Power and Data Center Sectors Join Forces to Resolve Mounting Electricity Demand Uncertainties

The Electric Power Research Institute (EPRI) has launched an ambitious new initiative alongside power companies, grid operators, and several tech giants to establish five to 10 “flexibility hubs” that will demonstrate how data centers can be leveraged as flexible grid resources starting in the first half of 2025.

EPRI’s three-year Data Center Flexible Load Initiative (DCFlex) will demonstrate “how data centers can support and stabilize the electric grid while improving interconnection and efficiency,” the industry research organization says. “The Initiative will drive a cultural, taxonomic, and operational shift, creating a blueprint for data center stakeholders, utilities, market operators, technology innovators, and policymakers to adopt.”

The measure kicks off with 15 founding members listed, including several major power companies and utilities: Duke Energy, Pacific Gas and Electric (PG&E), Southern Co., Vistra Corp., and NRG Energy. Members also include data center operators and technology firms, including Google, Meta, Compass Datacenters, NVIDIA, and QTS Data Centers.

Grid participants include PJM Interconnection and the Electric Reliability Council of Texas (ERCOT), which are experiencing dramatic load growth fueled by data centers. “PJM’s current summer peak load forecast predicts a significant increase in load growth, with the share of total load attributable to data centers (currently 4%) expected to rise to 12% by 2030 and 16% by 2039,” said Aftab Khan, PJM executive vice president of Operations, Planning, and Security during the Federal Energy Regulatory Commission’s (FERC’s) annual Reliability Technical Conference on Oct. 16.

ERCOT in April estimated an additional 40 GW of load growth by 2030 compared to last year’s forecast, driven by large industrial projects, increased electrification, and the rapid expansion of data centers and cryptocurrency mining operations. At least one public entity, the New York Power Authority (NYPA), is listed as a founding member.

Mounting Uncertainties Poised to Afflict Both Power and Data Center Sectors

At the core of the issue is the uncertainty in power demand created by the rapid proliferation and expansion of data centers to support the growing requirements of artificial intelligence (AI), large language models (LLMs), and heightened computing power. Hyperscale facilities, built by companies like Amazon, Google, Meta, Microsoft, and OpenAI, now require massive power loads—often between 300 MW to 1,000 MW or larger—on accelerated timelines, with lead times of up to one to three years. That’s stretching the capacity of utilities, which must plan ahead, and often require four or more years to build new electric infrastructure to serve these significant point loads.

Utilities, power companies, and grid operators are already grappling with a range of uncertainties, steeped within what the North American Reliability Corp. has called a “hypercomplex risk environment.” While the industry has been bracing for higher demand from electrification of the transportation, residential, commercial, and industrial sectors, it is also confronting retirements of coal and gas dispatchable resources, driven by environmental regulations and policy. And, it is contending with existing transmission limitations, supply chain issues, and sustainability conflicts. The economics of new resources and new infrastructure costs add yet another layer to risk, and utilities, traditionally risk-averse, are wary of making long-term financial commitments without strong signals. AEP Ohio earlier this month introduced a structured tariff for hyperscalers that attempts to address some of these uncertainties.

Forecasting data center power demand growth is, meanwhile, a complex endeavor, with the added wildcard of data center efficiency improvements. Given notable efficiency advances in server and cooling technology improvements, the power usage effectiveness (PUE)—a measure that quantifies a data center’s energy efficiency—appears to have stabilized, averaging 1.6 in recent years, according to EPRI, but more improvements may be on the horizon. (Tech giants like NVIDIA, notably, have pushed for a new productivity-based metric that reflects real-world energy consumption per useful work.)

The Data Center Coalition (DCC), which represents several tech titans, told POWER it does not conduct load forecasting. “Utilities and RTO/ISOs are best positioned to conduct forecasting,” said Aaron Tinjum, DCC director of Energy Policy and Regulatory Affairs. “However, in a period of load growth, there should be greater communication, collaboration, and transparency with load forecasting to ensure that grid investments are based on robust and accurate data, right-sized, and that all customers are protected from any unnecessary costs,” he said. “To that end, DCC is engaging with utilities, grid operators, regulators, and other key stakeholders across the country to provide information on the data center industry—including growth drivers and siting considerations—and help inform forecasting efforts.”

EPRI, in a May 2024 white paper, explored four scenarios for potential data center load growth, combining estimates of increased data processing needs with assumptions about efficiency gains. It suggests that electricity usage by hyperscalers—data centers capable of rapidly scaling up their operations to meet the vast computing needs—more than doubled between 2017 and 2021. “This increase is expected to continue, with data centers projected to consume 5% to 9% of U.S. electricity generation annually by 2030, up from 4% today,” EPRI says. That boils down to energy consumption that ranges widely, from 214 TWh to 404 TWh.

A related EPRI survey of 25 utilities, including three from Canada and Europe, suggested that the power industry is already beginning to struggle with the integration of large new loads from data centers. The majority of responding utilities projected data center loads “to be a significant portion of their peak load in five years,” EPRI concluded. Almost half of respondents predicted that 10% or more of their five-year forecasted peak will come from data centers. In addition, EPRI noted data center interconnection requests “are trending to larger sizes, with 60% of responding utilities having requests for 500 MW or large and 48% with requests for 1,000 MW or larger,” the survey reported.

Already, 26% of responding utilities said they experiencing operational impacts from connected data centers, including thermal violations and voltage violations, related to ramp issues.” Currently, there is no industry consensus on incorporating these loads into long-term forecasts, which complicates grid planning, EPRI noted.

