▲ 20 points
back
15 comments
I'm curious if anyone could expand on the numbers here. Global data centres running at max capacity are estimated to require ~45GW of electricity. That's a lot, but not necessarily for the whole world, so it really depends on the rate of growth, and in particular, its national/regional distribution. For reference, demand on the British electricity grid peaked at 37.5GW today:
Every article I have seen doesn't even provide numbers, which makes me think the claim is totally bogus if they are unwilling to quantify it. Plus there is massive flexibility in training on where and when it is run, they could easily pair it to areas and times of oversupply.
How else are they going to feed you information when they expect you to take rolling blackouts lying down. It's [looks down at notes on current tech thing] the AI from Big Tech causing it all.
https://www.economist.com/technology-quarterly/2024/01/29/da... has some actual data on use.
A recent analysis in Joule, an energy journal, suggests that at the rate they are being produced, ai servers could use about 100 terawatt-hours of electricity each year by 2027, between a quarter and a fifth of total data-centre consumption. Schneider Electric, a French firm that builds power supplies for data centres, projects that ai’s share of data-centre electricity use will jump from 8% in 2023 to 15-20% in 2028.
There an interesting line of thinking i've heard (apologies, i forgot which podcast i heard it on), that states that forecasts of electric load growth due to AI are overblown wildly, since they are based on very recent data that only reflects the early-stage explosive growth, and neglects to consider eventual efficiencies that tend to get built into such activities.
Electrification of other more 'boring' areas like transport, heating/cooling etc are order of magnitude higher at least.
Anyone know if crypto proof of work solving is being lumped in with “AI”?
Wonderful opportunity to kill two birds with one stone: legislatively require that all large scale AI training be powered with carbon free energy only. Not with “carbon offsets” or other scam like bullshit, but by actual, real energy derived from solar, wind, hydro, or nuclear.
Let the price go up. Increase welfare programs to compensate for poor people getting hurt. Charge for pollution at the source. The market will figure it out if you lay the incentive landscape properly.
The hard part is getting 100 million fools to vote for it.
AI is the necessary prerequisite of post-scarcity, which solves all of those problems and then some
Unfortunately for post-scarcity, I am made of matter and I occupy space
The headline is completely backward, it should be "global electricity supply threatens booming AI demand" if anything.
Mostly it's just FUD. Data centers are very flexible about where they can be established. The most generous read on this article is that it's gesturing at issues with capacity and sustainability of the grid, and using AI as a cheap hook to do so.
«the AI industry is forecast to expand “exponentially” and consume at least 10 times its 2023 demand by 2026, said the IEA»
10x in 3 years!
If it keeps growing at this rate AI will soon account for most of energy needs.
Raise prices for non critical consumer electricity supply.
I don't think power companies should be in the business of deciding what workloads are more critical for humans.