0:01
AI data centers are being built at such a speed that drawing a comparison to the rail world buildup of the late 1800s feels almost natural.
Search this video
0:10
The amount of money spent is staggering.
0:13
We're talking about billions and billions of dollars.
0:19
At the same time, the term AI data center is kind of misleading because what we're actually talking about are massive AI supercomputers with a power consumption that rivals that of large cities.
0:32
But where are these new AI data centers located?
0:35
Where is construction ongoing?
0:37
And which is the largest one of them all?
0:40
That's exactly what we will find out in this video.
0:43
Our rating system primarily focuses on the installed megawatt, but also takes into account how new the hardware is and if the site is still expanding or not.
0:54
Without further ado, let's take a look at the top 10 largest AI data centers in early 2026.
1:03
Starting at number 10, we have a data center in Memphis, Tennessee that's located inside a former home appliance plant.
1:10
When the Swedish company Electrolux closed their Memphis plant in 2022, it was first acquired by Phoenix investors and later sold to Elon Musk's XAI.
1:20
Now it houses the Colossus 1 data center.
1:24
Repurposing existing infrastructure is a reoccurring theme when it comes to XAI.
1:28
Where in the past or were built, now AI servers are hooking up the latest croc models.
1:34
Right now, Colossus 1 runs approximately 200,000 Nvidia Hopper GPUs and an additional 30,000 Nvidia Blackwell GB200.
1:46
A massive amount of GPU compute and the 300 megawatt power draw is testament to that.
1:52
Going by megawatt only, Polus 1 would rank even higher.
1:55
But the fact that it primarily uses last gen Hopper TPUs and isn't expanding right now puts it at number 10 in our ranking.
2:02
At number nine, we have Abilene, Texas.
2:08
This data center site is part of OpenAI's Stargate project.
2:09
Right now, it already contains about 100,000 Nvidia Blackwell GB200.
2:16
Each of those combines an ARMbased Grace CPU with two Blackwell GPUs.
2:22
They don't come cheap both in actual cost and in power consumption.
2:27
The Aene site already runs at about 200 megawatt, but that's only the beginning.
2:33
As you can see in these satellite images, construction is still ongoing.
2:37
An additional six data center buildings will be ready by mid 2026, which means not only is Ebene already a top 10 site today, but it will rise through the ranks next year.
2:48
For number eight, we are going to Canton, Mississippi.
2:51
That's where the Amazon AWS Mississippi AI data center is located, right next to an already existing Amazon fulfillment center.
3:01
And this one is special because it doesn't use Nvidia GPUs, but runs on Amazon's in-house Tranium 2 A6.
3:06
With a couple of hundreds of thousands Tranium 2 chips installed, the entire site runs at over 300 megawatt.
3:14
But of course, Jeff CH Basos has bigger plans.
3:17
The Mississippi side will ramp to over 1 GW by mid of next year.
3:22
Not only a lot of energy, but also a testament to how good Amazon's own tranium chips are.
3:27
Not everyone needs Nvidia.
3:32
The first Microsoft Azure AI data center comes in at number seven.
3:35
And this time we have to go to Mount Pleasant, Wisconsin.
3:38
Microsoft's first fairwater type campus runs more than 150,000 Nvidia GP200 at over 350 megawatt.
3:45
And Microsoft isn't stopping.
3:49
A path to a buildup in excess of 2 GW is already in the works.
3:56
The numbers just won't stop growing. even in Mount Pleasant.
4:01
Just like number 10, our number six on this list is located in Memphis, Tennessee.
4:05
Colossus 2 is the successor to its little brother we just talked about and another of Elon Musk's XAI projects.
4:12
This is the one with the giant macro hard riding on its rooftop. Yeah, that's Elon.
4:17
Still under construction right now, it already contains over 110,000 Nvidia GP200 at more than 200 megawatt.
4:26
But this is rapidly changing and it's changing so fast in fact that when we started working on this list, we placed Colossus 2 at number eight.
4:33
But looking at the cooling capacity of the data center right before Christmas, it's more likely already at around 350 to 400 megawatt.
4:44
I'm not kidding when I say the AI data center industry is moving fast.
4:50
The entire site will ramp to over 1 gawatt by mid 2026.
4:52
And with how fast Elon Musk and XAI are moving, the numbers will go up even more.
5:00
Number five is another Microsoft site, but this time in Atlanta, Georgia.
5:02
It's another fairwater type data center and very similar to its sister site in Wisconsin we just talked about.
5:08
It also contains more than 150,000 Nvidia GP200 at over 350 megawatt.
5:14
And just like its sister site, there are massive expansion plans.
5:19
A second building is already under construction which will more than double its capacity to over 700 megawatt before the end of next year.
5:30
For number four, we are going to the Amazon AWS site in New Carile, Indiana in spitting distance to Lake Michigan.
5:37
But this site is spread across multiple campuses.
5:39
All part of Amazon's project reineer.
