Artificial intelligence has become essential in business, financial transactions, medicine, technology development, research, and more. Without realizing it, consumers rely on AI to stream videos, bank online, and perform online searches. Behind these capabilities are more than 10,000 data centers around the world, each of which is a huge warehouse containing thousands of computer servers and other infrastructure to store, manage, and process data. . There are currently more than 5,000 data centers in the United States, and new data centers are being built every day in the United States and around the world. Dozens of power companies are clustered in close proximity to where people live, often drawn by policies that offer tax breaks and other incentives and by the seeming abundance of electricity.
And data centers consume a lot of electricity. According to the Electric Power Research Institute, U.S. data centers will consume more than 4% of the nation’s total electricity in 2023, and that share could rise to 9% by 2030. One large data center can consume as much power as 50,000 homes.
The sudden need for so many data centers poses major challenges for the technology and energy industries, government policy makers, and everyday consumers. MIT Energy Initiative (MITEI) researchers and faculty are investigating various aspects of this problem, from power procurement to grid improvements to analytical tools that increase efficiency. Data centers have quickly become today’s energy problem.
Unexpected demands bring unexpected solutions
Several companies that use data centers to provide cloud computing and data management services have announced some surprising measures to power it all. Proposals include building its own small nuclear power plant near its data center and even restarting one of Three Mile Island’s undamaged nuclear reactors, which have been idle since 2019. (another reactor at that power plant partially melted down in 1979, causing the nation’s worst accident). The need to supply electricity to the United States has delayed planned shutdowns of some coal-fired power plants and raised prices for residential consumers. Meeting the needs of data centers not only strains the power grid, but also sets back the transition to clean energy needed to stop climate change.
From a power perspective, there are many aspects to the data center problem. Here are some of the things MIT researchers are looking at and why they’re important.
Unprecedented surge in electricity demand
“Before, computing didn’t consume much electricity,” says William H. Green, director of MITEI and the Hoyt C. Hottel Professor in MIT’s Department of Chemical Engineering. “Electricity has been used to run industrial processes, power domestic appliances such as air conditioning and lighting, and more recently to power heat pumps and charge electric cars. But now, suddenly, The power used in computing in general, and in data centers in particular, is becoming a huge new demand that no one expected.”
Why the lack of foresight? Electricity demand typically increases by about 0.5 percent per year, and utilities install new generators or make other investments as needed to meet the expected new demand. But data centers that are now coming online are causing an unprecedented surge in demand that carriers weren’t expecting. Moreover, new demands are constantly being created. It is important that your data center provides service all day, every day. You can’t interrupt the processing of large datasets, access to stored data, and the cooling equipment needed to keep all your packed computers running without overheating.
And even if enough power is generated, getting it where it’s needed can be a problem, explains MITI researcher Deepjyoti Deka. “The power grid is a network-wide operation, and while the grid operator may have sufficient generation capacity elsewhere or in other parts of the country, the power lines carry the power to where it is needed. ” Transmission capacity therefore needs to be expanded, but that process will take time, Deka said.
Next is the “interconnection queue”. In some cases, adding a new user (a “load”) or a new generator to an existing grid can cause instability or other issues for everyone already on the grid. In such situations, there may be delays in bringing new data centers online. If the delay is too long, new loads or generators may have to wait their turn in line. Currently, much of the interconnection queue is already filled with new solar and wind projects. The current delay is approximately five years. Meeting the demand from newly installed data centers while ensuring that quality of service elsewhere is not disrupted is an issue that needs to be addressed.
find clean power
Further complicating the challenge, many companies, including so-called “hyperscalers” such as Google, Microsoft, and Amazon, have pledged to reduce their carbon emissions to net zero within the next 10 years. is. Many people are making progress toward achieving their clean energy goals by purchasing a Power Purchase Agreement. They may, for example, enter into contracts to purchase power from solar or wind farms and finance the construction of the facilities. However, this approach to harnessing clean energy has its limits when faced with the extreme power demands of data centers.
Meanwhile, many states are delaying the closure of coal-fired power plants as electricity consumption soars. There just aren’t enough renewable energy sources to serve both hyperscalers and existing users, including individual consumers. As a result, the need for conventional plants powered by fossil fuels such as coal is greater than ever.
As hyperscalers seek clean energy sources for their data centers, building their own wind and solar facilities could be an option. However, such facilities can only generate electricity intermittently. Given the need for uninterruptible power, data centers must maintain expensive energy storage units. Alternatively, they can rely on natural gas or diesel generators for backup power, but they do not require equipment to capture carbon emissions and a nearby location to permanently dispose of the captured carbon. must be combined with
Because of these complexities, some hyperscalers are turning to nuclear power. “Nuclear power plants can reliably generate large amounts of electricity without interruption, making nuclear energy a good fit for data center demands,” said Green.
