The Impact of GenAI on Electricity: How GenAI is Fueling the Data Center Boom in the U.S.

Data centers are the modern nerve centers of the digital world, and play a critical role in the storage, management, and delivery of the vast amounts of information that power our interconnected society. This includes cloud computing, big data processing, AI algorithms and, more recently, complex Generative AI (“GenAI”) models. Data center development continues to grow significantly year-over-year to meet the increased demand for digital services driven by advances in technology. The rapid adoption of GenAI-related tools across many industries has accelerated the growing demand for digital services and data centers. Much of this growth will occur in hyperscale data centers – a specialized category of data centers designed to power immense amounts of digital information and computational tasks (including GenAI) at an unprecedented scale.

This growth in demand for data center services, particularly for GenAI, is driving up power usage and density. Data center electricity consumption was 2.5% of the U.S. total (~130 TWh) in 2022 and is expected to triple to 7.5% (~390 TWh) by 2030. That’s the equivalent of the electricity used by about 40 million U.S. houses – almost a third of the total homes in the U.S. 

GenAI is expected to account for at least 1% of this increased power usage.

The projected uptick in incremental electricity usage for GenAI is driven by two key factors:

1. Training GenAI models: With the emergence of new Gen AI use cases around text, image, and videos, we expect an accelerated increase in large models that will be trained and re-trained on a daily basis in the next five years. This electricity-intensive training is projected to use ~400-1300 kWH per day per model.
2. Servicing higher GenAI queries: For example, ChatGPT uses 3-30 times more electricity than a Google search query and often leads to follow-up queries (thus also increasing the overall volumes of queries)

Given the magnitude of electricity needed to power data centers, the question arises: Where will these facilities be located?

Currently, most of the data centers in the U.S. are in PJM and MISO. New builds are also expected in CAISO and Southeast regions. Combined, these four ISOs / regions, will likely host more than 60% of the total data center capacity in the U.S. by 2027. Primary locations for new builds include Ohio, Illinois, California, Virginia, Florida, and Georgia. Decisions about where to build data centers are based on a number of pivotal criteria, including access to lower cost power and energy infrastructure and favorable state tax incentives. Additionally, reliable telecommunications infrastructure is key to ensuring low latency enabling data centers to be located further away from actual use. Environmental factors – such as cooler weather conditions and limited exposure to natural disasters – also play a role.

However, there are several emerging risks and concerns surrounding data centers, including:

• Energy availability: High electricity demand from data centers, especially in already congested areas, is making it hard for the grid to service the required needs. For example, in February, PJM warned that the number of high-demand data centers in Virginia was “driving unsustainable growth in energy demand.”
• Energy reliability: Data centers need 24/7 electricity. While there have been attempts to support battery electricity storage in data centers, an economically viable solution has yet to emerge.
• Public sentiment: Stakeholders have raised concerns about the level of carbon emissions from data centers, which account for 2% of CO2 emissions globally, and the amount of water they need to operate, especially in areas affected by recent droughts.

Data center owners (primarily major tech companies and colocation providers) have set ambitious clean energy goals to manage these emerging risks. These objectives include:

• Moving away from RECs to PPAs/VPPAs focused on new green energy builds
• Using more self-generated energy and local backup (batteries) where feasible vs grid-based electricity
• Pursuing 24/7 load matching  

The expected growth of GenAI and digital technologies create an opportunity for renewable developers and utility players to work with key data center players and provide innovative green energy solutions to help shape the future of data center development in the U.S. These efforts have the potential to make a sizable impact on economic development and support decisive climate actions to reduce carbon emissions and safeguard energy and water supplies.

For more, you can reference additional details on BCG on Energy page here.

Authors: Vivian Lee , Ross LaFleur , Khushboo Goel , Bassem Khoury