Meta to spend hundreds of billions dollar to build AI data centres

Meta founder Mark Zuckerberg has announced that the company will invest hundreds of billions of dollars in building massive AI-focused data centres across the United States.

One of the first, a multi-gigawatt facility named Prometheus, is scheduled to launch in 2026. According to Zuckerberg, one of the sites will be nearly as large as Manhattan, which spans 59.1 square kilometers (22.8 square miles).

Meta is heavily investing in the development of what it refers to as "superintelligence" — AI systems that could surpass the intellectual capabilities of the most intelligent humans. In 2024 alone, Meta generated over $160 billion in revenue, primarily from online advertising.

Posting on his platform Threads, Zuckerberg revealed that the company is constructing several large-scale AI data clusters. One of these, called Hyperion, could eventually reach five gigawatts in capacity and will be located in Louisiana, with full deployment expected by 2030. Prometheus, meanwhile, will be situated in New Albany, Ohio.

“We're building multiple titan-class clusters. Just one of them will span a footprint comparable to a large portion of Manhattan,” Zuckerberg said, noting that the data centres have been given names reflecting their massive scale and ambition.

Karl Freund, lead analyst at Cambrian AI Research, told the BBC that Zuckerberg appears determined to become a leader in AI through massive investment. He added that Meta’s AI talent will have access to some of the most advanced hardware available.

Following the announcement, Meta’s stock rose by 1%, with shares up more than 20% since the start of the year, according to Reuters.

Currently, there are over 10,000 data centres worldwide powering cloud infrastructure, most of them in the U.S., followed by the UK and Germany.

However, AI data centres require vast amounts of energy and water. A study projects that they could use up to 1.7 trillion gallons of water globally by 2027. Even a single AI query—such as one made to ChatGPT—can consume as much water as a small bottled drink.