Meta plans massive AI data centers requiring 1,000+ megawatts each

Meta CEO Mark Zuckerberg announced plans to build multiple massive data centers dedicated to AI training, each requiring over 1,000 megawatts of electricity and covering significant portions of Manhattan if placed in New York City. The ambitious infrastructure push represents Meta’s aggressive bet on developing “AI superintelligence,” with the company investing “hundreds of billions of dollars into compute” to compete in the race for artificial general intelligence.

What you should know: Meta is constructing several multi-gigawatt data center clusters, each far exceeding the power consumption of current supercomputers and AI facilities.

• The first cluster, called “Prometheus,” will come online in 2026 and consume multiple gigawatts of electricity.
• A second facility named “Hyperion” will scale up to 5 gigawatts over several years.
• Zuckerberg confirmed Meta is building “multiple more titan clusters” beyond these two initial projects.

The big picture: This infrastructure expansion reflects how aggressively major tech companies are racing to build AI that promises to surpass human intelligence.

• For perspective, the world’s leading supercomputer El Capitan uses only 30 megawatts of power.
• Elon Musk’s xAI “Colossus” data center in Memphis consumes an estimated 150 megawatts, though that’s expected to grow.

Why this matters: The scale of Meta’s investment raises questions about both the financial viability and environmental impact of the superintelligence race.

• Meta has been offering compensation packages exceeding $100 million to poach top AI researchers from companies including OpenAI and Apple.
• According to research firm SemiAnalysis, Meta is preparing to build natural gas plants to power the Prometheus data center in Ohio.

What they’re saying: “We’re actually building several multi-GW (gigawatt) clusters,” Zuckerberg wrote in a Facebook post. “We’re calling the first one Prometheus and it’s coming online in ’26. We’re also building Hyperion, which will be able to scale up to 5GW over several years.”

Key details: Each data center will feature hundreds of thousands of AI-focused GPUs and require power equivalent to what entire cities consume.

• Zuckerberg noted that “just one of these covers a significant part of the footprint of Manhattan.”
• The facilities represent a massive escalation from Zuckerberg’s January announcement about building a single Manhattan-sized data center.