Meta To Spend Hundreds Of Billions To Build Ai Data Centres

Meta's laminitis Mark Zuckerberg has said nan societal media elephantine will walk hundreds of billions of dollars connected building immense AI information centres successful nan US.

The first multi-gigawatt information centre, called Prometheus, is expected to travel online successful 2026, Zuckerberg said.

He said 1 of nan sites would screen an area astir nan size of Manhattan (59.1 sq km/22.8 sq miles).

Meta has invested heavy successful efforts to create what it called "superintelligence" - exertion that it said could out-think nan smartest humans.

The company, which has made astir of its money from online advertising, generated much than $160bn successful gross successful 2024.

In a station connected his societal media platform, Threads, Zuckerberg said Meta was building respective multi-gigawatt clusters, and that 1 cluster, called Hyperion, could standard up to 5 gigawatts complete respective years.

"We're building aggregate much titan clusters arsenic well. Just 1 of these covers a important portion of nan footprint of Manhattan," he added.

Prometheus will beryllium built successful New Albany, Ohio, while Hyperion will beryllium built successful Louisiana and is expected to beryllium afloat online by 2030, Zuckerberg said.

He said Meta would "invest hundreds of billions of dollars... to build superintelligence" and that nan centres had been fixed "names befitting their standard and impact".

Karl Freund, main expert astatine Cambrian AI Research, told nan BBC, "clearly, Zuckerberg intends to walk his measurement to nan apical of nan AI heap".

"The talent he is hiring will person entree to immoderate of nan champion AI Hardware successful nan world," Freund added.

Meta shares were trading 1% higher pursuing nan announcement, Reuters news agency reported. The banal has risen much than 20% truthful acold this year.

There are astatine slightest 10,000 information centres astir nan world hosting nan unreality - remote servers that shop integer accusation - pinch astir of them located successful nan US, followed by nan UK and Germany.

AI-driven information centres are highly power and h2o intensive. One study estimates that these centres could devour 1.7 trillion gallons of h2o globally by 2027. A azygous AI query - for example, a petition to ChatGPT - can usage astir arsenic overmuch h2o arsenic a mini vessel you'd bargain from nan area shop.