Perspectives on Energy and AI Data Centers

Clean Power and Industrial Efficiency, NCCETC Blog, Renewable Energy

By: Isaac Panzarella, Associate Director – Technical Services

With the boom of artificial intelligence (AI), data centers and the unconstrained integration of these computing systems in government, business and daily life will have long lasting implications from the global right down to the individual scale. A key issue is meeting huge power demands of proposed AI data centers. The direction we choose at this complex crossroads must consider local community impacts, broader energy affordability and long-term climate protection.

An event at NC State University, Workshop on Energy Needs for AI Data Centers, focused on meeting energy demands, grid integration challenges, technology pathways and policy considerations, explored the decisions we are faced with. Hosted by the FREEDM Center in the College of Engineering, the event brought together leaders from the data center industry, government, utilities and the academic research community.

A starting point for the discussion was the question of how much energy will be needed. Sasha Weintraub, executive vice president and chief customer officer for Duke Energy, presented several charts showing projected demand. One showed global demand for AI data center energy consumption growing 95% over a five-year period from 485 TWh in 2024 to 945 TWh by 2030. The International Energy Agency report, Energy and AI, that this chart was borrowed from further explains that this base case model shows that the sector will consume 3% of global energy in 2030. This seems like a limited amount, but the fact that it is new demand represents an opening to decide how and where the energy will be generated; onsite at the data centers or on the larger utility grid.

For that matter, we need to consider the scale of data centers being developed to perform AI workflows of training, fine-tuning and inference. Referred to as hyperscale, these data centers can start at 10 MW in electric demand, scaling to over 100 MW, and as much as 1 GW in demand. As examples, Amazon is building a site in Richmond County, North Carolina that will require up to 400 MW of electricity and Microsoft recently announced a 600 MW data center site in Person County, North Carolina.

The utility grid choice preference was one represented by Weintraub in a slide titled Advancing Carolinas Energy Transition which showed the following “near-term” resource plans, heavily weighted towards natural gas generation, based on Duke Energy’s 2025 integrated resource plan:

Seeing the timelines above for new grid generation capacity, it makes sense that the onsite generation choice for data centers, sometimes referred to “bring your own generation” (BYOG) is on the table. The onsite energy generation lineup includes similar technology and fuel choices as the grid approach, just at a scale customized to the specific site, and also adds fuel cells, geothermal and thermal energy storage to the mix. On a panel discussion at the workshop, Jim Smith, President of PowerSecure pointed out that to solve the capacity gap, [PowerSecure] can ramp up to meet the demands with a combination of generation and storage behind the meter while data centers wait for utility capacity, as well as transmission & distribution issues. After 5-10 years these assets could be repurposed to be grid assets, providing peak capacity as well as resilience for the data center.

Typically, the generation located at data centers may take the form of engines or turbines that spin electric generators to produce electricity. The state of Virginia recently passed legislation, HB323, that prioritizes heat capture and reuse from such generation at data centers. It is possible to develop such projects, such as the Joule Energy data center in Millard County Utah, which may scale up to 4 GW and operate independent of the power grid, and the DataOne Vineland, New Jersey data center which will be up to 300 MW. Both of these data centers are scarce examples of facilities using combined heat and power that will use a single fuel source to produce electricity and chilled water and reaching efficiencies over 70% (compared to utility scale gas turbines which are typically from 30-45% efficient). Another option would be to capture heat for distribution to district energy heating networks, sometimes located in urban districts, industrial parks or on university campuses.

Perspectives on AI capabilities and the importance of human intelligence and critical thinking in the energy infrastructure decisions we are faced with is something that dialogues like those at this workshop can help us keep a grasp on. According to the U.S. Department of Energy’s website “AI is a broad technology area that includes machine-based systems that can make predictions, recommendations, or decisions. These systems are trained to identify patterns in data, and they can then generalize to answer new questions, generate text or other media, or make autonomous decisions”. The AI data centers of today use 1000 times the energy that the human brain does to complete simple tasks; it may be faster, but nowhere near as efficient or elegant.

References

FREEDM Systems Center, Workshop on Energy Needs for AI Data Centers, https://engr.ncsu.edu/data-center-workshop/, March 2026.

International Energy Agency, Energy and AI, https://www.iea.org/reports/energy-and-ai/, April 2025.

Data centers; Department of Energy shall lead efforts to accelerate use of waste heat, report. https://lis.virginia.gov/bill-details/20261/HB323/text/HB323, March 2026

U.S. Department of Energy, https://www.energy.gov/topics/artificial-intelligence, accessed March 2026

Post navigation

Previous Post: Harness the Power of Connection & Innovation at the 2026 State Energy Conference of North Carolina

Next Post: Engineers Explore Powering Data Centers with Onsite Energy in NCCETC’s Energy Insights Webinar Series