Meta is planning to spend USD hundreds of bns on some of the biggest data centers in the world, as the company looks to develop “superintelligence” — AI meant to surpass human-level intelligence — infrastructure across the US, starting with a cluster of mega-sized AI hubs, Meta CEO Mark Zuckerberg said in a post on Threads. Meta is in such a hurry to get there that it’s using tents to host temporary data centers while permanent facilities are still under construction, according to Techcrunch.

Prometheus is first in line, and more are on the way: The first mega-site, Prometheus, is a multi GW data center set to go live in 2026 in New Albany, Ohio. Hyperion, another site in Louisiana, is expected to scale up to 5 GW by 2030, with “titan clusters” also in the pipeline, Zuckerberg said. To put it in perspective, Zuckerberg said some of these clusters will cover areas nearly the size of Manhattan.

The urgency comes as Meta strives to catch up with rivals like OpenAI, xAI, and Google. To close the gap, Meta is poaching top AI researchers and fast-tracking massive infrastructure projects. It is using ultra-light, prefabricated structures with limited redundancy to get temporary compute capacity online. These setups prioritize speed over durability, and in some cases, skip backup systems entirely.

But there’s a catch: These data centers are extremely resource-intensive. AI-focused facilities use large amounts of electricity and water, with the World Economic Forum estimating they could collectively consume 6.4 tn liters of water by 2027. Even a single AI query can use as much water as a small bottle.

Tags: