How much land mass would need to be covered by solar panels to power this future AI infrastructure. Yes, I'm implying that solar would be impractical, but I'm also genuinely curious.
Your implication is misguided, solar is in fact the most practical way to add more electricity for most countries.
The US generated an additional 64Twh of solar in 2024 compared to 2023. To get the same amount from nuclear you would need to build 5 large reactors in one year.
As for land mass, we can re-use already spent land mass, like rooftops, parking lots, grazing farmland and such. Solar can also be placed on lakes.
So for the foreseeable future there is no actual need for new land to be dedicated to solar.
Since America is so in love with car infrastructure, just turning open parking lots into covered lots would be more than enough.
Just converting all Walmart parking to covered solar would meet almost half of all US electrical demand.
4,070,000,000,000 kWh US electric use in 2022
Using 330W panels it would require 8,447,488 panels (4,070,000,000,000 kWh / (330W * 4 hours/day * 365 days/year)) which is 164,726,016 sq ft at 19.5 sq ft per panel.
Walmart has 4,612 stores in the US, averaging 1,000 parking spaces per store, and 180 sq ft per parking space (does not include driving lanes, end caps, etc.) giving us 830,160,000 sq ft.
Yes, hence my use of the word "dispersion". Over a wide enough area, the brine shouldn't have a noticeable impact on sea life. But concentrated release can be really damaging.
Even though I think solar is impractical as a primary source for various reasons, it doesn't take a lot.
David MacKay in "Sustainable Energy: Without the Hot Air" did a calculation circa 2010. To fulfill the world's energy needs back then, a 10 km^2 area in the Sahara desert would be sufficient. Even if you scaled that to 100 km^2, it's absolutely tiny on a global scale, and panels have only become more efficient since then.
The challenge of course is storage and distribution, but yeah, in terms of land area, it's not much.
I was curious about this number, so: 10 km^2 is 10mil square meters, Googling suggests that the theoretical maximum energy captured by a square meter of solar panel is well under 0.5 kW, so well under 12 kWh per day. Say 10 kWh for neatness. Then multiplying by 10mil gives 100mil kWh. More Googling suggests that 10 TWh is a comfortable lower bound for daily world energy usage, but 100mil kWh is 0.1 TWh.
So maybe 1000 km^2 is more like right order of magnitude. That's still tiny, about Hong Kong-sized. Even 100000 km^2 is about South Korea.
3,000-4,000 acres per GW of production capacity in the US Southwest. According to AI :)
Considering how little use there is for most of that land anyways, it seems like a good option to me.
Also AI training seems like the perfect fit for solar. Run it when the sun is shining. Inference is significantly less power hungry, so it can run base load 24/7.
> Also AI training seems like the perfect fit for solar. Run it when the sun is shining. Inference is significantly less power hungry, so it can run base load 24/7.
If you're talking about just not running your data center when the sun isn't out, that effectively triples the cost of the building+ hardware. It would require a hell of a carbon tax to make the economics of this make sense.
> Inference is significantly less power hungry, so it can run base load 24/7.
All major AI providers need to throttle usage because their GPU clusters are at capacity. There is absolutely no way inference is less power hungry when you have many thousands of users hammering your servers at all times.
Furthermore NVIDIAs 80% profit margin makes idling your biggest capital expense a huge ROI problem. Google and Apple should have a big advantage in this regard.
If the balance between capital outlay and running costs was more balanced - then optimising the running cost becomes a big line item on the accounts.
