If it costs you money to generate power at a time you cannot sell it, that is losing money. Wind and solar are unique in that they have low marginal costs, whereas a nuclear plant requires more staffing and fuel burn.
I don't understand why you're coming about this problem from a "100% nuclear" situation. That's not the case for the world, and likely never will be the case. But heck, let's assume all you've got to work with is nuclear.
You have a baseload. That's basically the minimum load that always exists. You don't have to worry about selling the power, it's already sold. And then you have peak. Both of these are time / weather dependent but we're still quite able to plan days, if not weeks and months, in advance for what those two values will be. As an example, let's get back to the original topic of the article.
If you're Google and you have a particular datacenter at a location, you know what that baseload is. You know what the peak is. It's a pretty simple calculation to figure out what is most cost-effective here. It's probably even easier for you than the local power company, as base and peak loads likely don't fluctuate much for you outside of HVAC keeping up with weather. It might be to use nuclear for just base load and use the local grid for the rest. It might be to over-produce occasionally because that's still cheaper than buying from the local system operator. Hell, you can probably sell any access, but that's more of a problem than most people think.
But the point is nuclear is king when it comes to baseload power supply. And a datacenter, which consumes a lot of power consistently, is almost entirely baseload.
Also, if you're serious about the problem of "our nuclear power plants are producing more power than we need and we can't turn them off because then we'll have a power shortage", there are applications to dump "bonus" energy into. You'd run into the same problem with renewables too; the difference is that with nuclear, you can avoid the "not enough energy" problem.
We have a version of that in Texas, too. There are some commercial cryptocurrency mining operations that pay for a certain level of power capacity, and if the grid is running low on power, the grid just buys the capacity back from them and they stop mining for awhile.
If you find cryptocurrency mining objectionable for some reason, you could apply the same basic principle to things like aluminum refining or desalination.
You have a baseload. That's basically the minimum load that always exists. You don't have to worry about selling the power, it's already sold. And then you have peak. Both of these are time / weather dependent but we're still quite able to plan days, if not weeks and months, in advance for what those two values will be. As an example, let's get back to the original topic of the article.
If you're Google and you have a particular datacenter at a location, you know what that baseload is. You know what the peak is. It's a pretty simple calculation to figure out what is most cost-effective here. It's probably even easier for you than the local power company, as base and peak loads likely don't fluctuate much for you outside of HVAC keeping up with weather. It might be to use nuclear for just base load and use the local grid for the rest. It might be to over-produce occasionally because that's still cheaper than buying from the local system operator. Hell, you can probably sell any access, but that's more of a problem than most people think.
But the point is nuclear is king when it comes to baseload power supply. And a datacenter, which consumes a lot of power consistently, is almost entirely baseload.