SitePower

The Interconnect

Issue 01  ·  May 2026

Money stopped being the hard part. The scarce thing now is the one almost nobody is pricing.

The Read

May settled an old argument: money is no longer the hard part. In a single fortnight the market produced a retail-accessible data-centre REIT, a $10B AI-infrastructure fund, a $3.25B commingled vehicle, and enough off-balance-sheet mega-JVs to make $27B look routine. Capital didn’t just show up — it industrialised.

And yet the largest transaction of the month wasn’t a data centre at all. It was a ~$67B utility merger, framed openly as a play for power access in Northern Virginia. When the biggest cheque of the month buys a power company, the message is hard to miss: the constraint has migrated from capital to electrons.

That’s now the consensus — secure the power footprint first, the real estate second. Microsoft banking ~3,200 acres in Wyoming, Amazon ~1,300 in Bastrop County, Equinix contracting three-quarters of a gigawatt of advanced nuclear, AWS locking a 17-year, 1.92 GW power purchase agreement at Susquehanna. Power-first is no longer contrarian. It’s the playbook.

Which is exactly why it’s worth saying the uncomfortable thing: if you think power is the binding constraint, you’re already one cycle behind.

Capital is being committed, and generation contracted, far faster than anyone can secure the right to build and interconnect. The signatures are racing ahead of the permissions. St. Charles, Missouri voted to exclude data centres from every zoning category — an effective citywide ban. Harford County, Maryland fast-tracked an emergency one. Utah’s “Stratos” megasite — entitled for up to ~9 GW — cleared its approval and promptly ran into organised local opposition. None of this made the headlines the way the nuclear deals did, and that’s the tell.

Because the nuclear-for-AI story everyone celebrated in May is the right story at the wrong horizon. The contracts are real — Equinix’s ~774 MWe, Terrestrial Energy and Riot co-locating up to 4 GW, Oklo and Vertiv integrating reactor steam straight into cooling. But the delivery dates don’t land inside the crunch. The first commercial onshore small modular reactor is China’s Linglong One, this year; Western nth-of-a-kind sits twelve to eighteen months behind that as a leading indicator, not a deliverable. The megawatts that actually energise 2026 and 2027 are unglamorous: existing nuclear PPAs, on-site natural gas, and demand flexibility.

So here is the position I’d defend in the comments. The binding constraint of the next two years is not gigawatts of nameplate generation. It’s the permission bottleneck — the throughput of permitting, interconnection, and community consent that turns a signed deal into an energised megawatt. Capital is solved. Power is being solved on paper. Permission is the part almost nobody is pricing. The winners of this cycle won’t be whoever contracts the most nameplate. They’ll be whoever can convert permission and interconnection into delivered power on the demand timeline — and they’ll quietly pay a premium for any site that already has it.

The Month in Power

If you track five moves, track these.

NextEra moves on Dominion, ~$67B. The largest US energy deal since Exxon–Mobil in 1998, openly a bet on Data Center Alley. The supply, not the building, is the prize — though completion is far from settled, with close odds quoted anywhere from 50% to 75%.

A BlackRock-led consortium to take Aligned, ~$40B. GIP, the sovereign vehicle MGX, and the Nvidia/Microsoft partnership buying ~5 GW across 50 campuses. The largest data-centre transaction on record — and confirmation that sovereign capital is now a structural buyer, not a tourist.

Blackstone brings BXDC toward the public market, ~$1.75B. The first retail-accessible, pure-play AI-data-centre REIT — a new and permanent class of buyer for stabilised, hyperscaler-leased assets.

Blackstone and Google stand up a $5B TPU joint venture. A hyperscaler ceding majority equity to fund compute-as-a-service. When the money flows into chips rather than walls, capital — not silicon — is the binding input.

Meta’s “Hyperion” goes off-balance-sheet with Blue Owl, $27B. Two gigawatts scaling to five, financed through private credit. The template for any project north of $10B.

From the Desk

A pattern, anonymised.

This month every developer in the market will tell you they have “secured” power. Most haven’t. The single most useful question a diligence desk can ask is what “secured” means on paper. A signed PPA, a filed interconnection position, a docket number — those are secured. A constructive conversation with someone at the utility or the regulator is not. We see permission-by-conversation constantly: a verbal opinion cited as though it were a permit. It isn’t — and it routinely costs six to eighteen months the model never priced.

Two adjacent tells travel with it. The first is the favourable revision: challenge a number and it moves in the developer’s direction every single time — the price suddenly higher, the offtake suddenly bigger. Three of those in one conversation and the model is aspirational, not conservative. The second is the data-centre anchor that appears late in a pitch with no NDA and no named tenant — a hyperscaler logo dropped in to justify the economics. In a month when everyone is racing to claim power, the discipline that matters most is telling a filed position from a hopeful one.

The Call

Stop optimising for nameplate. The scoreboard everyone’s watching — gigawatts contracted, dollars raised — measures the part of the problem that’s already being solved. The scarce thing is delivered power on the demand timeline, and that runs through permission and interconnection.

Watch the permissions, not just the megawatts.

— Nigel Broomhall, SitePower

Get next month’s issue.

One read a month on where power, capital, and sites actually meet. No noise.