Multi-faceted Challenges of Dynamic AI | MIT News

Artificial intelligence has become crucial in business and financial transactions, healthcare, technology development, research, and more. Without realizing it, consumers rely on AI when streaming videos, doing online banking, or conducting online searches. Behind these capabilities are over 10,000 data centers around the world, each with a huge repository containing thousands of computer servers and other infrastructures for storing, managing and processing data. There are now more than 5,000 data centers in the United States, and new data centers are built every day in the United States and around the world. Often, dozens of people gather near where people live, attracted by policies that provide tax breaks and other incentives, and look like a rich power.

Data centers do consume a lot of electricity. According to the Electric Power Institute, U.S. data centers consume more than 4% of the country’s total electricity in 2023, and by 2030, that percentage could increase to 9%. A large data center can consume as much electricity as 50,000 homes.

The sudden demand for so many data centers presents a huge challenge to the technology and energy industry, government policy makers and everyday consumers. Research scientists and faculty at the MIT Energy Program (MITEI) are exploring multiple aspects of this problem – from procurement capabilities to grid improvements to analytical tools for improving efficiency, and more. Data centers have quickly become our energy problem today.

Unexpected demand brings unexpected solutions

Some Companies that use data centers to provide cloud computing and data management services are announcing some surprising steps to deliver all power. Proposals include building its own small nuclear power plant near the data center and even restarting one of the three-mile island’s undamaged nuclear reactors that have been shut down since 2019. (Another reactor at the plant partially melted in 1979, and in 1979, which led to the sale of coal caused by AI and caused some coal prices and sold on some coal. Meeting the demand for data centers is not only about emphasizing the power grid, but also about the resumption of the transition to the clean energy needed to stop climate change.

From a power perspective, there are many aspects of data center issues. Here are some of the key focus of MIT researchers and why it is important.

An unprecedented surge in demand for electricity

“In the past, computing was not a significant user of electricity,” said William H. Green, professor at MITEI’s Mitei and Hoyt C. Hotel Hottel. Electricity is used to run industrial processes and power household equipment such as air conditioners and lighting, and has recently been used to power heat pumps and recharge electric vehicles. But now suddenly, electricity used for computing, especially in data centers, is becoming a huge new demand that no one is expecting. ”

Why lack of vision? Typically, demand for electricity increases by about half each year, and utilities bring new generators and make other investments as needed to meet the expected new demand. But now data centers that are about to access the Internet are creating an unprecedented leap, something operators have not seen. Furthermore, new demand is constant. It is crucial that the data center serves all day every day. Process large data sets, access stored data, and run the required cooling equipment so that all packaged computers stir the required cooling equipment when they are not hot.

Additionally, Mitei research scientist Deepjyoti Deka explains that even if enough electricity is generated, it can be transferred to where it is needed. “The grid is a network-wide operation, and the grid operator may have enough power generation in another location or even elsewhere in the country, but the wires may not have enough capacity to transport the power to the desired location.” Therefore, transmission capacity must be expanded – a slow process, Deka said.

Then there is the “interconnect queue”. Sometimes adding a new user (“load”) or a new generator to an existing grid can cause instability or other problems for everyone else on the grid. In this case, the new data center is delayed online. Enough delays may cause new loads or generators to have to stand and wait for their reincarnation. Currently, many interconnect queues are already full of new solar and wind projects. It’s been delayed for about five years. Meet the needs of newly installed data centers while ensuring that the quality of service is not hindered elsewhere is an issue that needs to be addressed.

Looking for a clean power supply

To further challenge the challenges, many companies (including so-called “super standards” such as Google, Microsoft and Amazon) have publicly promised to obtain net zero carbon emissions over the next 10 years. Many people have been achieving their clean energy goals by purchasing a “power purchase agreement.” They signed contracts to purchase electricity from solar or wind facilities, sometimes providing funding for facilities to be built. However, when faced with extreme electricity demand in data centers, the approach to obtaining clean energy has limitations.

Meanwhile, soaring power consumption is delaying the closure of coal plants in many states. There is simply not enough renewable energy sources to serve high-priced and existing users, including individual consumers. As a result, conventional plants emitted from fossil fuels, such as coal, are needed more than ever.

When Advanced Standards finds sources of clean energy for their data centers, one option is to build your own wind and solar installations. However, such facilities will only generate electricity intermittently. Given the need for uninterrupted power, the data center must maintain the energy storage unit, which is expensive. Instead, they can rely on natural gas or diesel generators to get backup power, but these devices need to be combined with the devices to capture carbon emissions, plus nearby sites to permanently dispose of the captured carbon.

