AI Datacenters Projected to Consume One-Fifth of All US Electricity by 2035

The proliferation of artificial intelligence (AI) infrastructure is set to dramatically reshape the energy landscape of the United States, with a new report from research firm BloombergNEF projecting that AI-driven datacenters will consume approximately 20 percent of all electricity generated in the country by 2035. This staggering forecast underscores the immense power demands of advanced computing and highlights a critical challenge for energy planners, policymakers, and grid operators nationwide. The growth trajectory is primarily fueled by the burgeoning requirements for training and operating sophisticated AI models, necessitating an unprecedented expansion of datacenter capacity.

According to BloombergNEF’s latest analysis, the total capacity of datacenters across the US is expected to surge to nearly 200 gigawatts (GW) within the next decade. Crucially, roughly half of this colossal capacity – an estimated 100 GW – will be exclusively dedicated to AI workloads. This represents a monumental increase in energy consumption from current levels and signals a profound shift in demand patterns, placing immense pressure on existing electrical grids and accelerating the need for substantial investments in new generation and transmission infrastructure.

The Unprecedented Surge in AI-Driven Demand

The rapid evolution of artificial intelligence, particularly the development and deployment of large language models (LLMs) and other complex machine learning algorithms, is the primary catalyst behind this projected energy surge. These AI systems require vast computational resources, often relying on specialized hardware such as Graphics Processing Units (GPUs) and Application-Specific Integrated Circuits (ASICs), which are inherently more power-intensive than traditional CPUs. The processes of training these models, involving billions of parameters and trillions of operations, consume extraordinary amounts of electricity over extended periods. Furthermore, the subsequent inference phase, where AI models are used to generate predictions, responses, or analyses, also demands significant, continuous power.

The United States is anticipated to remain the global epicenter for AI infrastructure, solidifying its leadership in technological innovation. BloombergNEF projects that by 2033, the US will account for a remarkable 64 percent of all AI chips globally, measured by their power consumption. This concentration of AI compute power within the US amplifies the domestic energy challenge.

This upward revision in energy demand projections has been swift and substantial. Compared to an earlier estimate from BloombergNEF in December, the anticipated electricity demand for 2035 has now been raised by a staggering 83 percent. This dramatic adjustment in a matter of months reflects the accelerating pace of AI development and adoption, coupled with a clearer understanding of the immense energy footprint of next-generation AI technologies. Other prominent research organizations, including the International Energy Agency (IEA) and consulting firms like McKinsey and Boston Consulting Group, have also issued revised, higher expectations for datacenter power consumption, affirming the trend identified by BloombergNEF. The consensus among analysts is that previous forecasts significantly underestimated the energy intensity of the AI revolution.

Strain on America’s Electrical Grids

The growth of AI datacenters is not evenly distributed across the US; instead, it is expected to concentrate in regions where the electricity grid is already experiencing considerable strain. This geographical concentration exacerbates existing infrastructure challenges and raises significant concerns about grid stability, reliability, and electricity prices.

One such critical region is served by the PJM Interconnection, which operates a wholesale electricity market and manages the transmission grid in 13 states across the Mid-Atlantic and Midwest, along with the District of Columbia. By 2035, approximately 34 percent of the electricity within the PJM network is projected to be consumed by datacenters. PJM has historically grappled with an overburdened interconnection process for new power plants and large electricity consumers, leading to significant delays in bringing new capacity online. The surging demand from datacenters has already contributed to a 76 percent increase in electricity prices within the PJM network over the past year, directly impacting businesses and residential consumers. The sheer volume of applications for new generation and load connections in PJM’s queue highlights the systemic challenges in upgrading and expanding grid infrastructure at the necessary pace.

Another critical region is ERCOT, the Electric Reliability Council of Texas, which manages the flow of electric power to more than 26 million Texas customers, representing about 90 percent of the state’s electric load. In ERCOT, datacenter consumption is projected to reach 22 percent of the total electricity supply by 2035. Texas, with its abundant land, relatively lower energy costs, and a growing renewable energy sector, has become an attractive location for datacenter development. However, ERCOT’s grid has faced well-documented reliability challenges, particularly during extreme weather events, making the prospect of such a substantial increase in demand a major concern for energy security and system resilience.

