Top 10 largest AI Datacenters in 2026

Watch on YouTube ↗  |  January 06, 2026 at 21:44  |  11:50  |  SemiAnalysis
Speakers
Dylan Patel — Founder, CEO, and Chief Analyst at SemiAnalysis

Summary

The video ranks the ten largest AI data centers by installed megawatt as of early 2026, highlighting historic construction speed and massive power consumption. It covers XAI, OpenAI, AWS, Microsoft, Meta, and Google sites, with Google's Columbus, Ohio and Omaha, Nebraska campuses taking the top two spots. The speaker argues the AI data center buildout is still accelerating, with many sites expanding into the gigawatt range and 500 MW soon insufficient to lead the ranking.

  • SemiAnalysis ranks top AI data centers by installed megawatts, hardware recency, and expansion status.
  • XAI, OpenAI, AWS, Microsoft, Meta, and Google sites are detailed with capacity and chip deployments.
  • Amazon's AWS sites use in-house Trainium 2 chips at massive scale, showing a non-Nvidia path.
  • Microsoft's Fairwater campuses and Meta's Columbus site are among the largest, with active expansion.
  • Google's Omaha and Columbus sites rank #2 and #1, using in-house TPUs and multi-data center training.
  • The top sites already reach over 1 GW total, with AI portions above 500 MW.
  • The speaker expects many sites to scale into the gigawatt range and says the ranking will look different next year.
  • The video promotes the SemiAnalysis Data Center Model for detailed AI infrastructure forecasts.
Ideas
Dylan Patel Founder, CEO, and Chief Analyst at SemiAnalysis 0:01
AI datacenter buildout just getting started
AI data centers are being built at a historic, railroad-like pace with staggering capital spend. The top sites already range from roughly 200 MW to over 1 GW, and many are in active expansion; the top three each exceed 500 MW of installed AI compute. The speaker expects the buildout is just getting started, with many sites scaling into the gigawatt range and 500 MW not nearly enough to lead a year from now.
Dylan Patel Founder, CEO, and Chief Analyst at SemiAnalysis 1:40
Nvidia dominates top AI datacenter compute
Nvidia is the dominant compute supplier across the largest AI data centers. Colossus 1 runs about 200k Hopper GPUs plus 30k Blackwell GB200s; OpenAI's Abilene site has about 100k GB200s; Microsoft's Wisconsin and Atlanta Fairwater campuses each have more than 150k GB200s; and XAI's Colossus 2 already has over 110k GB200s and is ramping. The breadth of Hopper/Blackwell deployments signals strong demand for Nvidia's AI platforms.
Dylan Patel Founder, CEO, and Chief Analyst at SemiAnalysis 2:53
Amazon Trainium scales massive AWS AI
AWS is deploying its own Trainium 2 chips at massive scale. The Mississippi AI data center uses hundreds of thousands of Trainium 2 chips at over 300 MW and will ramp to over 1 GW by mid-next year. The New Carlisle, Indiana Project Rainier site uses about half a million Trainium 2 chips, consumes about 420 MW now, has another 660 MW under construction, and will scale to at least 2 GW. The speaker presents this as a testament to how good Amazon's in-house Trainium chips are and that not everyone needs Nvidia.
Dylan Patel Founder, CEO, and Chief Analyst at SemiAnalysis 3:32
Microsoft rapidly scales Azure AI capacity
Microsoft's Fairwater-type Azure AI campuses are among the largest, with Mount Pleasant, Wisconsin and Atlanta, Georgia each running more than 150,000 Nvidia GB200s at over 350 MW. Microsoft has expansion plans: Wisconsin has a path to more than 2 GW, and Atlanta is adding a second building that will more than double capacity to over 700 MW before the end of next year. This shows Microsoft rapidly scaling AI compute capacity.
Dylan Patel Founder, CEO, and Chief Analyst at SemiAnalysis 7:30
Meta Columbus AI reaches half-gigawatt
Meta's Columbus, Ohio AI data center has over 500 MW of installed AI compute and ranks third. It uses a mix of old data center types, new high-density designs, and even tents because they are faster to build; time to market matters in the AI race, and the approach is working.
Dylan Patel Founder, CEO, and Chief Analyst at SemiAnalysis 8:16
Google leads top AI datacenter rankings
Google has the largest AI data center footprint in the ranking. Its Omaha, Nebraska and Columbus, Ohio sites each consist of multiple campuses using multi-data center training, hundreds of thousands of in-house TPUs, and latest-generation TPUs. Each site runs at over 1 GW total, with AI taking over 500 MW; Columbus is ranked #1 and Omaha #2. This shows Google's ability to scale AI compute with its own TPUs and multi-site training.
Up Next

This SemiAnalysis video, published January 06, 2026, features Dylan Patel discussing AI-SECTOR, NVDA, AMZN, MSFT, META, GOOG. 6 trade ideas extracted by AI with direction and confidence scoring.

Speakers: Dylan Patel  · Tickers: AI-SECTOR, NVDA, AMZN, MSFT, META, GOOG