Nvidia and openAI are talking about building a 10GW data centre.
So. We gots a big number. Let's look at it.
How much is 10GW? Well the UK has a typical day time electricity consumption of ~40GW. Meaning that this one single data centre is looking to use the same amount of electricity as a quarter of the UK. In a single installation. A quarter of the electricity of the 6th largest economy in the world. A single data centre.
Or put another way. It needs power from 3x sizewell C nuclear plant.
So how could we power this site?
Let's start with solar. Assuming it's not being built in a very sunny desert (which with water requirements, would be an utterly fucking stupid place to build it). Let's assume 1KW of sun per square meter. (Gonna ignore the whole night problem here). And a 20% efficiency for the panels. With everything else being "optimum". That's 50,000,000m² of panels. That's 50km². But that assumes perfect weather... And ignores night. And terrain...
2/n
So it's fair to say, this ain't gonna be solar powered. So what about wind? Well here the news is actually good. Taking the UK as an example (I know the data better than anywhere else, I know the DC isn't going to be here, I'm just using it for numbers). We actually have 70GW of offshore wind in various stages of construction. Spread variously across the north sea and other coastal waters. 15MW wind turbines are a norm now. Ignoring capacity factors, that's only 667 of these turbines...
3/n
667 15MW wind turbines would take up rather a lot of space. Fortunately theres a lot of sea for offshore wind to be installed in. The UK is actually doing this. But it's not a small install. And requires the DC be not too far from the sea. But we're ignoring capacity factors and what happens when the wind doesn't blow. The whole reason the UK wants so much wind is that it's unlikely for Scotland and Kent to be equally not windy. It's about resilience.
This 10GW DC isn't gonna use wind power
4/n
We established that this would need 3 Hinckley point C or sizewell C size nuclear plants, each if which takes many years to build and cost tens of billions. (No I don't want a debate on the rights and wrongs of nuclear here). Realistically this is the way that DCs can be powered without totally thrashing the atmosphere. Geoff Bennett has a great talk on the usefulness of nuclear for DC power. I recommend a watch.
5/n
What about gas? This is likely what they are going to use, it's just the quickest way they can deploy the amount of power needed. But even then, the largest gas power plant in the world is 8.8GW according to Wikipedia. So still not big enough to power this one DC. And the CO2 emissions would be mind boggling (literally, I'm really struggling to calculate a number for this). Geoff's talk explains what's happened with Musks 150MW DC and it's trailer mounted gas generation kit. And that's tiny
6/n
And this is before we get into the whole cooling and water usage of a facility this size. I need to double check the maths, but this is the sort of scale of facility that will change weather patterns in the area. It is that huge. If it's using a quarter the electricity of the UK. Just imagine how much water that's gonna be. And you just know they're gonna want to build this in a desert.
All so some manager doesn't need to write a full email.
Ye fucking gods.
7/7
@quixoticgeek "Since 2021, Texas has added nearly 40 gigawatts of new clean power generation, enough to comfortably power all of New York State."
from https://reader.foreignaffairs.com/2025/09/09/the-coming-electricity-crisis/content.html
Quixoticgeek
spara
Ariaflame
arclight