Repowering Data Centers for AI: Managing Energy Demand and Sustainability

This is the third article in our multi-part Confronting AI series, brought to you by Mouser Electronics. Based on the Methods: Confronting AI e-magazine, this series explores how artificial intelligence is reshaping engineering, product design, and the ethical frameworks behind emerging technologies. Each article examines real-world opportunities and unresolved tensions—from data centers and embedded ML to regulation, adoption, and ethics. Whether you're developing AI systems or assessing their broader impact, the series brings clarity to this rapidly evolving domain.

"AI Everywhere" explains how AI extends beyond LLMs into vision, time series, ML, and RL.

"The Paradox of AI Adoption" focuses on trust and transparency challenges in AI adoption.

“Repowering Data Centers for AI” explores powering data centers sustainably for AI workloads.

“Revisiting AI’s Ethical Dilemma” revisits ethical risks and responsible AI deployment.

“Overcoming Constraints for Embedded ML” presents ways to optimize ML models for embedded systems.

“The Regulatory Landscape of AI” discusses AI regulation and balancing safety with innovation.

When considering critical infrastructure, systems like the electrical grid, water treatment, roads, and utilities likely come to mind. But what about data centers?

Data centers already act as critical infrastructure for all kinds of digital services. However, the emergence of artificial intelligence (AI) is taking the importance of these computing hubs to a whole new level, driving data centers' power consumption to a dramatically higher level.

Addressing this rising power demand will require a combination of efficiency improvements, renewable energy sources, and advanced power solutions. This article unpacks current trends in data center electricity consumption, AI’s impact on energy demand, what solutions and advanced technologies are available to meet this demand, and environmental concerns surrounding AI power consumption and sustainability practices.

AI’s Surging Energy Demands

Data centers already account for a large share of global electricity usage, consuming 460 terawatt hours (TWh) in 2022.^[1] This demand is only expected to increase as the need for digital services and AI grows.

For example, Goldman Sachs estimates data center power use will skyrocket 160% by 2030 due to AI workloads. This will put a major strain on the grid, with Goldman Sachs predicting that Europe will need to invest $1 trillion to prepare its power grid for AI.^[2]

It’s no secret that AI training and inference workloads are computationally demanding. Even with some of the earliest AI models, we’ve already seen a notable spike in data center energy consumption. Generative AI and large language models (LLMs) have significant energy requirements, posing sustainability challenges for data centers. 

But what does this energy consumption look like? Here are a few things to consider:^[3]

• AI systems can use around 33 times more energy to finish a task than traditional software.

• Training GPT-3 used around 1,300MWh of electricity, while GPT-4 used more than 65,000MWh. Specific training energy consumption figures for recent models like GPT-4o have not been disclosed; however, they are designed to be more efficient.

• Between 2012 and 2018, the computational power needed to sustain AI’s growth doubled every 100 days. While that pace has since slowed, compute requirements for training models continue to grow rapidly.

Recommended reading: How ChatGPT Is Transforming the Data Center

This surging electricity demand creates significant challenges for scaling existing infrastructure. Data centers face shortages in power equipment and limited availability of skilled tradespeople and technicians to support rapid expansion. These constraints may ultimately force secondary and emerging markets to increase energy production.^[4]AI searches are estimated to consume nearly 10 times more energy than traditional searches. However, updated figures indicate that typical queries now require significantly less power, driven by advances in model efficiency and improved data center performance.

Solutions for Growing Data Center Electricity Demand

Clearly, data centers and the broader market need fast solutions to stay ahead of surging energy demands. But this isn’t easy when you need to account for both the pace of growth and regional variety for energy production.

So far, we’ve seen investment in renewables and alternate energy sources, innovation in energy efficiency and cooling technologies, and the rise of demand response programs and AI-powered energy optimization.

Solution #1: Renewable and Alternative Energy Sources

Renewable energy isn’t a new concept for data centers. However, the accelerated pace of energy consumption has seen renewables become a sensible solution to meet electricity requirements.

Many data centers have turned to solar, wind, and geothermal solutions to offset grid dependency and reduce emissions. Google has already achieved 100% renewable energy and hopes to produce 24/7 carbon-free energy (CFE) by 2030.^[5] Tech rival Amazon has similar ambitions, with plans to power operations and AWS data centers with 100% renewable energy and achieve its net-zero carbon emissions target by 2040.^[6] Meanwhile, Dubai’s Moro Hub, one of the largest solar-powered data centers in the world, already offsets more than 13,800 tons of CO2 emissions annually with solar panels.^[7]Dubai’s Moro Hub data center clears the skies, offsetting carbon dioxide equal to the weight of a large fleet of jumbo jets each year.

There’s also a push to develop alternative energy sources to meet growing energy demands in the near and long term. Some of the largest tech companies have invested in alternative energy sources like hydrogen and nuclear reactors. For example, Microsoft recently announced a zero-emissions hydrogen pilot project for one of its data centers in Ireland.^[8] The company also agreed to repower Pennsylvania’s decommissioned Three Mile Island nuclear plant.^[9] 

AI power needs may even help the nuclear power industry level up. Several tech companies are exploring partnerships with companies that are pioneering small modular reactors (SMRs). These compact nuclear facilities can be built faster and at a lower cost than traditional nuclear plants while still providing consistent, carbon-free power. SMRs could prove to be game changers not only for data centers but also for the entire power grid.

