Mark Zuckerberg says Meta's first AI supercluster will go online next year
Meta is playing catch-up in the AI arms race, spending big on superclusters and poaching top talent from competitors
.jpg)
Craig T Fruchtman/Getty Images
Mark Zuckerberg is going all in on AI. On Monday, the Meta CEO announced plans to bring the company’s first AI supercluster, dubbed Prometheus, online in 2026. And that’s just one piece of what he says will be a multi-year, multi-billion-dollar investment into the company’s AI future.
Suggested Reading
“We’re actually building several multi-GW clusters,” Zuckerberg said in a Facebook post. “We’re calling the first one Prometheus … We’re also building Hyperion, which will be able to scale up to 5GW over several years.”
Related Content
If you’re wondering what a supercluster is, think of it as a giant, hyper-connected network of computers — designed to train and run the kinds of next-generation AI models that will power smart assistants and supercharged search tools. Zuckerberg claims Meta’s clusters will rank among the largest and most powerful in the world, with some stretching across an area “comparable to a significant part of Manhattan.”
Meta’s AI play
The announcement comes as Meta pushes deeper into AI after a bumpy rollout of its Llama 4 models earlier this year, which failed to impress the developer crowd. In response, Zuckerberg’s revamping the company’s approach — building out massive infrastructure and hiring top-tier AI talent through Meta Superintelligence Labs, a new group launched in June.
He’s also been on an AI spending spree, including a $14 billion investment in Scale AI, and plans to keep the momentum going. “We’re going to invest hundreds of billions of dollars into compute to build superintelligence,” he wrote. “We have the capital from our business to do this.”
Prometheus and Hyperion
Prometheus may be Meta's first cluster to launch, but it won’t be the last. The social networking and ad giant says it's already working on other massive builds including Hyperion, which Zuckerberg says will scale to 5 gigawatts over time. These so-called “titan” clusters are being designed to deliver the most computing power per researcher in the industry — a metric Zuckerberg says will be key to staying ahead of the competition.
And that competition is stiff. Rivals such as Google and OpenAI have been racing to develop more powerful AI models, and it’s clear Zuckerberg doesn’t want Meta to continue lagging behind.
With AI shaping up to be the next big tech arms race, Meta’s building infrastructure that could define the next decade of machine learning. Whether Prometheus lives up to its name, however, remains to be seen.