Evan Lee Salim
The United States is on the precipice of a monumental energy challenge, as new analysis reveals that data centers are set to consume a staggering one-fifth of the nation’s total electricity generation by 2035, a fourfold increase from current levels. This dramatic surge, primarily fueled by the exponential demands of artificial intelligence (AI) compute, is detailed in a recent report by BloombergNEF, which underscores a rapidly escalating trajectory that is already forcing utilities and grid operators to re-evaluate infrastructure planning and accelerate modernization efforts. The report projects that data center capacity will swell to nearly 200 gigawatts over the next decade, with almost half of this colossal demand dedicated to the intensive processes of AI training and inference, a significant portion of which will remain concentrated within U.S. borders. By 2033, the nation is expected to host an overwhelming 64% of global AI chips by power demand, solidifying its position as the global nexus for advanced computing, yet simultaneously placing immense pressure on its aging energy infrastructure.
The Accelerating Forecasts: A Chronology of Underestimation
The current projections from BloombergNEF represent a stark upward revision, highlighting a pervasive underestimation of AI’s energy appetite across the industry. Just months prior, in December, the consultancy’s estimate for 2035 electricity demand was 83% lower than its latest figure, a testament to the unforeseen velocity of AI adoption and data center development. This pattern of rapidly escalating forecasts is not unique to BloombergNEF. Other prominent organizations have similarly adjusted their outlooks, reflecting a broader awakening to the scale of the impending energy challenge. The Electrical Power Research Institute (EPRI), a non-profit dedicated to the electrical industry, has more than doubled its 2024 estimate for data center electricity consumption. Concurrently, S&P Global’s forecast saw an increase of over a third between October and April. These repeated and significant revisions underscore the "fevered pace" of data center expansion across the U.S., driven by massive investments from tech giants and a competitive race to deploy AI capabilities.
Historically, data center energy consumption, while growing, followed a more predictable trajectory. For much of the 2010s, efficiency gains in computing hardware and cooling technologies managed to somewhat temper the overall increase in power demand, even as data volumes expanded. However, the advent of generative AI and large language models has fundamentally altered this dynamic. AI workloads are inherently power-intensive, relying on specialized graphics processing units (GPUs) that consume significantly more energy per operation than traditional CPUs. Training a single large AI model can require gigawatt-hours of electricity, equivalent to the annual consumption of thousands of homes. The continuous "inference" phase, where AI models are used to process requests, also demands substantial, always-on power. This shift from incremental growth to exponential demand has caught many forecasters off guard, leading to the rapid, successive upward revisions seen today.
The Grid Under Siege: Regional Hotspots and Challenges
The looming energy demand from data centers is poised to collide with an electrical grid already facing numerous stresses. The U.S. grid, a complex network of generation, transmission, and distribution assets, is characterized by aging infrastructure, slow permitting processes for new power plants and transmission lines, and the challenging integration of intermittent renewable energy sources. This confluence of factors creates a precarious environment where new, massive loads like data centers can quickly overwhelm regional capacity.
Two regions are particularly vulnerable, according to the BloombergNEF report: the PJM Interconnection and ERCOT. PJM, which oversees the electricity grid spanning from Virginia to Illinois, is projected to see a staggering 34% of its total electricity generation diverted to data centers. This area, particularly Northern Virginia, has long been a global epicenter for data center development due to its strategic location, robust fiber optic networks, and relatively stable power supply. However, the sheer volume of new connection requests has pushed PJM to its limits. The grid operator previously paused applications for new sources to connect to the grid for four years, a drastic measure indicating the severity of the congestion and the struggle to process the backlog of both new generation and new load requests. This pause created significant uncertainty for developers and threatened to stifle economic growth in the region.
While PJM reopened its queue to new generating sources in April, the underlying issues persist. The situation has grown so dire that American Electric Power (AEP), a major utility operating within the PJM footprint, has reportedly threatened to pull out of the interconnection, citing the immense strain and financial implications. Such a move would be unprecedented and could further destabilize an already stressed system. The supply-demand imbalance has already translated into tangible economic consequences, with electricity prices within PJM rising by an astonishing 76% over the past year. This increase directly impacts consumers, businesses, and the broader economy, potentially eroding the cost advantages that initially attracted data centers to the region. Despite these formidable challenges, data centers continue to seek connection to PJM, representing 38% of the charges in the grid manager’s most recent capacity auction, signaling the strategic importance of the region for these facilities, even at elevated costs.
ERCOT, the independent grid operator covering most of Texas, faces similar, albeit distinct, pressures. The report indicates that ERCOT will need to devote 22% of its generating capacity to data centers. Texas’s unique, largely isolated grid structure makes it particularly sensitive to demand fluctuations and supply constraints. The state has experienced rapid population growth, intense industrial development, and a booming renewable energy sector, all of which contribute to complex grid management challenges. Past events, such as Winter Storm Uri in 2021, exposed the grid’s vulnerabilities to extreme weather and highlighted the critical need for robust, reliable generation and transmission infrastructure. The influx of data centers adds another layer of complexity, demanding consistent, high-quality power in a region prone to both heat waves and cold snaps, which significantly impact electricity demand and supply. Texas’s pro-business environment and abundant, relatively cheap land have attracted significant data center investment, but the energy demands are now testing the limits of its independent grid.