5:42
And these AI data centers are again using Tranium 2 A6, but this time we're talking about half a million chips. Yes, you heard me right. Half a million.
5:52
The whole site consumes about 420 megawatt at the moment.
5:54
But of course, like always, that's just the beginning.
5:59
Another 660 megawatt are already under construction.
6:02
The entire campus will scale to at least 2 GW and potentially much more in the future.
6:10
Before we move into the top three, let's answer the question of how we know all of this stuff.
6:14
Sure, some of the information is public, but the majority of it comes down to in-depth research and for example actually spending money on highresolution satellite images.
6:23
And with that, I mean much higher than the Google Earth images we use in this video.
6:30
The semi analysis data center team tracks over 5,000 data centers worldwide.
6:34
And they do much more than just look at the current state of affairs.
6:38
One of the most impactful insights of the semi analysis data center model is actually understanding the AI infrastructure built out and the competitiveness of companies like OpenAI, Microsoft, Google, AWS, Anthropic, XAI, Meta and so on.
6:53
The data center model has detailed megawatt numbers on a quarterly basis with data points from 2017 and forecasts up to 2032 and that includes all hyperscalers and neocloud providers.
7:07
I don't think there's anything that comes even close in terms of coverage and insights to what the AI data center model from semi analysis offers.
7:15
If you are working in or with the industry and are interested in a highly detailed overview of the current AI data center market, you have to check out the data center model.
7:24
You can find the link in the video description.
7:27
The first entry in our elusive top three goes to the Meta AI data center site in Columbus, Ohio.
7:32
Of course, any top 10 list would be incomplete without one of Mark Zuckerberg's projects.
7:38
And this one is special with over 500 megawatts of installed AI compute, half a gawatt.
7:46
Today, this crazy number is made possible with a mix of different data center types that Meta is using here in Columbus.
7:52
We can find Meta's old age type buildings, but also the new highdensity designs and even AI data center tents.
7:59
Yes, you heard me right, tents, because they are faster to build and time to market is everything in the AI race.
8:07
But remember, it's not crazy if it works, and it clearly does work.
8:13
That's why it's at number three.
8:13
At number two, we have the first Google data center on our list located in Omaha, Nebraska.
8:19
The Omaha site consists of multiple data center campuses and is using so-called multi-data center training.
8:26
But we are not talking about a couple of data centers located right next to each other.
8:31
There is some serious distance between them.
8:33
Omaha is just the closest larger city and used as a denominator, but these data centers are spread around a much wider area and are connected with massive fiber lines.
8:45
There's a lot of new technology involved to make that happen.
8:47
And Google isn't the only one using a multi-data center training approach.
8:50
If you want to know more about how tech giants connect multiple data centers with massive fiber lines to create mega AI clusters, I highly recommend the semi analysis article on multi-data center training.
9:03
It's an important technology to stay ahead of the competition.
9:05
The data centers around Omaha contain hundreds of thousands of Google's in-house TPUs.
9:10
And in Omaha, Google is using the latest TPU models.
9:16
The entire site is already well above 1 GW, of which AI takes well over 500 megawatt.
9:24
Finally, at the number one spot, we have Google's data center site in Columbus, Ohio.
9:29
Just like the Google site in Omaha, the site consists of multiple data center campuses and is using multi-data center training and not all of them are used for AI.
9:38
Other workloads in the Columbus area are, for example, Google Cloud and even YouTube.
9:43
Maybe this very video is streaming to you from this site right now.
9:48
But the data centers that run AI are no joke.
9:50
Google has installed hundreds of thousands of its in-house TPUs across multiple generations.
9:58
The entire site runs at over 1 GW, but again, not all of it is AI.
10:03
The AI part is once again over 500 megawatt.
10:06
As you can see, our top three is closely matched.
10:09
But right now, in early 2026, we place the Columbus cluster just ahead of the Google Omaha cluster and the Meta AI data center.
10:17
And the crazy part is in Columbus, Google is literally located right next to Meta.
10:24
so much AI power and energy consumption in such close proximity.
10:30
The data center buildup is just getting started.
10:33
Right now, our top 10 list starts with Elon Musk's Colossus 1 and ends with Google in Ohio.
10:40
But you heard how close these data centers are, both in installed power and sometimes even in their physical location, especially our top three with three data centers at over 500 megawatt of installed compute.
10:53
The AI data center industry is moving so fast.
10:58
This top 10 list only captures a moment.
11:01
And with how many of these sites are in active expansion, many of them will scale well into the gigawatt range.
11:08
In a year from now, this list will look very different and 500 megawatt won't be nearly enough to make it to first place.
11:15
If you are interested in an actual deep dive into AI data centers, check out our video on how AI data centers eat the world.
11:23
And if you want much deeper insights into the current AI data center industry, including a detailed analysis of every site we just talked about and much more, plus the industry's best forecast of the AI infrastructure buildout, check out the semi analysis data center model.
11:38
It's the best way to get ahead of the curve and understand how and where the entire industry is moving and who is moving fastest.