Meta might not at present lead the race for AI superintelligence, but it surely’s drawing closely on its money reserves to pursue the expertise. Founder Mark Zuckerberg introduced Monday that the corporate will spend a whole bunch of billions of {dollars} to construct a number of monumental AI knowledge facilities.
The primary of those facilities—anticipated to be on-line subsequent yr—is named Prometheus. Zuckerberg says will probably be able to producing one gigawatt of power. However Meta isn’t stopping there.
“We’re additionally constructing Hyperion, which is able to be capable of scale as much as 5GW over a number of years. We’re constructing a number of extra titan clusters as effectively. Simply one in all these covers a major a part of the footprint of Manhattan,” he wrote on Fb. (A Meta spokesperson tells Quick Firm that Hyperion is at present underneath development in Louisiana, whereas the Prometheus challenge is an enlargement to an information middle campus in Ohio.)
5 gigawatts is roughly equal to the overall power consumption of Miami and equal to the output of 5 nuclear reactors—sufficient to energy about 3.5 million houses for a yr.
There’s a sure irony in naming the middle Prometheus. In Greek mythology, Prometheus defied the gods by stealing hearth from Olympus. Meta, in the meantime, has been aggressively poaching prime AI expertise from rivals, providing multimillion-dollar salaries that far exceed what rivals pay. Zuckerberg is reportedly overseeing these recruitment efforts himself. With knowledge facilities of this scale, he might additional strengthen the corporate’s hiring enchantment.
Meta has additionally acquired a 49% stake in Scale AI, bringing its CEO, Alexandr Wang, on board. Nat Friedman, former CEO of GitHub, has additionally joined Meta. Daniel Gross—the CEO and cofounder of Secure Superintelligence, which he launched with OpenAI cofounder Ilya Sutskever final June—is now a part of the corporate as effectively.
With this wave of infrastructure bulletins, Zuckerberg is signaling Meta’s intent to outpace OpenAI and different synthetic intelligence initiatives. This comes after the corporate’s Llama 4 mannequin underperformed following a number of high-profile employees departures. In response, Meta reorganized its AI division and renamed it Superintelligence Labs.
And Meta definitely has the funds to assist this push. In April, it raised its anticipated capital expenditure to as excessive as $72 billion. On Monday, Meta’s inventory rose slightly below 1%, to $724 per share. The corporate at present has a market capitalization of $1.82 trillion—vastly increased than OpenAI’s $300 billion valuation.
However the competitors isn’t sitting idly by. OpenAI is constructing a 5-gigawatt knowledge middle of its personal, known as Stargate. Introduced in January, the corporate stated it will make investments $500 billion over 4 years to construct the ability, with assist from Oracle, SoftBank, and MGX.
OpenAI dedicated to deploying $100 billion instantly. Texas has been designated because the flagship location for the info facilities, with the primary website anticipated to start operations later this yr in Abilene.
“This challenge is not going to solely assist the re-industrialization of america but additionally present a strategic functionality to guard the nationwide safety of America and its allies,” OpenAI wrote in a weblog put up.
Alphabet, in the meantime, is spending $3.3 billion on two new knowledge facilities in South Carolina.
One subject Meta didn’t tackle: the environmental influence of those huge knowledge facilities. A scientific paper printed final yr on Cornell College’s preprint server arXiv (titled “The Unpaid Toll: Quantifying the Public Well being Affect of AI”) estimated that air pollution from AI knowledge facilities might trigger as much as 1,300 untimely deaths yearly by 2030. It additionally estimated that public well being prices associated to their air air pollution are already at $20 billion per yr.
By 2030, the researchers forecast, the general public well being burden from AI knowledge facilities might be twice that of the U.S. metal trade, and will rival emissions from all of the vehicles, buses, and vehicles in California.
In the meantime, the Trump administration has dismantled dozens of local weather applications in its first 100-plus days. The Environmental Safety Company (EPA) can be contemplating overturning earlier findings that classify greenhouse gasoline air pollution as dangerous—probably weakening its capability to control carbon emissions.
