Related Articles
The rapid growth of artificial intelligence (AI) is placing unprecedented demands on U.S. power infrastructure, raising questions about the sustainability of this growth. Our recent analysis evaluates the risk of power availability for AI, projects potential deficits, and identifies measures to help mitigate these constraints.
Background
U.S. utilities operate within strict regulatory frameworks, limiting their flexibility to add capacity quickly. This approach has sufficed historically, as U.S. power demand remained largely stagnant—rising only 4 percent from 2013 to 2023. However, data centers have recently accelerated their energy consumption, now accounting for approximately 3 percent of the nation’s total. This figure is expected to increase with AI adoption.1
AI-related power usage grew threefold last year, reaching about 1 percent of total U.S. electricity demand.2 This rapid growth heightens the risk of power deficits and necessitates a proactive response from utilities.
Key Findings
- Manageable Deficits: AI-driven power deficits will likely emerge regionally but should remain manageable through 2027.
- Chip Power Efficiency is Critical: Advances in GPU efficiency are essential to meeting AI’s future energy needs.
- Utility Preparedness: Utilities have a window to adapt, potentially mitigating worst-case scenarios.
- Stop-gap Measures: Temporary solutions can address deficits until new baseload power becomes available.
In summary, while power deficits could present long-term challenges, we believe they are manageable in the medium term. Assuming continued chip efficiency gains, power constraints are unlikely to significantly hinder AI growth in the United States.
Projected Power Deficits
Using a conservative model, we predict AI-driven power deficits could emerge as soon as 2025, initially accounting for 0.9 percent of U.S. power demand. This figure could grow to 1.9 percent by 2028 and 3.2 percent by 2030. These deficits correspond to shortfalls of 4–12 GW between 2025 and 2027, increasing to 16–42 GW by 2030.
These figures, while material, are far less extreme than some projections suggesting the need for hundreds of additional gigawatts. For context, the average U.S. coal plant has a capacity of 400–500 MW, while a nuclear plant typically ranges from 1 to 1.2 GW.3
Utilities can address these deficits with existing measures such as demand curtailment, demand response, and increased baseload power utilization. Nationally, baseload utilization sits at 44 percent, down from 48 percent two decades ago, offering room for modest increases.4
Mitigation and Adaptation Strategies
Advances in Chip Efficiency
Innovations in chip architecture and AI model design have driven significant gains in power efficiency. For instance, training energy costs for advanced AI models like GPT-4 have decreased by 99 percent since 2016.5 Emerging technologies, such as Gate-All-Around Field-Effect Transistors (GAAFETs) and chiplets, promise further efficiency improvements, albeit at a potentially slower rate due to atomic-level physical limitations.
These advances allow AI compute to grow exponentially while power consumption increases more linearly, providing utilities time to plan capacity expansions in an orderly manner.
Temporary Solutions
Several interim strategies can bridge the gap until new infrastructure becomes operational:
- Data Center Site Selection: Building data centers in power-surplus regions like Ohio, Texas, and Atlanta can alleviate strain on high-demand areas, such as Northern Virginia and California.
- Offshore Model Training: AI model training, which is not latency-sensitive, can shift to regions with more affordable or available power.
- Delayed Plant Retirements: The U.S. Energy Information Administration (EIA) estimates 69 GW of coal and natural gas capacity will retire in the next decade; delaying retirements could help offset deficits.
- On-site Power Generation: Data centers may deploy containerized natural gas generators, bypassing grid connection delays.
- Increased Natural Gas Utilization: Although natural gas utilization is expected to
decline as renewables come online, utilities may temporarily leverage this capacity to support power needs. - Grid-scale Power Storage: Long-term, grid batteries can store off-peak energy for use during high-demand periods, improving baseload power utilization.
AI Power Constraints Are Significant but Manageable
Our analysis suggests that power constraints represent a notable but addressable challenge for AI growth. While deficits may emerge in the late 2020s, utilities have time to adapt by deploying a mix of temporary solutions and capitalizing on efficiency gains. Although grid investments and capacity additions will be necessary, the scope of these efforts may be less extensive than widely assumed.
1 Sands Capital research and https://www.eia.gov/consumption/commercial/data/2018/pdf/2018_CBECS_Data_Center_Pilot_Results.pdf. Most recent data as of September 30,2024.
2 https://www.goldmansachs.com/insights/articles/AI-poised-to-drive-160-increase-in-power-demand
3 https://www.energy.gov/ne/articles/nuclear-power-most-reliable-energy-source-and-its-not-even-close
4 https://www.eia.gov/todayinenergy/detail.php?id=60984
5 https://openai.com/index/ai-and-efficiency/
Disclosures:
The views and opinions expressed herein are those of individuals and may differ from the views and opinions expressed by Sands Capital. They are not intended as a forecast, a guarantee of future results, investment recommendations, or an offer to buy or sell any securities. The views expressed were current as of the date indicated and are subject to change.
This material may contain forward-looking statements, which are subject to uncertainty and contingencies outside of Sands Capital’s control. Readers should not place undue reliance upon these forward-looking statements. There is no guarantee that Sands Capital will meet its stated goals.
References to “we,” “us,” “our,” and “Sands Capital” refer collectively to Sands Capital Management, LLC, which provides investment advisory services with respect to Sands Capital’s public market investment strategies, and Sands Capital Ventures, LLC, which provides investment advisory services with respect to Sands Capital’s private market investment strategies, including the Global Innovation strategy, which is available only to qualified investors. As the context requires, the term “Sands Capital” may refer to such entities individually or collectively. As of October 1, 2021, the firm was redefined to be the combination of Sands Capital Management, LLC and Sands Capital Ventures, LLC. Both firms are registered investment advisers with the U. S. Securities and Exchange Commission in accordance with the Investment Advisers Act of 1940. The two registered investment advisers are combined to be one firm and are doing business as Sands Capital. Sands Capital operates as a distinct business organization, retains discretion over the assets between the two registered investment advisers, and has autonomy over the total investment decision-making process.
Information contained herein may be based on, or derived from, information provided by third parties. The accuracy of such information has not been independently verified and cannot be guaranteed. The information in this document speaks as of the date of this document or such earlier date as set out herein or as the context may require and may be subject to updating, completion, revision, and amendment. There will be no obligation to update any of the information or correct any inaccuracies contained herein.
Notice for non-US Investors