Meta moves to sell wholesale electricity through new subsidiary Atem Energy

Meta has filed an application with federal regulators to sell wholesale electricity through its newly formed subsidiary Atem Energy, marking the tech giant’s entry into energy markets. The move reflects Meta’s growing need to secure reliable power for its expanding AI data center infrastructure, which is driving unprecedented electricity demand across the industry.

What you should know: Meta’s energy subsidiary seeks broad market authorization to manage the company’s massive power requirements.

• Atem Energy has requested Federal Energy Regulatory Commission (FERC) approval to sell energy, capacity, and ancillary services by November 16.
• The wholly owned subsidiary was created earlier this year specifically to manage energy sales tied to Meta’s growing data center footprint.
• If approved, Atem would operate as a power marketer, allowing Meta to secure large-scale energy commitments and sell unused capacity back to the grid.

The scale is massive: Meta’s Louisiana data center project illustrates the enormous power demands of AI infrastructure.

• The facility is expected to exceed 2 GW of capacity, powered by three combined-cycle gas turbines totaling 2.26 GW built and operated by utility Entergy.
• Separately, Meta launched a request for proposals last year to identify nuclear developers capable of providing 1.4 GW of new nuclear generation for U.S. operations.
• These projects underscore how AI workloads require substantially more power than traditional data centers due to intensive computational demands from graphics processing units (GPUs) and tensor processing units (TPUs)—specialized chips designed for AI tasks.

Why this matters: The energy demands of AI are reshaping America’s electricity consumption patterns and forcing tech companies to become energy players.

• The U.S. Energy Information Administration (EIA), a federal statistics agency, revised its electricity demand projections upward, expecting consumption to rise from 4,097 billion kilowatt-hours in 2023 to 4,283 billion kWh by 2026.
• Data centers, particularly those supporting AI workloads, are identified as the primary driver behind these significant upward revisions.
• Training models like GPT-4 or Google’s Gemini requires processing trillions of parameters across thousands of high-performance processors, consuming far more power than traditional enterprise applications.

Strategic positioning: Meta’s energy market entry positions the company to better manage both supply security and cost optimization.

• Operating as a power marketer would give Meta more flexibility in securing energy commitments while potentially generating revenue from excess capacity.
• The company may need to apply for membership in competitive markets like the Midcontinent Independent System Operator (MISO), a regional electricity grid operator where its Louisiana facility is being developed.
• This approach mirrors strategies by other tech giants seeking greater control over their energy supply chains amid surging AI infrastructure demands.