Data Centers, Generative AI, & The Power Struggle

Altman Solon is the largest global telecommunications, media, and technology consulting firm. Below, Partner Charles Cieutat and Manager Elisabeth Sum break down how data center operators are dealing with power constraints in an era of generative AI adoption.

The data center industry is at an inflection point, and power is at the heart of this inflection point. Generative AI adoption across enterprises has thrust data center (DC) power demand in the U.S. into overdrive. According to a Lawrence Berkeley National Laboratory study, DC energy use in the U.S. has doubled over the past three years, driven by continued cloud expansion and generative AI growth, and significantly more is expected in the next few years. Similar trends are also affecting DC markets in Europe.

AI hardware requires a lot of power through much higher rack power density. Utilities can’t keep up with this explosive, sudden demand given deficient planning and capacity (a study found that six out of 10 major grid regions scored low in capacity planning), aging and under-dimensioned transmission infrastructure, and lengthy time frames to develop new power generation and interconnections due to labor and equipment shortages. As a result, many data center buildouts have stalled due to the three-to-seven-year lead times to procure utility power.

This is particularly true for major Tier 1 markets like Northern Virginia and Silicon Valley, where energy constraints have become the largest bottleneck to developing data centers. Because of this, operators and investors have been forced to explore new opportunities and arrangements to secure power. These include scouting out locations with less grid congestion, adaptive reuse, and "behind the meter” (BTM) and other power source solutions.

Scouting new locations

Power constraints and the location-agnostic nature of some AI workloads have contributed to shifting data center construction out of Tier 1 metropolitan areas and into new Tier 2 and 3 markets like Charlotte, North Carolina; Minneapolis, Minnesota; and Reno, Nevada. In Europe, these markets would include Nordic countries and Southern European metropolitan areas like Madrid and Milan, among others. These geographies historically had small retail and enterprise-oriented data center footprints, and their grids are often unencumbered.

Adaptive reuse

Repurposing existing sites or converting old power stations is gaining popularity. Data center operators are looking at closed industrial sites, nuclear sites, enterprise-owned data centers, crypto mining data centers, office buildings, warehouses, and retail stores for reconversion. Decommissioned coal power stations are particularly attractive, as they have attributes needed for a data center campus, like high-power usage. Adaptive reuse allows data center operators to tap into the existing power infrastructure of the original building, decreasing the time to full operation. On the flip side, adaptive reuse presents new challenges, such as retrofitting sites to meet data centers' resiliency, power density, cooling, and security requirements, which are heightened in an AI world.

Behind the meter solutions and moving the power source

Some operators are bringing their data centers directly to power sources or indirectly through Power Purchase Agreements. Indeed, on-site or near-site energy generation can ease the power burden. In the same vein, operators are experimenting with on-site microgrids powered by natural gas to provide additional power and run as a supplement to the traditional power grid. Major scalers are also interested in small nuclear reactors (SMR), but these will likely be available by the next decade at the earliest.

BTM battery energy storage systems are also emerging as pivotal tools for data centers, mainly to bridge imminent demand for data center capability. BTM systems store energy (either from the grid or renewables) and can be used as backup power when needed.

These novel power solutions mean that data centers will need to tackle new problems, including how to build, operate, and maintain a local power plant; how to manage regulatory issues (e.g., FERC approval, transmission and distribution fees, emissions); and utilities (e.g., fuel procurement, power pricing, electricity arbitrage).

Complete the form to receive our in-depth report on the evolving data center landscape and the impact of generative AI on power demands.

Altman Solon's data center expertise

Altman Solon has expanded its existing data center and analytics innovation capabilities to help operators and investors navigate this power crunch. Our services include:

• Performing commercial due diligence (CDD) on all segments of data center services
• Performing technical due diligence (TDD) of data center power assets; estimating regional power security likelihood with proprietary AI tools, build plans, and capital requirements.
• Assessing utility exposure risks and mitigation strategies
• Assessing local electric utilities to meet power obligations for planned data center deployments
• Leveraging proprietary generative AI and behind the meter data center forecasts
• AI geospatial mapping of nearby power grid assets (e.g., substations, transmission & distribution lines, natural gas pipelines, power plants)
• Assessing the quality of nearby power grid assets (e.g., renewable mix, power price, etc.)
• Strategic, commercial, and technical viability review of alternative power source plans such as gas plants, wind farms, and risk mitigation
• Performing detailed site-by-site analysis of wind farm asset compatibility with either greenfield DC builds or BTM solutions
• Contract review
• Stabilized and Development Companies (“DevCo”) joint ventures