The implications of this concentrated demand extend beyond mere consumption figures. It necessitates massive investments not only in new generation sources but also in the transmission and distribution infrastructure to deliver that power reliably. Upgrading existing lines, building new substations, and developing smart grid technologies capable of handling dynamic loads become imperative. The lead times for such infrastructure projects can span years, creating a potential mismatch between the rapid deployment of datacenters and the slower pace of grid modernization.

Global Implications and the Race for AI Compute

While the United States is poised to house the largest share of AI computing power, the demand for energy-intensive datacenters is escalating rapidly worldwide. If the current trajectory of AI development and adoption continues unabated, datacenters globally are projected to consume an additional 1,935 terawatt-hours (TWh) of electricity annually by 2033. To put this into perspective, this figure is nearly equivalent to the entire annual electricity consumption of India, the world’s most populous country and its third-largest emitter of greenhouse gases.

This global surge has profound implications for international energy markets, climate targets, and geopolitical dynamics. Countries in Europe, Asia, and other regions are also witnessing significant growth in datacenter construction, driven by domestic AI ambitions, cloud computing expansion, and data localization requirements. The global race to build AI capabilities is intrinsically linked to the race for secure, affordable, and sustainable energy supplies. Nations that can effectively power their AI infrastructure will gain a competitive edge in the evolving technological landscape.

The Energy-Intensive Nature of AI: A Deeper Dive

The extraordinary power demands of AI datacenters stem from several factors:

1. Specialized Hardware: AI workloads heavily rely on GPUs and other accelerators designed for parallel processing. These chips are significantly more power-hungry than conventional CPUs. A single high-end AI accelerator can draw hundreds of watts, and thousands of these chips are clustered together in modern AI datacenters.
2. Computational Intensity: Training large AI models involves billions of mathematical operations, often running for weeks or months continuously. Inference, while less intense than training, still requires substantial power for real-time processing of queries.
3. Cooling Systems: The immense heat generated by high-density AI hardware necessitates robust and energy-intensive cooling systems. Traditional air cooling systems consume significant electricity for fans and chillers. While liquid cooling technologies (e.g., direct-to-chip or immersion cooling) offer greater efficiency and heat removal capabilities, they still require energy for pumps and heat exchangers.
4. Power Delivery Infrastructure: Datacenters require sophisticated power distribution units (PDUs) and uninterruptible power supplies (UPS) to ensure stable and reliable electricity delivery, which introduces additional energy losses.
5. 24/7 Operation: Datacenters operate continuously, meaning their energy demand is constant, unlike many other industrial or residential loads that fluctuate significantly.

Chronology of Escalating Projections and Market Dynamics

The timeline of these escalating energy projections is telling. Just a few years ago, estimates for datacenter energy consumption were considerably lower. However, the rapid advancement and mainstream adoption of generative AI since late 2022, particularly the widespread accessibility of tools like ChatGPT, have fundamentally altered expectations. The BloombergNEF report’s December estimate, already substantial, was quickly deemed insufficient, leading to the 83% upward revision for 2035 demand. This rapid recalibration reflects how quickly the AI industry is deploying infrastructure and how analysts are struggling to keep pace with its unprecedented growth rate.

This accelerated timeline means that infrastructure planning, which typically operates on longer cycles, must adapt at an unprecedented speed. The construction of a new power plant can take several years, as can the deployment of major transmission lines. This mismatch between AI’s rapid deployment cycle and energy infrastructure’s slower development poses a significant logistical and financial challenge.

Reactions and Preparations from Stakeholders

The implications of this energy forecast are reverberating across various sectors, prompting diverse reactions and strategic preparations:

• Grid Operators (e.g., PJM, ERCOT, ISO New England): These entities are under immense pressure to accelerate grid modernization efforts, streamline interconnection queues, and forecast demand with greater accuracy. They are exploring options for new generation capacity, including both conventional and renewable sources, as well as investing in advanced grid management technologies, energy storage solutions, and demand-side management programs to maintain reliability.
• Government and Regulatory Bodies (e.g., Department of Energy, FERC, State Public Utility Commissions): Policymakers are examining how to incentivize sustainable datacenter development, ensure energy security, and manage electricity prices. This could involve direct investments in infrastructure, regulatory frameworks for energy efficiency, or incentives for locating datacenters near renewable energy sources. The Inflation Reduction Act, with its provisions for clean energy development, may play a role in steering datacenter power sources.
• Tech Giants and Datacenter Developers (e.g., Google, Amazon, Microsoft, Meta, NVIDIA): Major technology companies are acutely aware of their environmental footprint and are increasingly committing to powering their operations with 100% renewable energy. They are investing heavily in more energy-efficient hardware, exploring advanced cooling technologies (like liquid immersion cooling), and strategically siting new datacenters in regions with abundant renewable resources or stable grid infrastructure. Many are engaging in direct power purchase agreements (PPAs) for renewable energy projects. Some are even exploring investments in small modular reactors (SMRs) or geothermal energy.
• Environmental Advocates: Groups concerned with climate change and resource scarcity are highlighting the environmental risks associated with increased energy consumption, particularly if it relies on fossil fuels. They are advocating for stringent energy efficiency standards, mandatory use of renewable energy, and responsible water management practices for datacenters, as cooling systems can also be water-intensive.
• Local Communities: The development of large-scale datacenters impacts local communities through increased energy demand, potential strain on local water supplies, land use changes, and noise pollution. There are growing calls for greater transparency and community engagement in datacenter siting decisions.

