Why Mark Zuckerberg is positioning Meta as an AI infrastructure giant

The new organization is designed to secure the massive amounts of computing power—measured in gigawatts, each of which could power hundreds of thousands of homes—needed for Meta’s drive to build AI models that lead to “superintelligence.”

“Meta is planning to build tens of gigawatts this decade, and hundreds of gigawatts or more over time,” Zuckerberg wrote. “How we engineer, invest, and partner to build this infrastructure will become a strategic advantage.”

Under the new structure, Zuckerberg said longtime Meta executive Santosh Janardhan will continue to run the company’s technical architecture, software, custom chips, and the day-to-day building and operation of Meta’s vast data center network. Meanwhile, Daniel Gross—one of Zuckerberg’s high-profile AI hires from last summer, who was previously cofounder of Safe Superintelligence with former OpenAI chief scientist Ilya Sutskever—will lead a new group focused on the long game: how much computing power Meta will need years from now, where it should be built, how to secure scarce chips and energy, and how to model the business impact of those bets.

Zuckerberg also announced a new Meta president and vice chair, Dina Powell McCormick, to work on developing partnerships with governments to finance and deploy data centers around the world. She was previously deputy national security advisor for strategy to President Trump.

A growing perception that Meta is playing catch-up

For some Meta-watchers, the Meta Compute announcement was puzzling. After all, Meta is already an AI infrastructure giant. It has been a year since the company broke ground on its Hyperion site—a 4-million-square-foot data center campus in northeast Louisiana that Zuckerberg famously told President Trump was roughly comparable to the size of lower Manhattan. Why, then, did Meta suddenly need to declare a new top-level organization to do what it already appears to be doing at historic scale?

“This was kind of a head scratcher, I didn’t understand it at first,” said Patrick Moorhead, founder and chief analyst at Moor Insights and Strategy. He suggested the message about Meta Compute was really meant for investors and employees, signaling that Meta remains a serious contender in a club led by Microsoft, Google, Oracle, OpenAI, and xAI. “It’s Meta saying, ‘Here is our strategy to roll this out,’” he said.

Rick Pederson of Bow River Capital agreed that the announcement responds to a growing perception among market analysts that Meta is playing catch-up with Google and OpenAI in the AI arms race.

“It was a way to discuss their focus, their intentionality surrounding the build-out of AI, computational capacity and infrastructure,” he said. “I would imagine that the other major hyperscalers have organizations something like this. But he took the opportunity to define it.” Even though the company spent over $70 billion on AI infrastructure last year and plans to spend another $600 billion in the next two years, Google and OpenAI are spending at least that amount, he added. “So I think this gave Zuckerberg a chance to talk about not just the emphasis, but also how they’re going to do it.” 

Doubling down on infrastructure as an investment portfolio

Some experts said they are not at all surprised by the move to announce Meta Compute. “Meta is doubling down on infrastructure as an investment portfolio rather than simply a cost center,” said Lane Dilg, former head of infrastructure policy at OpenAI and founder of boutique advisory firm Apeiro. 

As the AI boom accelerates, Meta no longer sees data centers, GPUs, power contracts, and custom chips as mere plumbing for its products. It treats them as strategic assets—more like a capital allocator than a tech company. In effect, Dilg said, Meta is positioning itself not just against other hyperscalers, but against the world’s most sophisticated investment platforms.

In that landscape, she explained, choosing Daniel Gross to co-lead the initiative makes sense. “Gross’s experience building AI-native and agentic platforms matters, paired with his compute expertise,” she said, pointing to the supercomputer he built with NFDG coinvestor Nat Friedman—now Meta’s head of products. Their effort became the Andromeda Cluster, a network of computing stockpiles totaling more than 4,000 GPUs, which they made available to their portfolio companies at below-market rates.

Gross has dropped hints about his own recruiting push for Meta Compute, saying in a recent post on X that he is hiring people with backgrounds in “deep learning, supply chains, commodities, semiconductors, sovereigns, energy, Excel, prediction markets, monitoring situations, etc.” This suggests Meta is preparing to hedge against volatility in power and hardware costs—and to make long-term bets shaped not just by technology, but by energy markets, supply chains, and geopolitics.

Powell McCormick is also a key strategic hire for the Meta Compute effort, Umesh Padval, an experienced investor and board member, told Fortune. “Hyperscalers are now focused on how to get power and are investing in power projects financed by cash and debt,” he said. “With her banking and political background, she would work with the Meta data center group to finance and get approvals to enable faster building of the compute capacity.” 

Critics say Meta’s capital-intensive model would drag down returns

Not all are supportive of Meta’s move, however. In a social media post shared the day Meta Compute was announced, “Big Short” investor Michael Burry wrote: “Meta gives in, throwing away its one saving grace. Watch ROIC crash.” Burry’s warning reflects a fear that Meta is abandoning its ability to generate enormous profits without sinking vast sums into physical infrastructure. By embracing gigawatt-scale data centers, he argues, Meta is shifting toward a far more capital-intensive model—one that could drag down returns and make the company look more like a utility. 

But since Meta has already poured tens of billions into AI data centers and signaled hundreds of billions more in long-term infrastructure commitments, it looks like that train left the station long ago.