We operate nearly 900 servers, and I believe the capital of cloud infrastructure is Amsterdam, NL. The concentration of ASNs in Amsterdam is incredible. In contrast, Ashburn, Dallas, and LA seem to lack ASN diversity, primarily being dominated by singular big tech companies. Cities like Amsterdam, Singapore, Hong Kong, Frankfurt, Istanbul, and New York, however, have a greater concentration of smaller cloud hosting businesses, offering more diversity.

The real answer: https://en.wikipedia.org/wiki/MAE-East

I worked as a developer for an enterprise hosting company back in 2005. I took servers out to Ashburn a couple of times.

I had visited many data centers in Michigan but this place was at a completely different level. High level of security and more cages than I had ever seen before. It was like a little town with street signs so you could find your cage. Lots of logos of famous Internet companies.

The fans were so loud my ears were ringing for hours afterward. This was before the common use of hearing protection and I have no idea how people worked in there all day.

MAE East is there.

If you wanted good interconnect to the west coast and the rest of the world, you needed to be in or near MAE East.

This is a 5-year-old article, and it's still very pertinent. This past summer I did a lot of bike-riding around Loudon County surveying trails for the Park Service (volunteer) and it's like riding through a cyberpunk future out there. Endless windowless buildings humming loudly, lots of electrical substations, and wires crisscrossing everywhere. The Washington Post reported that Dominion Power is now running electrical lines deep into West Virginia and bringing defunct coal plants back online to meet the energy demands (1), but a recent independent report found that it won't be enough as energy demands are going to double at a minimum (2).

I live in eastern Prince William County and the spread of datacenters is a hot-button topic here. People living on the Western side of the county, near Loudon, are getting rezoned for data centers, which means dramatic increase to their property taxes. The same report that highlighted energy demands also found that the increased tax revenue and jobs created are only really during the construction of the buildings, once completed they don't take many techs to maintain them.

I feel this last point may underestimate the jobs created because it doesn't consider all the folks connecting to these data centers to do work. For example, Amazon is expanding here to be close to the datacenters and my friends who work for Amazon have to be within driving distance of the campus. I could be overestimating this effect though.

(1) Gift WaPo article: Internet data centers are fueling drive to old power source: Coal https://wapo.st/40A4SBm

(2) 2024 Data Centers in Virginia Report https://jlarc.virginia.gov/pdfs/reports/Rpt598-2.pdf

I know we are in AI cycle and likely need a lot more compute. But if we took out AI from the equation, would we still need that many datacenter , given the amount of compute and storage per server is increasing. By 2030 we should be able to buy a Dual Socket 1024 total CPU Core Server. 2 / 4 PB SSD per 1U space. CPU core getting faster. Even PHP is getting faster.

Surely someday, if it hasn't happened already compute density will outpace compute requirement growth ( Again excluding AI ). And we should have more Rack space than demand? Or is that still too far fetched?

Convenient to CIA HQ and "Liberty Crossing"?

The article states it is from 2019. So not sure if things have changed.

I'm surprised how many data centers there are in Hillsboro, Oregon. And they have more under construction at the moment. I wonder where Hillsboro ranks?

Brings back memories of visiting northern Virginia as a teenager and recognizing all the town names from IRC servers, like Ashburn, McClean, Vienna, and several others.

I saw the VA governor speak at some point in the late '90s, and he touted Virginia "internet capital of the world". I naively thought it was because of all the IRC servers.

The formatting of the article ("Blog Image 3Until recently...") strongly suggests this was copied from somewhere else.

As for the content - I'd say a tl;dr is "first mover advantage" followed by network effects. Assume you wanted to provide a service to the Internet at large, so you went to where you can get the best and cheapest connectivity, and the first movers had all the advantage, at least in a time where latency didn't matter because all end-users had was a 56k modem. And ideally, you went to a place where you knew other big dogs are, because they will have ironed out the kinks, making it less risky for you - in a city where there are tons of existing large names you can expect a way more fault tolerant network than if you go to some small town in the outback where one drunk backhoe driver can take out the entire town's electricity or network.

These days the calculation is different - customers have high bandwidth these days but also they are highly sensitive to latency, so you gotta be as close to your audience as possible, and you build your entire architecture to be fault-tolerant as compute has gotten incredibly cheap.

[dead]

> [...] surpass 1 gigawatt of overall data center capacity.

> [...] with only about half the capacity, at 559 megawatts (MWs) of inventory.

I didn't know that the physical unit for the rate of energy transfer, or more simple just power, is also an unit for computing power. After all it's the same word, right?

</s>