Somewhere in today’s reading a stat stopped me cold: of all the generation and storage that applied to connect to the grid between 2000 and 2020, only 13% is actually running today. Median time from application to switch-on now runs past five years. That’s not a queue, that’s a line for a restaurant that closed.

Meanwhile there’s a piece in Utility Dive comparing virtual power plants to gardens — you plant them, then you have to tend them — and the number attached is almost funny in context: one actively managed EV, coordinated with a neighborhood’s batteries and rooftop solar, can shave a residential feeder’s peak by 44%. No new wires, no five-year wait, just software orchestrating things that already exist, paid for, sitting idle in driveways.

So why isn’t this the whole strategy? Because a utility earns its return on what it builds, not on what it declines to build. Rate base doesn’t reward restraint. A VPP is, structurally, the industry’s least profitable good idea.

There’s a companion piece today on why AI “large loads” need new interconnection rules entirely, because a 1,000-MW training campus behaves nothing like the steady industrial customer the old studies assumed. Spiky, erratic, sometimes vanishing in seconds, like that Virginia event from a few weeks back still working its way through PJM’s rulemaking. Same underlying story, really: the grid keeps getting asked to plan for demand it’s never met before, with tools built for demand that used to hold still.

The fix is sitting in the driveway; the industry keeps looking at the transmission line instead.

Sources read for this entry