Broader Economic and Environmental Ramifications

The projected energy demands of AI datacenters carry significant broader implications:

• Economic Impact: Higher electricity costs could impact other industries and residential consumers. The sheer scale of investment required for new generation and grid infrastructure will be immense, potentially creating new economic opportunities in the energy sector but also posing significant financial burdens.
• Environmental Impact: If the increased electricity demand is met by fossil fuel-fired power plants, it will lead to a substantial rise in greenhouse gas emissions, undermining climate goals. The transition to renewable energy sources for datacenters is therefore not just an environmental imperative but also an economic necessity for long-term sustainability. Water consumption for cooling also presents an environmental concern, especially in drought-prone regions.
• Resource Scarcity: Beyond electricity, the construction of these datacenters and their specialized hardware requires vast amounts of raw materials. For instance, the original article’s referenced link points to a potential shortage of copper, a critical component in electrical infrastructure and datacenter cabling, underscoring the broader resource implications.
• Energy Security and Grid Reliability: A massive increase in concentrated load could strain grid stability, potentially leading to brownouts or blackouts if not managed effectively. Ensuring a resilient and reliable power supply for critical AI infrastructure becomes a matter of national security.
• Geopolitical Ramifications: The concentration of AI compute power in the US, and the energy required to sustain it, could influence international power dynamics, trade relationships, and global technological competition.

Mitigation Strategies and the Path Forward

Addressing the colossal energy demands of AI datacenters requires a multi-faceted approach involving technological innovation, policy intervention, and strategic investment:

1. Accelerated Renewable Energy Deployment: Large-scale investment in solar, wind, geothermal, and hydropower projects is paramount. Corporate Power Purchase Agreements (PPAs) by tech companies are already driving significant renewable energy development, but this must be scaled up dramatically.
2. Advanced Nuclear Power: Small Modular Reactors (SMRs) and other advanced nuclear technologies offer a promising pathway for reliable, carbon-free, baseload power, which could be ideal for powering large datacenters.
3. Grid Modernization and Energy Storage: Investments in smart grid technologies, enhanced transmission capacity, and grid-scale battery storage are essential to handle the intermittent nature of renewables and ensure grid stability under increased load.
4. Enhanced Energy Efficiency in Datacenters: Continued innovation in AI hardware design (more efficient chips), software optimization, and cooling technologies (e.g., liquid immersion cooling, waste heat recovery) can significantly reduce power consumption per unit of compute.
5. Strategic Siting: Locating new datacenters in regions with abundant renewable energy resources, robust grid infrastructure, and access to sustainable water sources can mitigate environmental and grid impacts.
6. Policy and Regulatory Frameworks: Governments can implement policies that incentivize energy efficiency, mandate renewable energy procurement for large loads, and streamline permitting processes for both renewable energy projects and critical grid upgrades. Carbon pricing mechanisms could also encourage cleaner energy choices.
7. Resource Management: Focusing on water-efficient cooling solutions and exploring closed-loop systems to minimize water consumption.
8. Decentralized Energy Solutions: Exploring microgrids and distributed generation projects that can provide power directly to datacenters, reducing strain on the broader grid.

The projected energy consumption of AI datacenters represents one of the most significant energy challenges of the 21st century. While AI promises transformative advancements across industries, its foundational energy requirements necessitate a proactive and concerted effort from governments, industry, and the energy sector to ensure a sustainable and resilient future for both technological progress and the planet. The next decade will be critical in shaping how the world powers its intelligence.