SitePower

The Interconnect

Issue 03  ·  July 2026

The money is still arriving from everywhere. What turned scarce this month is the right to build the power — and to connect it.

The Read

The binding constraint on this build-out has moved once already this cycle — from land, to capital, to who is willing to hold the capital’s risk. July made the next move unmistakable, and it’s a physical one. The constraint is now power: not the megawatts in the abstract, but the right to build generation and actually connect it. The month’s largest moves led with generation, not real estate — and the whole industry is quietly reorganising around that fact.

Start with the scale. SoftBank committed to a 5 GW AI data centre in northern France worth up to €75B ($85B), with a 3.1 GW first phase — the largest single-site announcement of the cycle, and one that establishes Europe as a genuine mega-campus destination. Read the headline number and it looks like a real-estate story. Read what had to be true for it to happen and it’s a power story: 5 GW is a mid-sized national grid’s worth of demand landing on one parcel.

Then look at how the smart operators are routing around the grid entirely. QTS and Lancium unveiled an 11-building, 465-acre, ~1 GW campus in Hall County, Texas, with a potential self-powered expansion — dedicated solar-plus-battery, closed-loop cooling, non-municipal water. The off-grid model isn’t a sustainability flourish; it’s an engineering answer to a queue. US data-centre demand is rising from 23 GW in 2023 toward 42 GW in 2026 against an interconnection queue exceeding 2,600 GW with 5–7 year waits. When the grid can’t take you for half a decade, you bring your own.

And here is the tell that this is a permitting problem, not a zoning one. New Mexico regulators rejected the gas pipeline serving Oracle and Stream’s 2.5 GW Project Jupiter. Not the campus — the pipeline. The energy approval, not the land use, is what stalled a multi-gigawatt site. That’s the position I’d defend in the comments: energy permitting is now the top site-selection risk for a mega-site, ahead of land, ahead of capital, and the market is pricing it in real time.

None of which is a capital shortage — if anything the opposite. BlackRock closed its record ~$40B acquisition of Aligned and committed a further $5B, inside a vehicle targeting up to $100B including debt; the six largest US hyperscalers are projected to spend ~$700B on data-centre capex this year. The money is deep and getting deeper. The caution flag is elsewhere: AI-related debt is tracking toward ~$570B in 2026 and bond investors have started to push back — the first repricing signal of the cycle. Capital isn’t the gate. The right to build the power is.

The Month in Power

If you track five moves, track these.

SoftBank — northern France, 5 GW / €75B. Confirmed by Jul 24 (was ~80% likely on Jul 15). A 3.1 GW first phase, the largest single-site announcement of the cycle, and the clearest sign that the mega-campus is now led by its generation footprint, not its floor plate.

QTS + Lancium — Hall County, TX, ~1 GW, self-powered. Base 1 GW campus confirmed; the self-powered expansion — solar-plus-battery, closed-loop cooling, non-municipal water — reported and ~65% likely. Off-grid built specifically to sidestep the interconnection queue. Watch whether this becomes the template.

Project Jupiter — the pipeline gets rejected. Confirmed. New Mexico regulators knocked back the gas pipeline serving Oracle/Stream’s 2.5 GW site. The single most important line in the month: energy permitting, not zoning, is now the gating item for a mega-site.

9.8 GW of nuclear now committed to AI data centres across 13 projects. Confirmed — and every major US hyperscaler has now signed a nuclear deal. The supply-side answer to a blocked grid is being written in firm, dispatchable generation, not in more queue positions.

The debt caution: ~$570B of AI issuance, and the first pushback. Recurring through the month’s analysis, not a single transaction. Set against a ~$40B Aligned buyout and a vehicle reaching for $100B, it confirms the shape of the risk: capital is abundant, the exit market for its debt isn’t proven yet. This is the number to watch into the second half.

From the Desk

The question that changed.

A year ago, the first diligence question on a site was where the power was coming from. This month, on the deals crossing our desk, the first question has become sharper: who holds the right to build and connect that power, and is the approval real or assumed. Project Jupiter made the distinction expensive — the megawatts were never in doubt; the pipeline permit was, and the permit is what stalled the site.

“The queue will clear by then” is not a power plan; it’s a hope dressed as an assumption — the same species of flag as the favourable-revision tell we’ve named in earlier issues. Where a developer or a technology partner has a genuine, permitted power path, that answer belongs in the diligence pack from the first conversation, not surfaced under questioning in month three. Where it isn’t there yet, that isn’t a reason to walk — it’s a reason to price the permitting gap honestly rather than assume someone else closes it later.

The Call

Don’t track the gigawatts announced. Track who holds the permitted right to build and connect them — that tells you more about where this cycle is headed than any headline capacity number. If you’re building technology, structure it to shorten time-to-power or move behind the meter. If you’re developing a site, bring a permitted power path to close, not a queue position and a hope. If you’re deploying capital, the opening is real — but underwrite the permitting gap as carefully as you underwrite the return.

Track the right to build the power, not the megawatts announced.

— Nigel Broomhall, SitePower

Get next month’s issue.

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