You're welcome!
The life expectancy is a really important point, and one I hadn't taken into account in my calculations! The thread at https://news.ycombinator.com/item?id=26231021 talks a bit about this. Laptop Li-ion batteries typically last only a few hundred cycles, but presumably you can do better than that if that's what you're optimizing for; Sir Bearington claims 1500 to 2000 discharge cycles, which sounds vaguely plausible and matches the 5 years you're suggesting, but I'd be interested in a deeper dive. Do you know the name of the Hawai‘i project?
I think US$111 billion over 5 years works out to US$22 billion per year, not US$111 million per year, which is probably just past the "manageable but not trivial" level.
I mean, suppose California needs 45 GW of power at the worst moments, as the CAISO report linked above suggests, and you want to provide that with nuclear power at a ballpark 01970s figure of US$1.50/watt (rather than the current contentious figures of around US$7/watt). Further, let's suppose that USA nuclear power's typical 90% capacity factor https://en.wikipedia.org/wiki/Capacity_factor#Nuclear_power_... is not something you can control (for example by scheduling refuelings to not coincide with summer) and that opex is zero. So you need 50 GWe of nameplate capacity, which hypothetically costs US$75B using 01970s practices and wages. If we simply divide by 30 years (rather than using debt financing and calculating an IRR) we get US$2.5 billion per year. At modern USA nuclear construction costs of US$7 per peak electric watt we end up with US$17.5 billion per year, which would still be less than the US$22 billion cost of replacing US$111 billion of batteries every 5 years. Debt financing makes these numbers a little better, but not a lot, and it reduces the cost of short-term projects like 5-year batteries a lot more than it reduces the cost of medium-term projects like 30-year nuclear plants.
(Does that seem backwards? At 5% interest, an infinite-lifetime annuity can be purchased for 21 years of its yearly earnings. If the annuity expires after 30 years, like a nuclear plant, the NPV is reduced to 16 years' worth of earnings, and if it expires after 5 years, like a battery, it's worth 4.5 years of earnings. So if you debt-finance your nuclear plants you end up paying 46% of your money to the bank and only 54% to GE or whoever builds the plants, while if you debt-finance your 5-year-lifetime batteries you pay 91% of your money to Panasonic and only 9% to the bank. So a 5% yearly discount rate drops the value of nuclear-plant joules by about 40% because most of them are so far in the future.)
Right now, batteries and other storage are only used a couple of hours a day at most, rather than the 22 hours or so that add up to my "million MWh" figure, so the battery price is actually an order of magnitude lower. (And we can expect that when we're just paying the cost of recycling old batteries into new batteries, rather than the cost of mining all that lithium, the price will go down.) So, right now, Li-ion batteries are far more economical than nuclear plants, but to bear the base load over days of low sun, either:
· Li-ion batteries will have to get substantially cheaper;
· wind, other forms of generation, or other storage technologies like lead-acid and compressed-air storage would have to be able to supply most of the load, as gas and coal do at present;
· some of the batteries would have to be borrowed from other uses (such as electric vehicles) that pay for most of their cost;
· demand response and efficiency improvements will have to be able to cut demand by a factor of two or three; or
· some combination of these.
For example, you could very plausibly imagine a conjunction of Li-ion batteries getting 25% cheaper than the US$111/kWh (US$31/MJ) price cited above, down to US$83/kWh (US$23/MJ); wind supplying 25% of power demand when the sun is down or clouded; drawing on an additional 30% "spinning reserve" of parked Teslas feeding energy back to the grid; and demand response reducing off-peak power usage by 25%. So our 45 GW of power demand drops to 33.8 GW from demand response; parked Teslas supply 23% of that, and wind supplies another 25% of it; and so the demand on the utility-scale storage plants is only 18 GW. 24 hours at 18 GW is 430 GWh, which costs only US$36 billion rather than the US$111 billion cited above.
We can hope to do a lot better than that, but it seems like an eminently achievable set of improvements.