Hydrogen fuel cells represent another promising frontier. These systems can provide primary and backup power while producing only water as a byproduct. Although hydrogen power is currently not cost-competitive with traditional power sources, investments in hydrogen production could make this technology a viable option for data centers seeking zero-emission alternatives.

Solution #2: Energy Efficiency and Cooling Technology Innovation

While increased energy consumption may be unavoidable, data centers have ways to limit their appetite for power. Among other things, data centers are investing in energy efficiency solutions, including advanced cooling systems, high-efficiency servers, and energy management software.

Cooling accounts for up to 40% of a data center’s energy consumption,^[10] and improved cooling technologies can make a big difference. For example, liquid cooling holds much promise: One study found that fully implementing liquid cooling can reduce facility power consumption by 18.1% and total data center power by 10.2% compared to traditional air cooling.^[11]

One of the more promising approaches is two-phase liquid cooling. This technology absorbs heat when a liquid changes into a gas to cool data center components. It offers an efficient way to manage high-heat loads with minimal energy consumption.Cooling systems consume nearly as much electrical power as the IT equipment they support.

The US Department of Energy is funding several of these innovative cooling projects, including the following:^[12]

• NVIDIA’s modular data center design, which combines direct-to-chip, pumped two-phase, and single-phase immersion cooling

• Intel’s development of a two-phase cooling system using a novel refrigerant mixture

Some concerns must be addressed for two-phase cooling to become common. This technology requires specialized dielectric fluids, and some candidate refrigerants raise environmental concerns.^[13] Also, implementing these systems may require significant changes to existing infrastructure.

Hardware advances are helping, too. Industry titan NVIDIA’s latest Blackwell graphics processing units (GPUs) promise 30 times more performance with 25 times lower energy consumption than its earlier models.^[14]

Finally, data center infrastructure management (DCIM) software can be critical in optimizing operations. It can monitor power use and environmental conditions in real time, helping operators quickly identify and address inefficiencies. For example, DCIM software can pinpoint “ghost servers” that are burning power even when not loaded. The tools can also help plan more efficient data centers, project server loading, and simplify employing techniques like hot/cold aisle containment.

Recommended reading: Improved electrical interconnects for data centers with higher-power delivery and smaller-footprint

Solution #3: Demand Response Programs, AI, and Automation for Power Optimization

Beyond efficiency improvements, data centers have become increasingly proactive in managing their energy consumption using AI and demand response programs. 

Demand response programs offer a policy-based approach to ensuring energy availability. These programs use time and location shifting to balance electricity demand—particularly during peak usage—and ensure grid stability. For example, data centers leveraging carbon-aware models can shift workloads to other regions with lower carbon intensity. Demand response programs can reduce energy costs for low-emissions energy by up to 34%. For example, Google now achieves a 64% share of carbon-free energy in their electricity consumption.^[15]

Of course, AI is already being used for power optimization. AI-driven temperature regulation is one solution to data centers’ energy consumption challenges. These systems use AI, sensors, and algorithms to monitor heat and adjust cooling in real time. AI aims to predict temperature fluctuations based on equipment usage, weather conditions, and other factors. This method conserves energy by avoiding overcooling and reduces wear on cooling equipment.

Environmental Concerns and Sustainable AI Practices

While increasing energy availability for data centers is critical to AI’s future, there’s no avoiding the impact this will have on the environment.

AI’s energy consumption is already well-observed. The impact of manufacturing advanced GPUs, clearing land to build data centers, and powering AI model training and inference—plus the natural resources consumed and emissions produced to create electricity—all affect the environment. None of this accounts for the social impact this has on local communities where emissions are released.

Many of these environmental concerns have paved the way for developing sustainable AI practices, including:

• Carbon offset programs: Specialized programs that allow companies to purchase carbon credits from projects that reduce, prevent, or isolate carbon emissions

• Eco-friendly AI model design: Lean and efficient models that require less computational power to train and deploy

• Energy-efficient hardware development: Developing custom hardware, energy-efficient server racks, and AI-optimized GPUs like NVIDIA’s Blackwell line

While these advancements are significant, they mean little without the right policies, regulations, and industry standards to guide and shape sustainable AI practices.

Despite being in the early stages of AI, lawmakers at all levels and global governments are discussing this issue. Singapore has announced a sustainability standard for data centers, while the European Commission has adopted a scheme for rating data center sustainability.^[16] Meanwhile, the US government is exploring ways to manage the impact of the country's AI energy use.^[17]

There are also many government programs and tax credits in the US, Europe, and other regions designed to incentivize developing renewable energy sources, offsetting carbon emissions, and investing in energy efficiency.Governments are beginning to respond to AI’s rising energy demands with new policies, sustainability standards, and incentives aimed at improving efficiency and promoting the use of renewable energy.

