To chase superintelligence, Meta plans city-size clusters
Meta is pledging hundreds of billions of dollars to build massive AI data centers, underscoring how infrastructure is becoming central to the race for artificial general intelligence (AGI).
In a memo to employees last month, Meta announced Superintelligence Labs, a new division led by former Scale AI CEO Alexandr Wang and GitHub ex-CEO Nat Friedman. On Monday, the company unveiled plans to build multiple multi-gigawatt clusters, including one called Prometheus, slated to go online in 2026, and another called Hyperion, which Meta says could eventually rival the footprint of Manhattan.
“These agentic models aren’t lightweight—they require deep compute, long-context training and runtime adaptability,” Shobhit Varshney, Head of Data & AI at IBM, told IBM Think in an interview.
As AI systems expand beyond narrow tasks to broader reasoning and planning, the demands on computing infrastructure are increasing, Varshney said. “AGI” refers to AI capable of performing a wide range of intellectual tasks at a human level. “Superintelligence” refers to a hypothetical form of AI that surpasses human abilities across all domains. And Meta’s announcement suggests it is serious about building the large-scale infrastructure it says will be needed to support these developments.
“For Meta, chasing superintelligence means more than keeping pace,” Varshney said. “It means building a supercluster that supports the next form factor of intelligence, one that lives in your glasses, your assistant and your digital self.”
While Meta builds for a superintelligent future, most companies are still struggling to make practical use of today’s AI, Gartner VP Analyst Kjell Carlsson told IBM Think in an interview.
“The problem isn’t model capability,” he added. “It’s that we don’t yet know how to use them intelligently.”
Source: Sascha Brodsky Tech Reporter, Editorial Lead, IBM
IBM - Think newsletter

Comments
Post a Comment