The Grid Doesn't Read Press Releases

Every week another gigawatt campus shows up in my feed. 5 GW in Louisiana. 7.7 GW in West Texas. 9 GW somewhere in the Utah desert.

I've spent over 20 years pulling fiber, building out suites and fighting for power at the meter. Before I get excited about any of these, I ask one question. Who is delivering the electrons, and when?

So I went and counted. As of August, public trackers show about 319 GW of US data center capacity planned or under construction (dcmap.us). About 90 GW of that has actually broken ground. The rest is a rendering and a load request sitting in somebody's utility queue.

For scale, the whole country averaged roughly 506 GW of load last year (4.43 trillion kWh, per EIA). The announced pipeline equals about a quarter of the nameplate capacity of every power plant in America. That isn't going to happen, and honestly it doesn't need to. The part that worries me is the slice that is real.

I pulled every power plant and every 230 kV and up line out of OpenStreetMap and totaled them. OSM tags about 1,193 GW of generation across 15,304 plants. It misses most of the wind fleet, so the real number is closer to 1.3 TW.

Nameplate is the wrong number to stare at. What matters is what shows up at 6 pm on the hottest day of August.

This year developers plan a record 86 GW of new capacity (EIA). That sounds great until you break it out. 43.4 GW is solar and 24.3 GW is batteries. Gas is 6.3 GW. When I run typical peak capacity credits against that slate, I get maybe 25 to 30 GW of firm capacity. Then take away the roughly 46 GW of retirements EIA has on the books through 2029 (EIA Table 4.5).

My back-of-envelope through 2030 grosses data center load up 15% for reserve margin. That's before electrification, reshored manufacturing or anything else that wants power. So treat these as floors.

To be fair, Grid Strategies thinks NERC is overcounting data center load and that most regions make it to 2030 fine (Utility Dive). I hope they're right. They're probably right about the speculative stuff. My bet is the middle scenario.

The gas is in the ground. Land isn't the problem either. The bottleneck is equipment and paperwork.

If you order a heavy-duty gas turbine from GE Vernova today, you're looking at 2031. Their backlog plus slot reservations hit 116 GW last quarter, and the whole company builds about 20 GW a year for the entire planet (OilPrice). Siemens Energy is sitting on 69 GW, Mitsubishi on 35 GW (EUCI). Large power transformers are running around 128 weeks, and substation units are past 160 (Simple Thread). Anybody who has waited on a GSU knows that number is optimistic.