Energy & Power — Research Frontier

What's genuinely new and where the rung is heading. First compiled 2026-09-24 from three sources and one concept page carried over from datacenters; extended the same day with the LBNL queue census and PJM's capacity auction results.

The binding constraint, named

Connecting to the grid, not generating the power. The national census puts the median wait from asking to connect to switching on at 61 months for plants built in 2025, with 2,061 GW queued and only 13% of past requests ever built (LBNL). The carried-over concept page's 3–4 year figure for PJM and ERCOT load zones sits below that national median (Power: The Binding Constraint). The newest evidence adds a second kind of gate on top of that: a political check in Texas that stopped the queue's timeline without any physical cause (ERCOT notice).

The number this rung owns

$325 per MW-day — PJM's capacity price for 2028/29, cleared 2026-07-14 at its agreed cap, after $333.44 (2027/28) and $329.17 (2026/27), with the region 6,831 MW short of its reliability requirement (PJM 2028/29). It replaces the earlier headline number, the about 7 GW gap between announced and under-construction US data-centre power in Q1 2026, which is still held but at weak evidence (snippet-only ingest) (7 GW shortfall). The price is capped, so it is a floor on scarcity, not a measure of it.

Active Frontiers

1. Political gates on the queue

Status: Early stage Key sources: ERCOT notice, EIA STEO

Texas was the fallback when connecting elsewhere was slow. A governor's verification order made it discretionary. The EIA's forecast five weeks later still named the region the growth leader.

Open problems:

• How much load the verification covers, and how long it takes.
• What the commission granted on 2026-08-20.

2. The queue, counted

Status: Steady Key sources: LBNL Queued Up 2026

About 8,200 projects and 2,061 GW waiting at the end of 2025; the queue shrank 10% but gas grew 86% to 253 GW. Only about 2% of the 2024 queue came online in 2025.

Open problems:

• The matching count for data-centre load queues, which LBNL excludes.
• Whether FERC Order 2023 and the regional reforms shorten the 61-month median. The authors say it is too early to tell.

3. Getting around the queue

Status: Early stage Key sources: Carbon Direct, Tesla Cortex 2

The concept page records gas clearing the queue faster than renewables and on-site generation and storage as a way to skip the wait. Storage shifts load in time; it does not create generation.

Open problems:

• Whether behind-the-meter power changes what a data-centre megawatt costs.

4. What the power costs

Status: Early stage (was Stalled) Key sources: PJM 2028/29

The rung now holds one price: PJM's capacity price, three collared auctions at $325–333 per MW-day, the last two at the cap and short of the reliability requirement. That is the price of being available at peak, roughly $119,000 per MW-year by our arithmetic. It is not the price of the electricity used, and there is still no PPA price or pass-through figure.

Open problems:

• A primary source for a data-centre electricity or PPA price, with a date.
• What PJM's December 2026 auction (2029/30) clears at, and whether the collar is extended.
• How much of the capacity price a data centre actually pays, given self-supply and bilateral contracts.

What this rung does not yet hold

• No price for the electricity itself. Only a capacity price. See frontier 4.
• No national gigawatt figure for data-centre load from a primary source. The EIA gives energy, LBNL excludes load queues; PJM's ~5,100 MW of one year's forecast growth is regional.
• Nothing on nuclear or gas generation deals, energy ownership, or transformer lead times as their own subjects, all named in scope.

Standing

Read it. Nothing on this rung has been measured or run in-house.