The US power industry,
built from the ground up

Ask an agent about the US grid and it will happily give you a number. The problem is knowing whether the number is real. So here is the industry built from primitives — four questions, each answered by a cited public dataset, each figure one call away from the raw EIA cell it came from.

The operating fleet is 1.39 TW of nameplate capacity across 27,971 generators. Natural gas alone is 42% of it — the US grid is, structurally, a gas-built machine with a large and growing renewable edge.

Capacity factor is output ÷ (nameplate × hours) — how hard a fuel actually runs. It's the number that turns a stock table into a working model, because every fuel runs at a wildly different intensity.

This is why the stock table lies. Nuclear is 7% of nameplate but ran flat-out to produce 18% of February's electricity. Solar is 12% of nameplate but produced 6%. You cannot reason about this grid in nameplate terms — you have to multiply by how hard each fuel works.

Run the whole fleet through that multiplier and you get what the grid actually delivered in February — 343 TWh, mixed like this:

All of that capacity exists to serve one thing: load. And load is not steady. Across the Lower 48 this June, average daily demand swung from a mild start to a heat event in ten days.

The defining physical constraint sits underneath that line: the grid must balance every second. Electricity isn't meaningfully storable yet, so generation has to equal demand in real time, always. That single fact is why capacity factor matters, why gas is irreplaceable today, and why the entire battery buildout exists — you must carry enough firm, dispatchable capacity to cover a peak that shows up only a few hundred hours a year.

And after two decades of flat US load, demand is now growing — with data centers as the primary new driver, a signal that shows up directly in the queue.

A snapshot can't show direction. For that you need three tiers, in ascending order of how likely each is to actually get built — what's retiring, what's financed and planned, and what's merely requested.

sources EIA-860M (retired, planned) as_of 2026-04 · ERCOT / MISO / CAISO / PJM / SPP queues, as_of 2026-05 to 2026-06 · all sha256-verified

The arrow points the same way across all three tiers: coal out; solar, storage, and gas in. Solar leads on energy, gas leads on firmness, and storage is the wildcard — which is where the whole model resolves.

Storage is being built faster than anything else in the system, because it's the only thing that breaks the every-second constraint. Read these three numbers in sequence:

If batteries scale to long duration — 4–8+ hours, not today's 1–2 — the entire model inverts. Solar's 20% capacity factor stops mattering, because you time-shift it into the evening peak. Gas peakers, the 40%-CF balancers, start to look like stranded assets. "Firm capacity" gets redefined around storage instead of fossil.

That hasn't happened yet. Today's storage is mostly short-duration — good for the evening ramp, useless for an overnight or a multi-day cold snap. Until duration scales, gas stays irreplaceable and coal retirements stay capped by reliability rules.

Meanwhile the demand side is being rewritten in real time. NYISO's load queue — 53 projects, ~14 GW of new demand requests — is the cleanest cited signal in the platform that the data-center buildout is hitting the grid as hard, physical load. No other ISO exposes it this directly; for the rest you infer it from gas requests and load growth.

Watch the storage-queue-to-operating ratio and the NYISO load tab, and you're watching the whole industry's future resolve in real time. That's the model. Four questions, one variable to watch — and not a single number you have to take on faith.