Balancing Power and Progress: The Future of AI-Driven Data Centers

Electricity demands for data centers, compounded with AI advancements, highlight the need for innovative power sourcing and management solutions. The current state of electrical grids will not be able to support AI’s exponentially growing power requirements.

To move forward, data centers need a combination of increased renewable energy production, advanced cooling technologies, and sophisticated power optimization. Each of these avenues will require significant investment and engineering advances, and the success of these efforts is far from assured.

AI’s future will require cooperation from federal, state, and local governments, technology companies, and utility providers to ensure the future is powered sustainably.

References

[1] Data Center Dynamics. Global data center electricity use to double by 2026 - IEA report: report [Internet]. 2024. Available from: https://www.datacenterdynamics.com/en/news/global-data-center-electricity-use-to-double-by-2026-report/

[2] Goldman Sachs. AI is poised to drive 160% increase in data center power demand [Internet]. 2024. Available from: https://www.goldmansachs.com/insights/articles/AI-poised-to-drive-160-increase-in-power-demand

[3] World Economic Forum. AI and energy: Will AI help reduce emissions or increase power demand? Here's what to know [Internet]. 2024 Jul. Available from: https://www.weforum.org/agenda/2024/07/generative-ai-energy-emissions/

[4] McKinsey & Company. How data centers and the energy sector can sate AI’s hunger for power [Internet]. 2024. Available from: https://www.mckinsey.com/industries/private-capital/our-insights/how-data-centers-and-the-energy-sector-can-sate-ais-hunger-for-power

[5] Data Center Knowledge. Data Center Power: Fueling the Digital Revolution [Internet]. 2024. Available from: https://www.datacenterknowledge.com/energy-power-supply/data-center-power-fueling-the-digital-revolution

[6] Amazon. Driving Climate Solutions [Internet]. 2024. Available from: https://sustainability.aboutamazon.com/climate-solutions

[7] Data Center Knowledge. Data Center Power: Fueling the Digital Revolution [Internet]. 2024. Available from: https://www.datacenterknowledge.com/energy-power-supply/data-center-power-fueling-the-digital-revolution

[8] Microsoft News. Microsoft announces pioneering green hydrogen pilot project with ESB [Internet]. 2024 Sep. Available from: https://news.microsoft.com/source/emea/2024/09/microsoft-announces-pioneering-green-hydrogen-pilot-project-with-esb/

[9] Reuters. Microsoft deal propels Three Mile Island restart, with key permits still needed [Internet]. 2024 Sep 20. Available from: https://www.reuters.com/markets/deals/constellation-inks-power-supply-deal-with-microsoft-2024-09-20/

[10] U.S. Department of Energy. DOE Announces $40 Million for More Efficient Cooling for Data Centers [Internet]. 2024. Available from: https://www.energy.gov/articles/doe-announces-40-million-more-efficient-cooling-data-centers

[11] Vertiv. Quantifying the Impact on PUE and Energy Consumption When Introducing Liquid Cooling Into an Air-cooled Data Center [Internet]. 2023. Available from: https://www.vertiv.com/en-us/about/news-and-insights/articles/blog-posts/quantifying-data-center-pue-when-introducing-liquid-cooling/

[12] Data Center Dynamics. Here are the 15 DOE-funded projects hoping to usher in the next generation of data center cooling [Internet]. 2023. Available from: https://www.datacenterdynamics.com/en/analysis/here-are-the-15-doe-funded-projects-hoping-to-usher-in-the-next-generation-of-data-center-cooling/

[13] Data Center Frontier. Understanding PFAS Concerns for Two-Phase Cooling of Data Centers [Internet]. 2024. Available from: https://www.datacenterfrontier.com/sponsored/article/33035570/understanding-pfas-concerns-for-two-phase-cooling-of-data-centers

[14] VentureBeat. Nvidia unveils next-gen Blackwell GPUs with 25X lower costs and energy consumption [Internet]. 2024. Available from: https://venturebeat.com/ai/nvidia-unveils-next-gen-blackwell-gpus-with-25x-lower-costs-and-energy-consumption/

[15] International Energy Agency. Electricity 2024: Analysis and forecast to 2026 [Internet]. 2024. Available from: https://iea.blob.core.windows.net/assets/6b2fd954-2017-408e-bf08-952fdd62118a/Electricity2024-Analysisandforecastto2026.pdf

[16] TIME. How AI Is Fueling a Boom in Data Centers and Energy Demand [Internet]. 2024. Available from: https://time.com/6987773/ai-data-centers-energy-usage-climate-change/

[17] U.S. House Energy and Commerce Committee. Energy, Climate, and Grid Security Subcommittee Hearing: "Powering AI: Examining America’s Energy and Technology Future [Internet]. 2024. Available from: https://energycommerce.house.gov/events/energy-climate-and-grid-security-subcommittee-hearing-powering-ai-examining-america-s-energy-and-technology-future

This article was originally published in “Methods: Confronting AI,” an e-magazine by Mouser Electronics. It has been substantially edited by the Wevolver team and Ravi Y Rao for publication on Wevolver. Upcoming pieces in this series will continue to explore key themes including AI adoption, embedded intelligence, and the ethical dilemmas surrounding emerging technologies.