Artificial Intelligence (AI) has become indispensable across various sectors, including finance, healthcare, technology, and research. From streaming videos and banking online to conducting searches, consumers unknowingly depend on AI every day. This reliance is supported by a vast network of over 10,000 global data centers, which serve as colossal repositories encompassing thousands of computer servers and infrastructure for data storage, management, and processing. In the United States alone, there are over 5,000 data centers, with new facilities springing up daily, motivated by favorable tax policies and the perceived availability of abundant electricity.
However, the energy demands of these data centers are staggering. In 2023, U.S. data centers consumed more than 4% of the nation’s overall electricity usage, and this number could surge to 9% by 2030, based on projections from the Electric Power Research Institute. To put this into perspective, a single large data center can use as much electricity as 50,000 households.
The rapid growth in data center construction presents significant challenges for the technology sector, energy providers, policymakers, and everyday consumers. Researchers and faculty at the MIT Energy Initiative (MITEI) are investigating numerous aspects of this dilemma—from energy sourcing and grid enhancements to analytical tools that promote efficiency. Data centers have rapidly emerged as a critical energy concern today.
Rising Demand Sparks Innovative Solutions
A number of cloud computing and data management companies are taking unexpected measures to meet the energy requirements of data centers. Some proposals include constructing small nuclear plants adjacent to data centers and even reviving one of Three Mile Island’s undamaged reactors, which has been inactive since 2019. The pressing need to supply energy for AI has led to delays in shutting down certain coal-fired power plants and is increasing electricity costs for residential users. As data centers exert pressure on the power grid, they also hinder the transition to clean energy needed to combat climate change.
There are multiple dimensions to the data center energy issue that MIT researchers are examining. Here’s a closer look at their importance:
A Historic Surge in Electricity Demand
“Historically, computing was not a major electricity consumer,” explains William H. Green, MITEI Director and the Hoyt C. Hottel Professor in MIT’s Chemical Engineering Department. “Electricity was primarily utilized for industrial processes and everyday household appliances. Recently, however, the demand from data centers for energy has skyrocketed, introducing unexpected challenges.”
This surge in demand stems from the unique operational needs of data centers, which require constant electricity for data processing, access, and crucial cooling systems, making interruptions unacceptable.
Moreover, even when sufficient electricity generation is available, the distribution network might struggle to deliver it where needed. “The grid operates as a collective unit,” says MITEI research scientist Deepjyoti Deka. “While generation may be adequate in one location, the existing infrastructure may lack the capacity to transmit electricity to data centers efficiently.” Expanding transmission capabilities is essential, yet it’s often a slow process.
Additionally, the interconnection queue poses complications; as new users or generators are added, stability issues may arise, causing delays in bringing new data centers online. Currently, this queue is heavily occupied with newly proposed solar and wind projects, resulting in a backlog of about five years.
Seeking Clean Energy Sources
To exacerbate the issue, many major companies such as Google, Microsoft, and Amazon have publicly committed to achieving net-zero carbon emissions within the next decade. While they are progressing toward these clean energy goals through Power Purchase Agreements (PPAs), where they finance renewable facilities, this solution has limits given their enormous electricity needs.
This growing power demand is stalling the closure of coal plants, as insufficient renewable energy sources are unable to satisfy the needs of both these large-scale users and everyday consumers. As a result, fossil fuel-powered plants are increasingly relied upon.
To meet the cleanliness and supply demands for their data centers, hyperscalers may consider building onsite renewable energy sources. However, solar and wind solutions provide only intermittent power. This would necessitate costly energy storage or backup generators that emit carbon unless combined with carbon capture solutions.
Thus, many hyperscalers are exploring nuclear energy as a reliable option. Green notes, “Nuclear energy is ideally suited to data centers due to its ability to provide substantial, dependable power without interruptions.” In a recent deal, Microsoft agreed to purchase power from a recently revived reactor at Three Mile Island, which is set to come online in 2028. Amazon and Meta are also pursuing nuclear energy contracts to support their operations.
In addition to traditional nuclear reactors, there’s emerging interest in small modular reactors (SMRs), which can be constructed near data centers with fewer delays and cost overruns. Google has ordered SMRs intended to supply power for its data centers, with the first unit expected by 2030.
Hyperscalers are also experimenting with advanced technologies. Google is developing next-generation geothermal projects while Microsoft plans to acquire electricity from an experimental fusion power plant commencing in 2028, even as this technology remains unproven.
Reducing Electricity Demand
To alleviate some of the energy pressures, efforts are being made to enhance the energy efficiency of data centers, allowing them to perform tasks while consuming less power. Utilizing faster chips and optimizing energy-efficient algorithms are already yielding positive results.
Another strategy involves shifting workloads to times and locations where carbon-free energy is more plentiful. Deka elaborates, “If a task has a deadline but isn’t urgent, could it be postponed or rerouted to a cleaner energy data center?” This concept, known as “carbon-aware computing,” presents potential but also faces limitations due to the interconnection queue challenges.
Community Impact Considerations
The introduction of data centers in residential areas raises significant concerns. Residents may wonder about the reliability of their local electricity supply, the placement of new transmission lines, and the costs associated with necessary upgrades. While new businesses typically bring job opportunities, this isn’t the case with data centers, which often require minimal staffing.
Utilities are facing the challenge of rethinking traditional cost-sharing structures to avoid unduly burdening local residents, as the infrastructure required to accommodate data centers can shift the financial responsibility significantly.
MIT’s Innovations
Researchers at MIT are actively exploring numerous strategies for supplying clean energy to data centers. They are looking into architectural designs that utilize natural ventilation, efficient layouts that enhance airflow, and innovative cooling systems. New analytical tools are being developed to assess the impacts of data center deployment on the energy sector and to determine effective clean energy sourcing.
Moreover, MIT is investigating efficient energy storage solutions and optimal backup power sources, while also working to streamline regulations for critical interconnections. As demand for computing surges, rapid advancements in technology and innovative approaches will be essential to meet energy needs while transitioning toward a decarbonized system.
As data centers continue to proliferate and computing demand escalates, scientists and engineers are racing against the clock to develop solutions that will satisfy this burgeoning need while promoting a sustainable energy future.
Photo credit & article inspired by: Massachusetts Institute of Technology
