Meta invests hundreds of billions in superclusters with 5 GW power consumption
Mark Zuckerberg announced unprecedented investments in next-generation AI infrastructure! The company will invest hundreds of billions of dollars in creating server superclusters. Which will consume 1 to 5 GW each. This is power consumption at the level of an entire nuclear power plant!
After freezing the Llama 4 project, all of Meta’s AI directions are now united under the roof of Superintelligence Labs. The company is betting on creating larger-scale models that require not just “more servers”, but real mini-energy systems.
One supercluster called Prometheus is planned to launch as early as 2026. It will be followed by the even more powerful Hyperion, which will consume 5 GW and unite tens of thousands of advanced graphics processors.
The new architecture includes power substations directly on campus grounds, heat recovery systems and industrial-scale liquid cooling. All this will ensure minimal network latency and significantly accelerate model training cycles.
Already this year capital expenditures will reach up to $72 billion and continue to grow. Zuckerberg is confident that the advertising business will bring sufficient funds to cover these risks.
