U.S. Data Center Power Outlook: Balancing competing power consumption needs

Surge in power demand following two flat decades

The commercialization of artificial intelligence (AI) and associated data center expansion is bringing rapid growth to previously flat U.S. power markets. The Boston Consulting Group (BCG) projects that total data center power demand will increase by 15-20% annually to reach 100-130 GW (800-1,050 TWh) by 2030. That’s the equivalent of the electricity used by about 100 million U.S. houses – about two thirds of the total homes in the U.S.

Data center power consumption, driven by a combination of AI, high performance computing, and traditional business computing, will have a lasting impact on U.S. power markets. The drivers of this growth indicate a sustained trend through 2030 and beyond as AI applications become integrated into daily operations, new power-intensive chipsets are developed, and new data center facilities are built.

Projections of data center-driven power demand continue to rise, and risk becoming an increasing cycle of hype. BCG conducted a detailed analysis, starting with the fundamental drivers of data center power demand (e.g., demand for computational services and power intensity of GPUs) and worked closely across our centers of industry expertise including our Technology, Media & Telecom, Principal Investors & Private Equity, and Energy practice areas.

We took two approaches to analyzing data center demand. For our first approach, we took a bottom-up look AI-related computational demand (160% increase from 2023-2028). Then, based on the expected performance of current and future hardware used for AI computing, we calculated the power required to meet this demand, subject to constraints in the semiconductor supply chain. For our second approach, we built a geospatial model of the U.S. data center development pipeline and determined the power required to serve the planned footprint.

Many assume that chipsets will demand less power as they become more efficient. However, although the compute power per watt of AI chipsets will continue to improve with each generation, this will be largely offset by increasing AI model complexity. This trend is as old as the steam engine – technological progress may increase a resource’s efficiency, but the lower relative cost leads to increased demand for the resource.

An inflection point for data centers and the power sector

As recently as 2023, the U.S. Energy Information Administration (EIA) forecasted annual power load growth of 1.5% from 2023 to 2025. Our analysis suggests growth in U.S. total power consumption will increase to 3% annually through 2030, reaching ~5,100 TWh annual demand. Data centers are projected to be the largest source of U.S. load growth. Data centers alone will contribute more than 60% of total growth, with the remainder projected to come from electrification of transport, buildings, and industrial operations.

The U.S. may face a shortfall of up to 80 GW of firm power to meet this demand by 2030, though gaps will vary in size across regional markets.

Closing this gap in firm power is a tall order. Not only will it require a step-change in capital deployment and the supporting investment conditions, but it also requires the ability to work through planning, permitting, and local regulatory challenges. Building new firm capacity will take years – which data centers do not seem to have given the opportunity cost of falling behind on advancements in AI. This is before we even start to layer in the commitments that many data centers (especially hyperscalers) have made to carbon free energy.

Therefore, data centers are having to navigate trade-offs across a range of considerations in this new environment:

• Supply sufficiency: Closing the potential shortfall in firm capacity would require significant incremental capital expenditures. However, generation assets are long-term investments with 20+ year lifecycles, and the risk of stranded capital is high. In a supply-constrained environment, data centers may need to consider and accept tradeoffs regarding the availability of power. Generation providers may need to retool to meet the challenge – some must rebuild dormant development skills, while others must balance market agility with greater financial discipline.
• Carbon intensity: The predominant sources of carbon free electricity currently available at scale are intermittent resources such as wind and solar power along with short duration battery storage. Going forward, as data center operators seek to both minimize greenhouse gas (GHG) emissions while also ensuring reliable supply they are increasingly likely to look to other sources of firm power generation. The most scalable, near-term firm and low-carbon option is natural gas with carbon capture and storage (CCS). We see CCS with natural gas as approaching technical readiness and economic feasibility, potentially making it a near-term generation source of choice for some data centers. Looking further out, other new low-carbon firm generation sources such as long duration energy storage (LDES) and modularized nuclear reactors also hold promise, though they remain costly and will take time to scale.
• Grid access: Timely grid connection is top of mind for data centers, and most are focused on solving for in-flight data center builds. Some customer segments are willing to pay premiums for expedited connection (up to 50%) and are even taking matters into their own hands and investing in private infrastructure (e.g., substations in addition to behind-the-meter generation solutions). Power utilities must consider how best to address underlying network constraints – navigating a fragmented planning landscape, cost allocation disputes, constrained supply chains, and financing challenges.
• Grid reliability: Increased intermittent generation on U.S. grids has created reliability concerns for operators requiring power for near-24/7/365 operations. As additional data centers are developed, grid operators must plan for effective onboarding of large new loads to ensure continued reliability and avoid overburdening increasingly strained networks. This will require managing an already tight supply chain and evolving planning processes to be more dynamic and cross-functional.
• Power pricing: Data centers seem to prefer long-term power contracts (e.g., ~10 years) and are willing to support innovative rate structures to secure stable pricing and support de-risking clean investments, but often have only traditional short-term options from utilities. In response, data centers are exploring alternatives to maintain predictable pricing, including self-generation and co-siting facilities with off-grid power generation. Utilities must partner with data centers on rate design to attract growth to their service territories without over-burdening the public on their power bills.