Flexibility A Core Pursuit at the DC Flex Initiative

Ultimately, what is becoming clearer is that flexible data center design and operation is emerging as a “key strategy for accelerating AI development and realizing its benefits while minimizing costs, lowering carbon emissions, and enhancing system reliability,” said EPRI President and CEO Arshad Mansoor.

Under EPRI’s vision, data center backup generators—powered by cleaner fuels than the diesel they currently rely on—could help shift the data center-grid relationship away from the current “passive load” model to a collaborative “shared energy economy,” it says. If grid resources powering data centers and data center backup resources can contribute to grid reliability and flexibility, they not only will help power entities “contend with the explosive growth of AI but also contribute to affordability and reliability for all electricity users,” it suggests.

“The goal is the complete integration of grid and data center power resources. Clean power generators co-located with data centers act as both grid and data center power sources,” it says. “During grid outages, these resources can seamlessly form a microgrid to provide uninterrupted power to data centers, eliminating the need and cost of standard diesel backup generators.”

Several stakeholders have underscored the urgency of this adaptive capacity. “Our energy system is built to handle the extreme demands of our hottest summer days and coldest winter nights but is often underutilized. The real challenge isn’t a lack of energy for data centers but managing the peak demand hours,” noted Joe Dominguez, Constellation’s president and CEO.

Tech firms agree. “As a nation, we cannot afford to fall behind on AI,” said Compass Datacenters CEO Chris Crosby. “We need to invest in sustainable infrastructure to support big data applications, but not at the expense of other consumers of electricity.”

The broad-based DCFlex collaboration will essentially pursue three workstreams. First, it will focus on creating adaptable data center design specifications that will be informed by grid needs for flexibility. It will also pursue the implementation of new “Transformational Utility” programs that incentivize data centers to serve as grid-supporting resources. This will include operator dispatch software modules for flexible load. Finally, and crucially, it will provide long-term planning models to ensure data centers can respond to varying load demands without compromising the grid. That measure will seek to produce data center forecast maps through 2030 and a granular load forecasting methodology for the evolving data center industry.

However, in addition, the initiative will deliver a “toolkit” to guide utilities, regulators, and data center operators in designing facilities that act as flexible grid resources. This should include best practices for economically integrating data centers with the grid and recommendations for utility programs that incentivize flexible operations to benefit all customers.

As a first step under DCFlex, EPRI, and its partners will establish five to 10 “Flexibility Hubs” across different locations. The hubs, designed to serve as “living laboratories” will test how data centers can operate flexibly and integrate with the grid. They will demonstrate innovative strategies under real-world conditions to provide an essential “blueprint” that stakeholders across the data center and utility sectors can adopt to expand flexible operations on a broad scale, it says. Demonstration deployment will begin in the first half of 2025, and testing could run through 2027.

The action items are informed by recommendations from the  U.S. Department of Energy’s Secretary’s Energy Advisory Board (SEAB) issued in July 2024. SEAB’s Working Group on Powering AI and Data Center Infrastructure  emphasized the need for closer coordination among utilities, technology firms, and policymakers to ensure high-power loads are managed efficiently, without compromising grid stability or affordability for other consumers. Among the body’s key recommendations is to encourage efforts to enhance data center energy efficiency, including the installation of advanced cooling technologies, power management, and server optimization. It also advises data centers to adopt optimized energy consumption and flexible load management practices that could contribute to grid peak load and critical stress management.

“Hyperscalers and technology providers state that temporal and spatial computational flexibility is possible if they are given appropriate signals,” the SEAB report notes. But, “Despite this perception of technical capability, we identified no examples of grid-aware flexible operation at data centers today other than the carbon minimizing geographic optimization that Google has employed for several years, recent efforts to respond to energy shortages in the European Union resulting from the Russian-Ukraine war, and flexibility requirements in Ireland,” it says. “This lack of flexible operations in the U.S. may result from the fact that electricity providers only recently started having to say no to data center interconnection requests.”

A Critical Cross-Industry Collaboration

Tech firms and power companies on Tuesday lauded the collaborative opportunity offered by the DCFlex inititative. Several companies suggested the effort aligns with their industries’ resilience and sustainability goals.

“AI can help solve the world’s greatest problems, and to enable this, NVIDIA is committed to our leadership in providing the highest performance per watt for AI computing infrastructure,” said Marc Spieler, head of global business development and strategy for the energy industry at NVIDIA. “The DCFlex initiative is another example of how industry leaders can leverage data centers as a flexible resource on the grid to help address peak loads.”

Bryce Dalley, director of Commercial Energy Supply at Meta, said that while the tech giant is committed to efficient data center operations matched by 100% clean and renewable energy, his company was aware “this work cannot be done alone.” Efforts like EPRI’s DCFlex initiative “are critical to these cross-industry efforts,” he said.

Caroline Golin, global head of Energy Market Development and Innovation at Google, hailed the initiative as a “generational opportunity for the public and private sector to work together to meet energy demand responsibly and unlock significant benefits for people, the economy and the planet.”

Power companies echoed these points. “As we serve our customers and meet their rapidly changing needs around reliability, affordability, and sustainability, we believe load growth from data centers presents a crucial opportunity to bring together stakeholders across the power and technology sectors to help solve some of the most complex grid management issues,” said Jim Burke, president, and CEO of Vistra. “Vistra is pleased to invest in EPRI’s DCFlex initiative and looks forward to working with other innovative companies to answer the call in ways that advance our national and regional interests and ensure the grid is reliable for all customers.”

—Sonal Patel is a POWER senior editor (@sonalcpatel, @POWERmagazine).