Meta and Google are the latest to trade blows in the AI data center arms race, promising billions of dollars in new investments for massive infrastructure buildouts.
Meta CEO Mark Zuckerberg on Monday said the first in a series of AI data "superclusters" will come online next year as part of a plan to invest "hundreds of billions of dollars" to secure its spot as a dominate player in artificial intelligence. Superclusters are massive AI data centers powered by the most advanced AI accelerators.
Zuckerberg said in a post on Threads, Facebook's social media platform, that Prometheus would be one of several sites to provide multi-gigawatt computing power. Another planned supercluster, Hyperion, could be scaled to 5 gigawatts of power.
Meta isn't just building data centers. It's building bargaining power.
Pradeep SanyalAI and data leader, Capgemini
"Meta isn't just building data centers," said Pradeep Sanyal, AI and data leader at Capgemini, in an email interview. "It's building bargaining power. … Compute supremacy is now the new battleground for frontier AI. With gigawatt-scale clusters, Meta is buying its way into a class of infrastructure that only a handful of players can match."
Big tech's appetite for AI advancement has hyperscalers and AI companies racing to build such clusters.
Meta's AI Research SuperCluster has been operational since 2022. That cluster uses 16,000 Nvidia A100 GPUs for large language model (LLM) training. Amazon and Anthropic announced a joint supercluster this year -- Project Ranier -- that will use AWS' Trainium2 chips. Elon Musk's xAI effort, Colossus, is powered by 200,000 Nvidia chips that require 300 megawatts of power.
Google's power play
The propulsion of AI computing requires a massive amount of power, and big tech companies are competing to shore up energy pipelines.
On Tuesday, Google said it would invest $25 billion in data centers and AI infrastructure within the nation's largest U.S. electric grid -- which covers 13 states -- over the next two years.
Google also said it would spend $3 billion to update two hydropower plants in Pennsylvania to help power the energy-hungry data centers. The Pennsylvania investment is part of an agreement with Brookfield Asset Management to buy 3,000 megawatts of carbon-free hydroelectric power in the U.S.
Google's parent company, Alphabet, earlier this year said it would spend $75 billion in 2025 on data centers and infrastructure. In June, the company said it would partner with CTC Global Corp. to use new technology to free up grid capacity across the country.
Sanyal said there's a stark contrast between the AI strategies from Meta and Google.
"While Meta is betting on bespoke AI superclusters, Google is taking a more balanced route," he said. "They are pairing aggressive data center expansion with long-term clean energy procurement. Their $3 billion hydro deal secures up to 3 gigawatts renewable power, giving them not just capacity but predictability."
While Meta is going in on scale-first buildouts, Google is threading compute, sustainability and grid alignment, he said, adding that "both are aiming for leadership."
Meta's AI gamble
Sanyal said Meta's strategy might be to horde AI infrastructure and talent. The company has already poached top talent from competitors, including OpenAI and Apple, offering lucrative signing bonuses. Meta hired former Safe Superintelligence CEO Daniel Gross after a failed attempt to buy the company, according to published reports.
"Meta is trying to force its way back to the center of the AI conversation using compute, capital and headcount," Sanyal said. "This also comes with serious execution risk. Scaling that kind of infrastructure is hard. Talent integration at this level is even harder. And alignment, safety and public trust will only get trickier the more power you throw at the system."
Whether Meta executes on its strategy is still to be determined, but the outcome could be critical either way.
"If Meta executes, this could reset the pecking order in AI," he said. "And if they don't, it could be the most expensive HR experiment in history."
Shane Snider, a veteran journalist with more than 20 years of experience, covers IT infrastructure at Informa TechTarget.
