Tech giants are frantically snapping up electricity—but no one sees the real battlefield.

Zhang Lei, Chairman of Envision Group, made a groundbreaking statement on-site: “The power system is becoming the core engineering foundation of artificial intelligence—not merely a supporting infrastructure.” When the Moore’s Law governing chips and models collides with the slow evolution of traditional power systems, the bottleneck for the entire AI industry is shifting—from algorithms and chips—to “end-to-end energy management in intelligence production.” The key to breaking this impasse lies in the organic integration of energy, chips, and computing centers; and the AI-powered power system is precisely what establishes this tripartite integration as the new foundational infrastructure for AI. After AI’s breakneck expansion, traditional grids can no longer keep up. In recent years, rapid advances in AI technology have transformed it into a multi-trillion-dollar industry attracting massive venture capital. Since 2022, the market capitalization of AI-related companies in the S&P 500 Index has grown by approximately $12 trillion. Especially over the past two years, global computing demand has surged significantly. According to the China National Energy Administration’s “China ‘AI+’ Energy Development Report 2026,” “AI + Energy” has become a critical strategic direction for major countries worldwide to seize future development initiative. AI’s rapid growth is continuously driving up global electricity demand from computing infrastructure. In its report “Key Issues at the Intersection of Energy and Artificial Intelligence” (2026), the International Energy Agency (IEA) forecasts that data center electricity demand will accelerate sharply after 2030. The United States is expected to account for the largest share of this growth, followed by China. Data center electricity consumption is projected to rise from 485 terawatt-hours (TWh) in 2025 to 950 TWh in 2030—exceeding Japan’s current total national electricity consumption. The core driver behind surging computing demand is the Artificial Intelligence Data Center (AIDC). As the core infrastructure hosting massive AI data storage, computation, and deployment, AIDCs have seen electricity consumption skyrocket due to the explosive growth of generative AI—far outpacing traditional data centers. For several years, AIDCs competed fiercely for GPUs and high-bandwidth memory (HBM); but recently, as AI has become an “electricity glutton,” a new bottleneck has emerged: traditional power systems are simply insufficient. Giants are buying electricity—but failing to address the fundamental question. Recognizing this, global tech giants have sprung into action. Meta has partnered with ultra-long-duration energy storage companies, reserving up to 100 GWh of storage capacity; Google has become the first U.S. utility company to commit to purchasing power from next-generation small modular reactors (SMRs); and Microsoft revived the former Three Mile Island nuclear plant in the U.S. as early as 2024… Yet these solutions only address the question of “having electricity available.” A deeper issue remains overlooked: AI’s load profile is fundamentally undermining the design logic of existing power systems. Grid capacity is finite—and both renewable energy projects and large-scale power users are competing for access. Grid congestion is widespread across multiple countries; even queueing for grid interconnection can take years. Meanwhile, lead times for critical grid components—including transformers and cables—have doubled over the past three years.

AI data centers have reignited interest in a previously dormant clean-energy sector! Meta and Google are betting big on long-duration energy storage—with storage durations reaching up to 100 hours.

The soaring electricity demand from AI data centers is thrusting a long-dormant clean-energy sector back into the spotlight. Tech giants including Meta and Google are ramping up investments in long-duration energy storage (LDES), opening markets for novel storage technologies previously hindered by high costs and sluggish commercialization—through large-scale procurement and project partnerships. On September 18, Bloomberg reported that since the start of 2025, at least 150 gigawatt-hours (GWh) of LDES projects in the U.S. have been announced, approved, or secured financing—more than five times the total planned installed capacity nationwide at the end of 2024. Most of this newly added demand originates from AI data centers. Unlike conventional lithium-ion batteries—which typically provide only a few hours of storage—long-duration storage can hold electricity for dozens of hours or even days. This makes it far better suited to meet the escalating electricity demands of AI data centers, and also helps bridge the temporal mismatch between generation and consumption for wind and solar power. The involvement of tech giants also brings much-needed commercial validation—a resource previously scarce in this field. Google is partnering with Italy’s Energy Dome to deploy a carbon dioxide-based energy storage system in Arizona capable of storing power for 10 hours; Meta has signed an agreement with California-based Noon Energy to install a system capable of storing energy for up to 100 hours. With rising order volumes, LDES startups stand to attract more funding and scale up production capacity. AI demand is catalyzing long-duration storage—and accelerating technology deployment. Long-duration energy storage encompasses diverse technological pathways, enabling electricity to be stored and discharged continuously for hours—or even days. Historically, this sector has suffered from high costs and a lack of large-scale validation, resulting in slow commercial progress. Today, however, the expansion of AI infrastructure is rapidly accelerating its real-world deployment. Cully Cavness, Co-Founder and President of Crusoe, stated that long-duration storage is becoming a vital tool for scaling AI infrastructure. Tech giants are backing different technical approaches: Google collaborates with Energy Dome to achieve 10-hour storage using carbon dioxide; Crusoe partners with Energy Vault to store electricity via gravitational potential energy; and Meta’s collaboration with Noon Energy extends storage duration to 100 hours.

Meta and Google Invest in Long-Duration Energy Storage for AI

4 days ago

Meta and Google are investing billions in long-duration energy storage to power AI data centers, driving a fivefold increase in US projects since early 2025. This shift targets reduced fossil fuel reliance and critical mineral dependency while scaling renewable integration.

bloomberg

2026-09-18

Meta and Google Invest in Long-Duration Energy Storage for AI

Meta and Google Invest in Long-Duration Energy Storage for AI

FAQ