Energy Policy, Infrastructure Investment, and Sustainability Implications
The unprecedented surge in data center electricity demand necessitates a multi-faceted response encompassing energy policy, infrastructure investment, and a renewed focus on sustainability. Policymakers at both state and federal levels are grappling with how to facilitate necessary grid upgrades while also ensuring reliability and affordability. Discussions are intensifying around streamlining permitting processes for new power generation and transmission projects, which can often take years, if not decades, to navigate regulatory hurdles. Incentives for grid modernization, including the deployment of advanced smart grid technologies, energy storage solutions, and demand-response programs, are becoming critical tools to manage the volatile balance of supply and demand. Furthermore, considerations for the optimal siting of new data centers are evolving, moving beyond mere land availability to incorporate grid capacity, proximity to renewable energy sources, and potential community impacts.
Utilities, on the front lines of this energy transformation, are struggling to keep pace. They face the immense capital expenditure required to build new power plants, upgrade transmission lines, and enhance distribution networks. Many are advocating for regulatory frameworks that allow for more proactive infrastructure investment recovery and more flexible planning horizons. Some are exploring innovative partnerships with data center developers to co-locate generation, including small modular reactors (SMRs) or large-scale battery storage, to alleviate strain on the broader grid.
Data center developers, for their part, are increasingly prioritizing clean energy sourcing. Many major tech companies have set ambitious goals to power their operations entirely with renewable energy. This has led to a proliferation of Power Purchase Agreements (PPAs) with wind and solar farms, and a growing interest in developing on-site renewable generation. However, the sheer scale of future demand means that meeting these renewable energy targets will require an unprecedented build-out of clean energy infrastructure. Beyond carbon emissions, data centers also present significant environmental concerns related to water usage, primarily for cooling systems. As facilities become denser and more powerful, their water footprint grows, raising concerns in regions already facing water scarcity. This necessitates innovation in cooling technologies, including liquid cooling solutions and advanced waste heat recovery systems.
Economically, the data center boom presents a dual-edged sword. While it brings significant investment, high-paying jobs, and reinforces the U.S.’s leadership in the digital economy, the escalating electricity costs and grid strain could have broader economic implications. Increased power prices could affect other industries, making the U.S. less competitive for manufacturing or other energy-intensive enterprises. Maintaining grid stability and affordable power is crucial for long-term economic health and the nation’s attractiveness as a global technology hub.
The Global Race for AI Compute and Energy
The energy implications of AI are not confined to the United States. The BloombergNEF report projects that by 2033, if AI adoption continues along an aggressive trajectory, data centers will generate an astounding 1,935 terawatt-hours of new electricity demand worldwide. To put this into perspective, this figure is nearly equivalent to the entire annual electricity consumption of India, the world’s most populous nation. This global surge highlights a looming energy crisis that will impact grids, economies, and environmental targets across the globe.
Regions like Europe and Asia-Pacific are also experiencing rapid data center expansion and facing similar challenges. European nations, with their ambitious climate goals, are under pressure to ensure new data center capacity is powered by renewables, while simultaneously addressing existing grid bottlenecks. In Asia, particularly in fast-growing economies, the demand for digital services and AI capabilities is driving massive data center investments, often in regions with less developed or already strained energy infrastructures. The geopolitical dimension is also becoming increasingly apparent, as energy security and access to reliable power become critical factors in the global competition for AI dominance and chip manufacturing capabilities. Countries that can reliably and sustainably power their AI infrastructure will gain a significant strategic advantage.
Conclusion: A Call for Innovation and Collaborative Action
The projected fourfold increase in U.S. data center electricity consumption by 2035, driven by the relentless march of AI, presents an existential challenge to the nation’s energy infrastructure. It is a dual imperative: to meet unprecedented demand while simultaneously ensuring grid reliability, economic affordability, and environmental sustainability. The rapid revisions in forecasts underscore the dynamic nature of this challenge and the need for agile, forward-looking planning.
Addressing this requires a collaborative effort from policymakers, utilities, technology companies, and research institutions. Innovation in energy generation, including advanced nuclear technologies and next-generation renewables, will be critical. Equally important are investments in grid modernization, including smart grid technologies, energy storage, and more resilient transmission systems. Data center operators must continue to push the boundaries of energy efficiency, exploring advanced cooling, optimized hardware, and intelligent workload management. Ultimately, the future of AI and the digital economy hinges on the ability to power this revolution responsibly and sustainably, transforming what appears to be an insurmountable obstacle into an opportunity for a more robust, intelligent, and greener energy future.
