Another way to look at it.. they say they have about 3.14 bil daily active users across all platforms. So that's about 13Wh per daily user session. At $0.15/kWh, that's about $0.002 of electricity which doesn't seem that horrible.
Another way to look at it.. they say they have about 3.14 bil daily active users across all platforms. So that's about 13Wh per daily user session. At $0.15/kWh, that's about $0.002 of electricity which doesn't seem that horrible.
I am also not shocked. But somewhat surprised by the choice of unit in the article.
these are newer and more efficient. so they are likely producing more than the average reactor. So its closer to 1.25 reactors worth of electricity.
but data centers are a HUGE driver of electricity consumption.
Or in the future - "how many SMRs is that?".
I should call it the law-of-something.
I feel reasonably comfortable with the estimated 2-3 reactors.
I could have said 1-4 but...
(there are lots of reactors around 4-500MW too)
2 to 3 is accurate for Canada, where the last nuclear plant to connect to the grid was 31 years ago. For the US, the range is 1.25 to 3.5 (four reactors in the last 30 years). For China, the range is 1.05 to 8 (56 reactors built in the last 30 years, but the 200 and 300 MW reactors are kind of special, so 1.05 to 4 is a reasonable range). For France, 1.15 to 2.0 (five reactors in the last 30 years).
One reactor, if you use a modern design! https://en.m.wikipedia.org/wiki/EPR_(nuclear_reactor)
(Or maybe 2 anyway, to deal with downtime)
My viewpoint (having involvement with many large (>$100M) construction projects, but not power plants): I don't believe the constructability issues with the EPR have anything to do with the size, specifically. I suspect the root issue is design and construction being too separated on the EPRs.
For example OL3 in Finland has to limit its max power because if it trips, the grid needs to shed enough load to stay up: (in finnish) https://www.fingrid.fi/kantaverkko/sahkonsiirto/olkiluoto-3-...
Kind of beside the point of meta using the equivalent of 1-4 nuclear plants of capacity.
Plus ~1/2000th of a server, divided by the life time of a server on their premises in days. Let’s say a millionth of a server. At $1000 a server, that adds about $0.001 a day, if my math is correct. Actual numbers may be higher. What do they pay for their hardware, at the scale they buy it?
Plus installation and management, plus the cost of the building.
I expect it to be even cheaper for example on Iceland with an abundance of hydro power.
[1] They are also trying to redefine steel-processing there and that will use a lot of electricity if they succeed.
That's about 2 in 5 people on the planet. Astonishing.