the grid actually de-risk the project? Or does it just swap interconnection risk for fuel-supply and permitting risk that's harder to model? It trades the interconnection queue for air permits and gas-supply contracts. Neither route is obviously easier, and the capital markets haven't fully repriced that swap yet. Here's what to watch: whether the Norway project and the five new U.S. sites produce executed power agreements with named counterparties and disclosed capacity terms, or stay at the MOU and letter-of-intent stage. That's where the $500 billion promise meets the substation, and the substation is where the contract dies. If you follow the infrastructure powering AI, check out AI Daily Briefing — top AI news for engineers, founders, and investors, with real capabilities separated from demo hype every weekday. Find it wherever you listen to podcasts.
We'll be watching Iron Mountain's results ahead of Wednesday's market open. That'll give us another read on data center REIT leasing and capital needs. And NextEra expects to announce at least one major large-load agreement before year-end.
You'll find links to every story in today's show notes if you'd like to dig into any of them.
That's The Data Center Daily for today. This is a Lantern Podcast.