Each U.S. power market is unique. Successful answers to these challenges will vary regionally and require targeted design.

Fundamental shifts for energy players

The energy industry should adapt to the new realities and uncertainties associated with data centers and evolve its historical approaches to growth and capital allocation across the value chain. Four fundamental shifts will be required:

1. Sharpen commercial focus: Data centers want power providers to be more proactive. This requires developing perspectives on available projects and their cost – even if imperfect. Many utilities are not equipped to manage this dynamic shift in context, which requires balancing new objectives for stakeholders, e.g., managing potential revenue upside associated with new growth with a balanced view of cost and community engagement to manage needs across multiple stakeholders. Commercial innovation is a critical new muscle to flex for many utilities. For example, can you define preferred development zones, refine rate design, or innovate contractually to incentivize the right load in the right places?
2. Build agility and move quickly: “Time to power” is of the utmost importance to data centers, and they seek partners who can move at the pace of their business. Utilities must become agile in planning, work cross-functionally, adapt their risk processes and considerations to balance discipline and speed, and streamline governance to meet evolving consumer needs. For the first time, many regulated utilities are having to “compete” for customer load, including with non-utility players.
3. Partner with data centers: Data centers are willing to explore partnership. Utilities’ ability to offer technologically agnostic solutions (renewables, gas, nuclear, etc.) is top of mind but will require balancing risk manageably, equitably, and transparently. As already demonstrated, data centers will find ways to meet their own needs if they believe power providers will not. Providers must engage in give-and-take, while data centers may be asked to pay premiums to fund risk capital research and development for commercialization of new clean firm technologies that are not yet at scale and/or to plan for flexible demand in support of regional stability.
4. Collaborate with regulators: Data center load has not been welcomed with open arms in many markets, in contrast to industrial or manufacturing load. Energy companies and data center players must work together to define the benefits associated with data center development including economic development, jobs, and tax revenue, alongside improving transmission planning, refining rate structures, expediting investment and interconnection reviews, and incentivizing new technologies.

The path forward

The power sector stands at an inflection point and the time to act is now. Power providers have a generational opportunity to fundamentally rethink assumptions about supply, demand, risk, and capital allocation to serve customers, support communities, and win in a dynamic market.

The choices made today will shape the future of power markets, determining whether they can support an important emerging economic driver and provide efficient, equitable, and sustainable systems to secure a robust energy future for all.

Authors: Vivian Lee, Ross LaFleur, Michael McKissack, and Ganesh Swaminathan

For more, we will publish a more detailed version of this report and our underlying analyses in the coming weeks. Please follow and stay tuned.

Special thanks to additional experts and team members: Pattabi Seshadri, Clark O’Niell, Braden Holstege, Chris Richard, Alex Wright, Lauren Powers, Joao Arujo, Nate Johnson, Juliana Sanford, Yang Hao, Khushboo Goel, Jahnvi Vaidya, Leon Wang, Lauren Taylor, Kelly Kutas, Laura Togut, Alex Dewar, Mike Matson

See our prior analysis cited in the New York Times.