Due to this complication, some high standards are turning to nuclear energy. As Green points out: “Nuclear energy is very matched with the needs of data centers because nuclear power plants can reliably generate a lot of power without interruption.”

In a highly-watched move in September, Microsoft signed a deal to reopen one of a non-damaged reactor at Constellation Energy that has reopened its nuclear power plant on Thrighted Island, the site of a highly-watched nuclear accident in 1979. If approved by regulators, the site of the stable nuclear accident will be brought to 2028 by Microsoft’s purchase of all products at Microsoft. Amazon also reached a deal to buy power arising from another nuclear power plant threatened by financial problems. In early December, Meta proposed a proposal to identify nuclear energy developers to help companies meet their AI needs and sustainability goals.

Other nuclear news focuses on small modular nuclear reactors (SMRs), modular power plants built by factories that can be installed near data centers, may not be cost-overflows, and are often experienced when building large plants. Google recently ordered a fleet of SMR to generate the power required for its data center. The first one will be completed in 2030 and by 2035.

Some high standards are betting on new technologies. For example, Google is pursuing the next generation of geothermal projects, and Microsoft signed a contract to buy electricity from a startup’s converged converged power plant starting in 2028, although converged technology has not yet been proven.

Reduce power demand

Other ways of providing enough clean power focus on making the data center and its operations it accommodates more efficiently to perform the same computing tasks using less power. Using faster computer chips and optimizing algorithms that use less energy has helped reduce loads and the heat generated.

Another idea to try involves transferring the computational task to time and place without carbon energy on the grid. “If a task does not have to be completed immediately, but can be delayed or moved to data centers elsewhere in the United States, or data centers elsewhere in the United States, and where electricity is richer, cheaper and/or cleaner,” Deka explained. “This approach is called “carbon awareness.” “We are not yet sure if each task can be moved or delayed easily.” If you think of a task based on generated AI, can it be easily divided into small tasks that can be taken to different regions of the country, solved with clean energy, and then reintegrated together?” How much is the cost of doing this kind of task division? ”

Of course, this approach is limited by the interconnect queue problem. It is difficult to obtain clean energy in another region or state. However, efforts are being made to mitigate regulatory frameworks to ensure that critical interconnections can be developed faster and easier.

What about the neighbors?

The main problem with all the options that power the data center is the impact on residential energy consumers. When data centers enter the community, not only are there aesthetic problems, but there are more practical concerns. Will local power services become less reliable? Where will the new transmission line be? Who will pay for new generators, upgrade to existing equipment, etc.? When new manufacturing facilities or industrial plants enter the community, the drawbacks are often offset by the availability of new jobs. This is not the case with data centers, which may only require a dozen employees.

Standard rules on how maintenance and upgrade costs are shared and allocated. However, the existence of the new data center completely changed the situation. As a result, utilities now need to rethink their traditional rate structures to avoid undue burdens on residents to cover changes in infrastructure required to host data centers.

MIT’s contribution

At MIT, researchers are considering and exploring options to address the problem of providing clean energy to data centers. For example, they are working on architectural designs that will use natural ventilation to promote cooling, equipment layouts that will allow for better airflow and power distribution and energy-efficient air conditioning systems based on novel materials. They are creating new analytical tools to evaluate the impact of data center deployment on U.S. power systems and find the most effective way to provide clean energy to facilities. Other work focuses on how to match the output of a small nuclear reactor to the needs of a data center and how to speed up the construction of such reactors.

The MIT team is also focused on identifying the best sources of backup power and long-term storage, as well as developing decision support systems for locating proposed new data centers, with the potential to use large amounts of waste that can be used to heat buildings, given the availability of electricity and water and regulatory considerations. Technology development projects include designing faster and more efficient computer chips and more energy-efficient computing algorithms.

In addition to providing leadership and funding for many research projects, Mitei also serves as a convener, bringing companies and stakeholders together to solve this problem. At Mitei’s 2024 Annual Research Conference, a team of representatives from two large standards and representatives from both companies designed and built data centers, discussing their challenges, possible solutions, and what MIT research is most beneficial.

Green said that as data centers continue to be built, computing continues to create an unprecedented increase in demand for electricity, scientists and engineers are competing to provide ideas, innovations and technologies that can meet that demand while continuing to advance the transition to decarbonized energy systems.