If you have so much wind and solar generation that they spend too much of their lifespans sitting idle and not generating income, they potentially never even pay for their initial build cost and basic maintenance, let alone be a better investment than just buying a treasury note yielding 1%. That's a recipe for long term stagnation and depression.
So yes, lots of solar might make it more cost effective to add some wind and/or batteries and vice versa but that's a good thing and what this model explores.
Bitcoin miners are the key players innovating in that space right now. It's something that can very easily be packaged up in portable containers. I believe Chinese miners for years had a yearly migration to soak up massive amounts of excess hydro power during the rainy season that would have been otherwise wasted.
There might be a place for more generic computation datacenters, but they would have to find a relatively unique customer that has compute loads that can be turned off and on at a moment's notice, and potentially be down for days, weeks, and even months. There's probably a startup to be founded here, if you can figure out how to severely undercut equivalent AWS services to justify the lack of uptime guarantees.
I don't know enough about hydrogen production to know if it can be quickly spun up/down to suck up variable load. I imagine gas transport infrastructure is going to be the biggest blocker.
Who is going to build a hydrogen generator or data center with a 80% idle time, operating only 12-4 PM?
The real solution, like you mention, is more capacity + storage. Pumped storage is my personal favorite, as it is much cheaper than batteries for anything more than 2 hours of power/day. Pumped seawater has a pretty small footprint.
I doubt that data centers will be built, but there are other uses beside hydrogen -> ammonia->urea. Desalination, for instance. Carbon capture, to make money in the carbon credit markets. Even at 4 hours a day, mechanical capture with plausible costs can compete with forests.
Meanwhile, solar output can go from 100% to 10% in three seconds. How is this not a problem?
While your example is solar at 100% exactly meeting demand, and then dropping to 10% in 3 seconds, and the fix for that is: overprovision (and some other stuff).
They do need to have decent predictions of demand and supply, like this graph that shows the 1-hour ahead and 24-hour ahead predictions:
https://www.caiso.com/TodaysOutlook/Pages/default.aspx
To get the whole grid working together, but we already have that, and renewables and batteries help in various ways.
https://www.pjm.com/~/media/committees-groups/task-forces/rm...
An energy storage regulating resource, such as a battery or flywheel, cannot burn fuels to produce electricity, but instead charge and discharge energy from the bulk electric system itself through a power inverter. Energy storage devices can match a desired output within 200 milliseconds of receiving the regulating signal; therefore these resources do not need to model any dynamics, ramp limitations, or dead-bands.
Australia is also using batteries in a similar way:
"South Australian Battery Responded in Just 140 Milliseconds After A Coal-Fired Power Plant Failed"
https://www.sciencealert.com/elon-musk-s-south-australian-ba...
It's an area that's ripe for some software innovation, if they can actually break through the legacy management club that currently runs the grid.
Also, medium term storage is a very, very young industry. Things like redox batteries, flow batteries, molten salt, compressed air, and many others are on the horizon. The better these work the more gas peakers are solely used as backup.