In a highly publicized move in September, Microsoft announced that Constellation Energy would be purchasing an undamaged nuclear power plant on Three Mile Island, the site of a highly publicized nuclear disaster. After restarting one of the reactors, a contract was signed to purchase electricity for 20 years. 1979. If it receives regulatory approval, Constellation plans to have its reactor online by 2028 and buy all the power Microsoft produces. Amazon also reached a deal to buy electricity produced by another nuclear power plant, which is on the verge of closure due to financial difficulties. And in early December, Meta announced a call for proposals to identify nuclear energy developers to help the company meet its AI needs and sustainability goals.
In other nuclear news, small modular reactors (SMRs), which can be built near data centers and potentially avoid the cost overruns and delays often experienced in building large plants, are Focuses on modular power plants. Google recently ordered a fleet of SMRs to generate the power needed for its data centers. The first construction is expected to be completed by 2030, and the rest by 2035.
Some hyperscalers are betting on new technology. For example, Google is pursuing a next-generation geothermal project, and Microsoft has signed a deal to buy power from the startup’s fusion power plants starting in 2028, even though fusion technology is not yet proven. did.
Reducing electricity demand
Other approaches to providing sufficient clean power focus on making data centers and the operations they house more energy efficient so that they can perform the same computing tasks with less power. . The use of faster computer chips and the optimization of algorithms that use less energy are already helping to reduce load and generate less heat.
Another idea being tried involves moving computing tasks to times and places on the grid where carbon-free energy is available. Deka explains: “If a task doesn’t need to be completed right away but needs to be completed by a certain deadline, it can be delayed or moved to a data center in the U.S. or overseas where power is plentiful and cheaper. ?Cleaner? This approach is known as “carbon-aware computing.” ”It remains to be seen whether all tasks can be easily moved or delayed, Deka says. “If you think about an AI-based generative task, can you easily break it down into smaller tasks that can be taken to different parts of the country, solved using clean energy, and then put back together? How much does it cost to split tasks?”
Of course, this approach is limited by interconnection queue issues. Accessing clean energy in different regions and states is difficult. However, efforts are underway to ease the regulatory framework to allow critical interconnections to be developed more quickly and easily.
What about the neighbors?
A major concern common to all options for powering data centers is the impact on residential energy consumers. When data centers are located in close proximity, there are not only aesthetic concerns, but also more practical ones. Will local electricity service become less reliable? Where will new transmission lines be installed? And who will pay for new generators, upgrades to existing equipment, etc.? New manufacturing facilities or industrial plants? When a neighborhood is built, the downside is usually offset by new job security. This is not the case with data centers, which require only a few dozen employees.
There are standard rules for how maintenance and upgrade costs are shared and distributed. However, the situation changes completely with the presence of new data centers. As a result, utilities are now having to rethink their traditional pricing structures to avoid placing an undue burden on residents to pay for the infrastructure changes required to host data centers.
MIT contribution
At MIT, researchers are considering and exploring different options to tackle the problem of providing clean power to data centers. For example, we are researching architectural designs that utilize natural ventilation to increase cooling, equipment layouts that allow for better airflow and power distribution, and energy-efficient air conditioning systems based on new materials. They are creating new analytical tools to assess the impact of data center deployments on the U.S. power system and find the most efficient ways to provide clean energy to facilities. Other research is looking at how to match the power output of small nuclear reactors to data center needs and how to accelerate the construction of such reactors.
The MIT team also considered power and water availability, regulatory considerations, and potential to determine optimal backup power and long-term storage, as well as proposed new data centers. We also focus on developing decision support systems to find out. This is to harness heat that could be valuable waste heat, for example for heating nearby buildings. Technology development projects include designing faster and more efficient computer chips and energy-efficient computing algorithms.
In addition to providing leadership and funding to many research projects, MITI serves as a convener, bringing together companies and stakeholders to address this issue. At MITEI’s 2024 Annual Research Conference, a panel of representatives from two hyperscalers and two companies that jointly design and build data centers will discuss their challenges, possible solutions, and where MIT’s research will be most beneficial. We discussed the place.
As data centers continue to be built and computing continues to drive unprecedented increases in power demand, scientists and engineers are racing to deliver ideas, innovations and technologies to meet this need, Green says. . The transition to a decarbonized energy system will continue to advance.
