Meta launches Meta Compute to build massive AI power network

Meta unveiled Meta Compute in mid-January, a new top-level initiative charged with planning, building and operating a dramatic expansion of the company’s artificial intelligence computing infrastructure. The program will report directly to CEO Mark Zuckerberg and is intended to unify decisions about power, land, equipment and networking as a coordinated engineering and strategic challenge.

Meta said it plans to add "tens of gigawatts" of capacity this decade and to scale to "hundreds of gigawatts or more over time," a pace of growth that industry analysts say would place AI workloads on a scale comparable to small cities or even countries. The company framed the expansion as central to its effort to accelerate frontier AI work and to create a long-term competitive advantage based on how it engineers, invests and partners for infrastructure.

Meta Compute will be co-led by Santosh Janardhan and Daniel Gross. Janardhan, who already runs Meta’s global infrastructure and co-heads engineering, will maintain oversight of system architecture, in-house silicon efforts, the software stack and the day-to-day operation of the company’s worldwide data center fleet and network. Gross will lead strategic capacity planning and business partnerships, defining future compute requirements, building supply chains capable of delivering hardware at multi-gigawatt scale and developing planning models that account for industry shifts and resource constraints.

In a parallel executive appointment, Dina Powell McCormick joined Meta as president and vice-chair. She will focus on government and sovereign partnerships and will handle government deals to finance and build capacity and to secure approvals for large-scale energy and facilities projects. Powell McCormick is a former White House adviser and national security official, and Meta has positioned her role as central to navigating the regulatory and political landscape that will accompany the company’s rapid buildout.

The announcement follows sustained investment and hiring around AI. Meta disclosed very large infrastructure spending in recent filings and commentary, including a reported $72 billion committed to AI infrastructure in its 2025 fiscal year and a separate commitment of $600 billion in U.S. infrastructure spending by 2028; the company has also spent months recruiting top AI talent. At the same time, the company is trimming other businesses, with cuts planned in its Reality Labs unit.

Meta currently operates roughly 30 data centers globally, most within the United States. To underpin the expansion, the company has secured long-term power arrangements cited as 20-year agreements with nuclear energy firms Vistra, TerraPower and Oklo. Those deals reflect an effort to lock in stable, high-density power for AI facilities but also underscore the scale of energy and permitting challenges ahead.

The scale and pace of Meta’s plan raise environmental and community concerns. Large new data centers require substantial electricity and water, engage local permitting processes and can reshape regional grids. Meta’s concentration on coordinated planning and government partnerships signals an awareness of those obstacles, but the company has not released independent estimates of total future electricity or water consumption.

Meta Compute represents a bet that infrastructure, power, real estate, hardware and supply chain, will be the defining battleground for the next phase of AI. If sustained, the program could reshape energy markets, create new geopolitical leverage and force communities and regulators to confront tradeoffs between economic investment and environmental impact.