How Meta’s CEO Is Pushing the AI Arms Race to Another Level

Jul 16, 2025 14:01:00 -0400 by Tae Kim | #AI #Barron's Tech

Meta CEO Mark Zuckerberg is spending big on artificial intelligence. (Anna Moneymaker/Getty Images)

This article is from the free weekly Barron’s Tech email newsletter. Sign up here to get it delivered directly to your inbox.

Game of Kings. Hi everyone. Thanks to Meta Platforms CEO Mark Zuckerberg, the price to compete in the AI arms race has just gone up this week. Way up.

On Monday, Zuckerberg outlined his new AI strategy on his social-media platform Threads. “I’m focused on building the most elite and talent-dense team in the industry,” he wrote. “We’re also going to invest hundreds of billions of dollars into compute to build superintelligence.”

The executive said Meta is building several multi-gigawatt data center clusters, with the first one named Prometheus coming online next year. He added that another cluster, Hyperion, will reach five gigawatts in a few years.

The sheer scale and scope of these figures should not be overlooked. Prior data centers for traditional cloud computing typically used about 50 megawatts of power. Today’s high-end AI infrastructure data center with 100,000 GPUs may use 150 megawatts. Multiple multi-gigawatt data centers represent an exponential step up for next year and beyond.

The moves come after Meta’s latest AI model, Llama 4, fizzled in the marketplace, with one prominent AI researcher calling the offering “entirely lost.” Other leaders might back down after such a setback and become more cautious. But not Zuckerberg.

“We’re all in on this,” Zuckerberg said in an interview with The Information, repeatedly saying Meta will do “whatever it takes” to win the AI race.

The executive has gone on a recruiting blitz, poaching top AI talent from OpenAI and Google by paying top dollar. Last Friday, semiconductor and AI research firm SemiAnalysis reported Meta has been offering employment contracts worth $200 million over four years to AI researchers to join the company. Meta declined to comment on the figure.

Zuckerberg is using the power of his company’s resources and his control over Meta as a competitive advantage against rivals. Meta has a dual-class share structure where Zuckerberg has a majority of voting power, giving him de facto control over the company’s strategy.

“Meta Superintelligence Labs will have industry-leading levels of compute and by far the greatest compute per researcher,” he wrote. “I’m looking forward to working with the top researchers to advance the frontier!”

Meta is in a unique position versus AI model labs in that it can use its AI infrastructure not only to create new models but also to serve its core social networking and messaging businesses, which then leads to a larger gusher of profits to do more investments.

Last year, the company generated $164.5 billion in revenue and $62.4 billion in net profit. In April, management increased its 2025 forecast for capex to a range of $64 to $72 billion versus a prior outlook of $60 billion to $65 billion.

Look for Meta’s capex to increase dramatically from there going forward, as Zuckerberg stated his goal is to develop advanced AI, which he calls superintelligence, within the next two to three years. Simple math shows that “hundreds of billions” spent over a few years equates to massive investment.

Meta’s higher level of spending will affect the entire industry. The top AI labs and large technology companies will be forced to compete with Meta’s spending, which means more AI infrastructure investment and more compensation for talent.

Earlier this month, New Street analyst Pierre Ferragu said the top four data center technology companies- Alphabet , Amazon.com , Meta, and Microsoft -will grow total capital expenditures to $1.7 trillion by 2035 from $341 billion this year.

Those forecasts may go even higher. One should not understate this trend. Never before have the most profitable companies in history all been investing in a new computing paradigm infrastructure at the same time at this scale.

Of course, this is good news for Nvidia , which is the primary AI hardware provider for Meta and the industry. Earlier this year, I asked Nvidia CEO Jensen Huang when we could expect one-million-GPU-sized superclusters inside data centers. He told me he’s confident that several one-million-GPU clusters would be built by 2027.

Zuckerberg is making that prediction a reality. The AI arms race is lifting off.

This Week in Barron’s Tech

Write to Tae Kim at tae.kim@barrons.com or follow him on X at @firstadopter.