- Meta’s first multi-gigawatt AI datacenter cluster, Prometheus, will go live next year
- Hyperion will follow, with plans to scale it up to a staggering 5GW
- Company revenue is strong, but Capex rises
Meta’s CEO Mark Zuckerberg has confirmed plans to build several multi-gigawatt AI datacenter clusters to fuel our growing and insatiable demand for artificial intelligence, with the first cluster, Prometheus, set to go online in 2026.
Another cluster, Hyperion, will follow, with Meta planning to scale this one up to 5 gigawatts over the course of several years.
More so-called titan clusters are also in the works, with the CEO sharing a visualization in a Facebook post showing how Hyperion covers the equivalent of most of Manhattan.
Boasting about the company’s progress, Zuckerberg shared: “SemiAnalysis just reported that Meta is on track to be the first lab to bring a 1GW+ supercluster online.”
The Facebook post summarizes how Meta will combine massive compute power and elite talent to advance superintelligence, backed by hundreds of billions of dollars in compute infrastructure investments. Meta’s new Superintelligence Labs promise to have the highest compute per researcher in the industry.
Despite consistent growth, with last year’s revenue hitting $165 billion, Meta’s spending spree in the world of enhanced intelligence continues. The company’s 2025 capital expenditure was raised to an estimated $64-$72 billion, but investors seem to be slightly more cautious about when these massive AI investments will become profitable.
With competition heating up against OpenAI, Google, Microsoft, and Amazon – among others – Meta aims to bring the first gigawatt-plus AI cluster online with next year’s Prometheus.
Speaking about the company’s most recent quarterly report, Zuckerberg wrote: “AI continues to transform everything we do.” He highlighted five key areas of long-term investment and opportunity: improved advertising, more engaging experiences, business messaging, Meta AI, and AI devices.
