Meta Compute: Zuckerberg's Tens-of-Gigawatts AI

What is Meta Compute?

Meta Compute is a new top-level initiative at Meta designed to build and operate AI infrastructure at a massive scale. CEO Mark Zuckerberg announced it on January 12, 2026, in a post on Threads. The goal: build tens of gigawatts of AI capacity this decade, and "hundreds of gigawatts or more over time," according to Axios.

Who is running Meta Compute?

Zuckerberg named three executives to lead the effort.

Santosh Janardhan is Meta's head of global infrastructure and has been with the company since 2009. He will oversee technical architecture, the software stack, Meta's silicon program, developer productivity, and the company's global data center fleet and network.

Daniel Gross joined Meta last year from Safe Superintelligence, where he was CEO and co-founder alongside former OpenAI chief scientist Ilya Sutskever. Gross will lead a new group responsible for long-term capacity strategy, supplier partnerships, industry analysis, planning, and business modeling.

Dina Powell McCormick, a former government official who joined Meta as president and vice chairman on January 12, 2026, will work with governments to help "build, deploy, invest in, and finance Meta's infrastructure," per TechCrunch.

What are Meta's infrastructure spending commitments?

Meta previously said it intends to invest $600 billion in American infrastructure and jobs — including AI data centers — by 2028. The company had not offered many details on how that capital fits into its long-term strategy before this announcement.

Meta's CFO Susan Li said during an earnings call last summer: "We expect that developing leading AI infrastructure will be a core advantage in developing the best AI models and product experiences."

You might also like

China's Humanoid Robots Stun at 2026 Spring Gala

Google Emissions Up 51% Since 2019 Despite Clean

OpenAI Cuts Inference Costs in Half on Key Models

Anthropic Launches Claude Science for Drug

Here's a snapshot of the key figures and roles tied to Meta Compute:

Will Meta sell compute to outside customers?

Yes — that is part of what is being explored. According to Investing.com, Meta is developing a dual-pronged approach to monetize its infrastructure:

• Model-as-a-Service: Selling access to AI models hosted on Meta's infrastructure — similar to AWS's Bedrock offering — where Meta operates the data centers and chips while charging developers for access.
• Raw Compute Infrastructure: Selling bare-metal computing capacity, positioning Meta against "neocloud" providers like CoreWeave.

This would put Meta in direct competition with Amazon Web Services, Microsoft Azure, and Google Cloud Platform.

How did markets react?

Meta shares jumped as much as 8% following a Bloomberg report on the cloud business plans. The stock surge reflected investor optimism that Meta could monetize its heavy capital spending.

Adam Crisafulli, analyst and founder of Vital Knowledge, laid out both sides of the trade. On the bull case, he wrote: "Meta has been one of the heaviest spenders (in terms of capex/revenue) and many feared it was building way more capacity than it could ever use internally, so this external cloud business will help monetize all that infrastructure, bolstering revenue, margins, and cash flow."

On the bear case, Crisafulli warned: "The formation of an external cloud platform is a tacit admission from mgmt. that it overbuilt capacity and/or is falling short on its own internal AI model initiatives."

He also drew a comparison to another company making a similar move: "Meta isn't the first company to make this transition — SpaceX's xAI also appears to be dialing back expectations for its internal AI tools (Grok and Cursor) and has started selling capacity to external customers." Our coverage of SpaceX's Grok engineers tracks that parallel shift in detail.

Crisafulli's bottom line: "The Meta news is great for that company specifically, but negative for sentiment toward pick-and-shovel providers (and it could weigh on other hyperscalers/neocloud companies too)."

Why does the gigawatt scale matter?

A gigawatt equals one billion watts of electrical power. AI data centers are among the most energy-intensive facilities on the planet. One estimate cited by TechCrunch projects that AI power demand in the U.S. could rise from 5 GW to 50 GW by 2030. Meta's stated target of tens of gigawatts this decade sits squarely inside that demand curve.

The energy footprint of AI infrastructure has become a major concern across the industry. For context on how other tech giants are handling the environmental side of this buildout, our piece on Google's emissions covers the tension between AI expansion and clean energy commitments.

As we read the announcement, the creation of Meta Compute formalizes what had previously been a collection of infrastructure efforts under a single strategic umbrella — with named leadership, defined responsibilities, and an external revenue angle that wasn't publicly confirmed before.

How does this compare to what Meta's peers are doing?

Meta is not alone in racing to build AI-ready infrastructure. Microsoft has been partnering with AI infrastructure providers. In December 2025, Google parent Alphabet acquired data center firm Intersect Power to work around energy grid bottlenecks. Meta's move into external cloud sales would add a fourth major competitor to a market currently dominated by AWS, Azure, and Google Cloud.

Developers already weighing their options on AI model access should also watch how Amazon's AI model strategy evolves alongside Meta's new offering.

The $600 billion commitment to U.S. infrastructure by 2028 remains Meta's most concrete public spending target tied to this initiative.