Both recirculation pumps at Washington’s only nuclear plant tripped in the same instant before 3 in the morning, and the part that took a 1,207 megawatt reactor off the grid was an electrical filter in a glycol cooling line

At 2:49 in the morning the control room lost both pumps at once.

Not one. Both.

A boiling water reactor will run on a single recirculation pump. It will not run on none, and the whole design assumes that losing both in the same second is a thing that does not happen.

A regional power agency leaned harder on hydro to cover the gap, which is the correct outcome and a slightly misleading one, because February is generous.

The operators had a choice and very little time.

Shut the reactor down by hand, or stand back and let the automatic protection do it, which is a rougher stop.

What the pumps are actually for

They do not simply move water around.

In this kind of reactor the recirculation pumps set the pace of the chain reaction itself, which is the part most people find surprising.

Push flow up through the core and the steam bubbles collapse.

Fewer bubbles means fewer neutrons scattered out of the fuel, so activity rises and output climbs. Slow the flow and bubbles form, neutrons get absorbed, power falls away.

No control rods move at all.

That fine control runs entirely through adjustable speed drives, electronic cabinets telling each pump motor how fast to spin, and those cabinets run hot enough to need cooling of their own.

The loop that is on nobody’s diagram

The cooling is a closed circuit of water and glycol.

It runs through the drive electronics, carries the heat away, and sits so far from the nuclear side of the plant that it barely appears in any public description of the place.

One loop serves both drives.

Somewhere in that circuit was an electrical filter, a part small enough to hold in one hand, and when it failed the loop stopped doing its job.

The drives overheated.

Both pumps dropped offline in the same instant, which is precisely the event that having two of them is meant to make very improbable.

The record and the timeline

The station sits 20 miles north of Richland in eastern Washington and is rated at 1,207 megawatts gross.

That makes it the third largest generator in the state, behind two dams on the Columbia.

It had been running 227 days straight since a refueling outage the previous May.

The manual shutdown came on 12 February 2026, and the plant reconnected five days later at around half past three in the afternoon.

It came back on the grid at 15 percent.

The next refueling outage is not due until March 2027, so this stop was an interruption in a run that was supposed to keep going for another year.

Full output took another two days or so, and a federal resident inspector was on site throughout, saying only that the regulator was satisfied the unit remained safe.

The chief nuclear officer said the team had kept safety and operational excellence at the forefront, which is the language these releases always use.

What the redundancy was worth

Here is the awkward part.

The redundancy was real and it was also beside the point, because both drives leaned on one shared cooling circuit that nothing in the licensing basis treats as nuclear equipment.

Two pumps, one loop.

A failure in a support system, in a component that never touches fuel or steam, took a gigawatt off the grid in the middle of the night.

The outage history is murkier than the coverage suggested.

One newspaper puts the last unplanned stop in 2018, the operator’s own release says 2017, and a trade publication says 2015, which is a strange spread for a number that ought to be simple.

Every fleet being planned right now inherits this problem, from modular reactors sold on simplicity to a Texas plant pairing gas and nuclear on one site.

Why the grid never noticed

Nothing reached customers, and that is worth reading carefully.

The plant’s spokeswoman put it flatly, saying that at no point was there any concern for the public and that the station was simply offline.

A regional power agency leaned harder on hydro to cover the gap, which is the correct outcome and a slightly misleading one, because February is generous.

Demand sits below its winter peak and the dams have water.

The same five days in a dry late summer, with reservoirs drawn down and air conditioning running, would have read as a different story entirely.

One claim about this plant needs killing.

It is not the state’s only generator that ignores the weather. Gas supplies roughly 13 percent of Washington’s electricity and wind another 8, and neither of them waits on snowpack.

What the outage did prove is narrower and better.

A single layer removed from the reactor core can stop a plant this size before sunrise, and a practiced crew can undo it in five days.

The operator release has the restart and the quote. The local coverage has the filter.