Egregoros

Okay so here's the question: Do these DCs plan on renting space and filling capacity over time, or do they not consider their work "complete" until they are filled with hardware owned by the DC owner.

If they're "AI datacenters", then one would presume the latter.

The low number is 65mn square feet. So the question is how much sub-10nm silicon (square meters) is required per square foot of datacenter?

And then the second question is how many square meters of sub-10nm silicon is the world producing?

Okay so I talked to the slop machine.

Post-2023 datacenter buildout is about 10GW (they measure in power consumption rather than floor space)

A rack of Nvidia GPU machines uses about 500kW, so 10GW of buildout is about 20,000 racks. Interestingly, even if you consider that each rack requires 50 square feet of space for "support
infrastructure", 10GW is still only 1 million square feet in total.

One million square feet is 17.4 football fields.
The Texas Tesla Gigafactory is 10 million square feet.

That's the entire AI datacenter buildout since 2023.

The reason why EVeRy CoMmUniTy nEeDs A DatAcEnTeR is (I'm guessing) because they use enough power that they need to colocate wherever power is cheap and available. But the actual needed building space is fairly small.

And nobody is really talking about floor space, but I'm going to guess there's a lot of "for future use" empty space in these buildings. So it's kind of what I said - empty buildings.

If there's 200 post-2023 completed projects (GPT's estimate), that's about 100 racks per project, each rack is technically 2 feet wide, so maybe 5 isles, 40 feet long (?) Basically the size of the hunting section in a Walmart - plus of course a bunch of power and cooling bullshit, but not a "massive" amount.

This picture seems to show 96 racks, so ALL post-2023 completed DC capacity is about 200 times that.