Meta is unveiling several new computing clusters, each boasting a capacity of 1 GW. The first, dubbed Prometheus, is slated for launch in 2026. Another cluster, Hyperion, will have the ability to scale up to 5 GW in the coming years. According to Mark Zuckerberg, the footprint of one such cluster will be comparable to that of Manhattan.
Zuckerberg also noted that Meta intends to invest "hundreds of billions of dollars in computing capabilities" aimed at developing superintelligence. He emphasized that the company has ample resources to execute these ambitious plans without external funding.
According to an analysis by SemiAnalysis, Meta is outpacing its competitors and could be the first to launch a cluster exceeding 1 GW. The company is even deploying clusters in specialized tents to expedite the development of its computing infrastructure. This is part of a shift from "AI incrementalism" to full-blown races for AGI (Artificial General Intelligence).
Meta is also assembling the "most elite" team of developers: the company has already recruited top talent from OpenAI and Apple, and has invested in Scale AI, with CEO Alexander Wang leading this team.
