Meta is building supercomputers with over 1 GW of power to develop artificial superintelligence

Meta is building several 1-gigawatt computing clusters. The first, Prometheus, is scheduled to launch in 2026. Another, Hyperion, will be able to scale to 5 GW within a few years. According to Mark Zuckerberg, one of these clusters will occupy an area comparable to Manhattan.

In his post, Zuckerberg said that Meta intends to invest "hundreds of billions of dollars in computing power" to create superintelligence. He also emphasized that the company has enough capital to implement these plans independently, without external funding.

Zuckerberg also cited SemiAnalysis’ analysis that Meta is ahead of its competitors and could become the first company to launch a cluster with a capacity of more than 1 GW. Meta is now even building clusters in special tents to accelerate the deployment of computing infrastructure. It’s all part of the transition from “AI incrementalism” to a full-fledged race for AGI (Artificial General Intelligence).

Meta is also building a "most elite" development team: the company has already attracted leading specialists from OpenAI and Apple, invested in Scale AI, and its CEO Alexander Wang has led the development team.