From Megawatts to Gigawatts: The 10 Largest AI Datacenters in the World (2026 Edition)

In our last deep dive into the world’s biggest AI factories, we marveled at the speed of xAI’s Colossus and the secretive might of Microsoft’s West Des Moines campus. Back then, a 100,000 GPU cluster was considered a "titan," and power draws of 200 MW were cutting-edge.

That was then. This is now.

Less than two years later, the landscape has shifted violently. We have moved from the "Megawatt Era" to the "Gigawatt Era," where AI data centers consume as much power as major metropolitan cities. Many of the facilities that dominated the headlines in 2024 would struggle to make the top half of today’s list.

However, some giants have only grown stronger. In 2024, we predicted the massive potential of Google's New Albany site in Ohio, estimating it already held over 350 MW of power. That prediction has proven prophetic. The New Albany site hasn't just survived the transition; it has expanded into the Columbus Cluster, which stands today as the largest AI supercomputer on Earth.

Here are the top 10 largest AI data centers globally in early 2026, ranked by installed power capacity and expansion velocity.

Note: This time the job was easier - the data comes from a video pubblished by Semianalysis - Thanks to Semianalysis the time intensive part of scrutinizing satellite pictures to count gensets and drycoolers was already done.

The Top 10 Ranking

10. xAI: Colossus 1 (Memphis, Tennessee)

• Power: ~300 MW

• Hardware: ~200k Nvidia Hopper GPUs + 30k Blackwell GB200s

• The Scoop: Housed in the former Electrolux home appliance plant, this facility was the headline grabber of our previous report. While still massive, it now sits at #10. It primarily runs on last-gen Hopper GPUs and, compared to the aggressive expansion of its rivals, has reached a plateau in growth.

9. OpenAI / Crusoe Energy: Stargate Project (Abilene, Texas)

• Power: ~200 MW (Rapidly Expanding)

• Hardware: ~100k Nvidia Blackwell GB200s

• The Scoop: This site is just getting started. It already houses 100,000 of Nvidia's powerful Grace-Blackwell superchips. With six additional buildings slated for completion by mid-2026, this site is expected to rocket up the rankings later this year.

Openai / Crusoe Energy Abilene's site

8. Amazon AWS: Mississippi AI Data Center (Canton, Mississippi)

• Power: >300 MW

• Hardware: Hundreds of thousands of Amazon Trainium 2 ASICs

• The Scoop: Located next to an Amazon fulfillment center, this site proves you don’t always need Nvidia. Running entirely on Amazon’s in-house silicon, this campus is set to ramp up to over 1 Gigawatt (GW) by mid-2027.

7. Microsoft Azure: Fairwater Campus (Mount Pleasant, Wisconsin)

• Power: >350 MW

• Hardware: >150k Nvidia GB200s

• The Scoop: Microsoft’s first "Fairwater" type campus is a beast, running over 150,000 Blackwell GPUs. Expansion plans are already in the works to push this site past 2 GW in the future, dwarfing the capacity of the West Des Moines facility we tracked in 2024.

6. xAI: Colossus 2 (Memphis, Tennessee)

• Power: 350–400 MW (Estimated)

• Hardware: >110k Nvidia GB200s

• The Scoop: The successor to Colossus 1, featuring giant "Macro Hard" writing on the roof. This site is moving at breakneck speed; initial estimates placed it lower, but recent cooling capacity checks suggest it is already pushing 400 MW. It is expected to hit 1 GW by mid-2026.

Xai's Macrohard Colossus 2. Note: Macrohard is the challenge to Microsoft

5. Microsoft Azure: Atlanta Site (Atlanta, Georgia)

• Power: >350 MW

• Hardware: >150k Nvidia GB200s

• The Scoop: Very similar to the Wisconsin site, this Atlanta facility is already constructing a second building that will double its capacity to over 700 MW before the end of 2027.

4. Amazon AWS: Project Rainier (New Carlisle, Indiana)

• Power: ~420 MW

• Hardware: 500,000 Trainium 2 chips

• The Scoop: Located near Lake Michigan, this site is staggering in volume, housing half a million AI chips. With another 660 MW under construction, the campus will eventually scale to at least 2 GW.

Rainer's AWS / Anthropic Data center. Amazon Project Rainier, the mass-scale data center link-up and unprecedented in-house AI chip deployment to support Anthropic’s Claude LLMs, has reached the finish line.

The Top 3: The 500 MW Club

The gap at the top is narrow, but these three behemoths are currently in a league of their own, each boasting over 500 MW of dedicated AI compute.

3. Meta AI: Columbus Site (Columbus, Ohio)

• Power: >500 MW

• Hardware: Mix of legacy and high-density designs

• The Scoop: Mark Zuckerberg’s team is prioritizing speed above all else. This site uses a mix of building types, including AI data center tents. It sounds crazy, but the "tent strategy" allowed them to deploy capacity faster than anyone else, securing them the #3 spot.

Meta's Columbus Cluster

2. Google: Omaha Cluster (Omaha, Nebraska)

• Power: >500 MW (AI portion) / >1 GW Total

• Hardware: Hundreds of thousands of latest-gen TPUs

• The Scoop: Google uses a unique "Multi-Data Center Training" approach. Instead of one giant building, they connect multiple campuses across the Omaha region with massive fiber lines to function as a single supercomputer.

1. Google: Columbus Cluster (New Albany, Ohio)

• Power: >500 MW (AI portion) / >1 GW Total

• Hardware: Hundreds of thousands of TPUs (Multiple generations)

• The Scoop: Taking the top spot is the validation of our 2024 analysis: Google’s New Albany site. Now encompassing the broader Columbus area, this cluster sits literally right next to Meta’s #3 facility, creating an unprecedented concentration of AI power. Like Omaha, it utilizes distributed campuses tied together to run massive AI workloads alongside YouTube and Google Cloud services. Two years ago, we identified this as a critical hub for Gemini's training; today, it is the uncontested king of AI compute.

  
  

Google data center in New Albany

Key Takeaways from the 2026 AI Data Center Landscape

1. The Gigawatt Era is Here: While 300–500 MW gets you on the list today, the leaders are all actively building toward 1 GW and 2 GW campuses. By next year, 500 MW likely won't even crack the top 5.

2. Custom Silicon Strikes Back: While Nvidia remains dominant (powering Microsoft, OpenAI, and xAI), Amazon and Google occupy 4 of the top 10 spots using their own custom chips (Trainium and TPUs).

3. Speed is the New Currency: Whether it’s xAI retrofitting old appliance factories or Meta building high-tech tents, the companies winning the race are those finding creative ways to deploy hardware now rather than waiting for perfect buildings later.