Right. Meta wants big enough plants that an AP1000 or two would be the right size. They're known to work. There are four in operation, two in the US, and another dozen or so under construction.

Most of the small nuclear reactor startups hand-wave the failure modes and argue that they don't need the hulking big expensive containment building. NuScale claimed that. They wanted multiple reactors sharing the same cooling pool. If they ever had a leak, the whole set of reactors would be contaminated, even without a meltdown.

If we look at the big reactor accidents so far, there's Chernobyl, with no containment building. There's Fukushima, with too small a containment unable to contain the pressure. And there's Three Mile Island, where a large, strong containment building contained a meltdown. Three Mile Island was an expensive disaster, but not hazardous outside the plant. That's the failure mode you want.

We might be better off at developing better techniques for welding thick sections to make hulking big, strong containment vessels. There's been progress with robotic welding of thick sections.[1]

[1] https://www.agrrobotics.com/trends-s-industry-analysis/roadm...