← All tech signals
Infrastructure

The New Compute Ceiling

Electricity, not silicon, is now the hard limit on how fast AI can scale, as data centres outpace grids that were built for decades of flat demand.

What is it?

AI’s real bottleneck has stopped being chips and become electricity — interconnection queues in dense markets now run four to seven years, and a boom risks breaking up PJM, the largest US grid operator. Whether power constrains growth, or growth reshapes power, is the open part.

Why now?

Grids were engineered for decades of flat demand, then one industry asked for gigawatts at once. Developers now bypass queues, chasing dedicated generation instead, so power security has replaced land and fibre as the first site filter. The constraint is physical, not financial — transmission can’t be built overnight.

Who’s building it?

Hyperscalers securing dedicated power, utilities rewriting rules, and regulators watching interconnection queues stretch further.

Hyperscale operatorsUtilities · PJMGrid regulatorsIndependent developersTransmission planners
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — Hyperscalers and independent developers competing for scarce power capacity.
Buyers — AI labs and enterprises whose compute plans now hinge on grid access.
Gatekeepers — Utilities and regulators who decide who gets a connection first.
What’s at stake
A two-tier build-out taking shape — hyperscalers negotiating dedicated power deals, everyone else queued years behind.
PJM’s structural integrity, strained by demand nobody planned transmission for.
Who ultimately funds new transmission capacity, and on what timeline.
What’s unsettled
Whether utilities can rebuild planning fast enough to match AI’s pace.
If stranded generation stays a workaround, or becomes standard practice.
Who ultimately absorbs new transmission costs — ratepayers, developers, or some negotiated split between them.
02 · How the demand matureda progression, not a timeline
Signal
Interconnection queues stretch from one year to seven, forcing a rethink of siting.
Workaround
Developers pursue dedicated or stranded generation rather than wait in the public queue.
Bifurcation
Hyperscalers with leverage secure long-term power deals; smaller operators face years-long handicaps instead.
Structural shift
Utilities begin treating gigawatt-scale AI demand as its own regulatory class.
03 · Where the market goesthe forks, not a prediction
→Utilities adapt or fracture — PJM’s breakup risk shows regulatory structures built for flat demand may not survive this load.
→Stranded and dedicated generation becomes standard practice everywhere, quietly sidestepping the public interconnection queue for good.
→Nuclear and other firm power get recommissioned specifically to serve AI’s growing appetite for it.
→Household electricity pricing gets entangled with decisions made inside corporate infrastructure planning.
The question underneath it all is whether grid capacity is genuinely fixed near-term, or just under-planned — that decides whether it’s a hard ceiling or a squeeze.
04 · The Infrastructure Supercycle deskadjacent markets, moving now
The Atom, Recommissioned
Nuclear restarts to feed AI’s power appetite.
A sixteen-billion-dollar restart backed by Microsoft and Amazon.
Forbes
The Plumbing Problem
Cooling now runs on a closed loop, not mains.
Nvidia’s new data centre design targets closed-loop cooling.
Fortune
Compute’s Water Bill
Water joins power as a hard AI constraint.
AI data centres draw water equal to 1.3 billion people’s worth.
UN News
05 · The Wire
01
PJM seeks 6.8 GW of backstop power after a data-centre-driven shortfall
Utility Dive ↗
02
Large grid connections now take seven to ten years in mature hubs
Brookings Institution ↗
The panel on this signal
They agreed on
–Power, not chips, is now the binding constraint on how fast AI infrastructure can actually scale.
–A two-tier market forms around who secures dedicated power capacity.
–Utilities and regulators, not AI companies, will set the real pace of build-out.
The competition economist disagreed on
–Grid capacity constraints have historically resolved through utility investment cycles once demand signals become clear enough to justify capital expenditure; whether the current shortfall is a lasting structural ceiling or a planning lag utilities eventually close remains unsettled.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.