AI's Growing Energy Demands Spur Innovation in Sustainable Computing

Can we make AI less power-hungry? These researchers are working on it.

At the beginning of November 2024, the US Federal Energy Regulatory Commission (FERC) rejected Amazon's request to buy an additional 180 megawatts of power directly from the Susquehanna nuclear power plant for a data center located nearby. The rejection was due to the argument that buying power directly instead of getting it through the grid like everyone else works against the interests of other users. Demand for power in the US has been flat for nearly 20 years. "But now we're seeing load forecasts shooting up. Depending on [what] numbers you want to accept, they're either skyrocketing or they're just rapidly increasing," said Mark Christie, a FERC commissioner. Part of the surge in demand comes from data centers, and their increasing thirst for power comes in part from running increasingly sophisticated AI models. As with all world-shaping developments, what set this trend into motion was vision -- quite literally. The AlexNet moment Back in 2012, Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton, AI researchers at the University of Toronto, were busy working on a convolution neural network (CNN) for the ImageNet LSRVC, an image-recognition contest. The contest's rules were fairly simple: A team had to build an AI system that could categorize images sourced from a database comprising over a million labeled pictures. The task was extremely challenging at the time, so the team figured they needed a really big neural net -- way bigger than anything other research teams had attempted. AlexNet, named after the lead researcher, had multiple layers, with over 60 million parameters and 650 thousand neurons. The problem with a behemoth like that was how to train it. What the team had in their lab were a few Nvidia GTX 580s, each with 3GB of memory. As the researchers wrote in their paper, AlexNet was simply too big to fit on any single GPU they had. So they figured out how to split AlexNet's training phase between two GPUs working in parallel -- half of the neurons ran on one GPU, and the other half ran on the other GPU. AlexNet won the 2012 competition by a landslide, but the team accomplished something way more profound. The size of AI models was once and for all decoupled from what was possible to do on a single CPU or GPU. The genie was out of the bottle. (The AlexNet source code was recently made available through the Computer History Museum.) The balancing act After AlexNet, using multiple GPUs to train AI became a no-brainer. Increasingly powerful AIs used tens of GPUs, then hundreds, thousands, and more. But it took some time before this trend started making its presence felt on the grid. According to an Electric Power Research Institute (EPRI) report, the power consumption of data centers was relatively flat between 2010 and 2020. That doesn't mean the demand for data center services was flat, but the improvements in data centers' energy efficiency were sufficient to offset the fact we were using them more. Two key drivers of that efficiency were the increasing adoption of GPU-based computing and improvements in the energy efficiency of those GPUs. "That was really core to why Nvidia was born. We paired CPUs with accelerators to drive the efficiency onward," said Dion Harris, head of Data Center Product Marketing at Nvidia. In the 2010-2020 period, Nvidia data center chips became roughly 15 times more efficient, which was enough to keep data center power consumption steady. All that changed with the rise of enormous large language transformer models, starting with ChatGPT in 2022. "There was a very big jump when transformers became mainstream," said Mosharaf Chowdhury, a professor at the University of Michigan. (Chowdhury is also at the ML Energy Initiative, a research group focusing on making AI more energy-efficient.) Nvidia has kept up its efficiency improvements, with a ten-fold boost between 2020 and today. The company also kept improving chips that were already deployed. "A lot of where this efficiency comes from was software optimization. Only last year, we improved the overall performance of Hopper by about 5x," Harris said. Despite these efficiency gains, based on Lawrence Berkely National Laboratory estimates, the US saw data center power consumption shoot up from around 76 TWh in 2018 to 176 TWh in 2023. The AI lifecycle LLMs work with tens of billions of neurons approaching a number rivaling -- and perhaps even surpassing -- those in the human brain. The GPT 4 is estimated to work with around 100 billion neurons distributed over 100 layers and over 100 trillion parameters that define the strength of connections among the neurons. These parameters are set during training, when the AI is fed huge amounts of data and learns by adjusting these values. That's followed by the inference phase, where it gets busy processing queries coming in every day. The training phase is a gargantuan computational effort -- Open AI supposedly used over 25,000 Nvidia Ampere 100 GPUs running on all cylinders for 100 days. The estimated power consumption is 50 GW-hours, which is enough to power a medium-sized town for a year. According to numbers released by Google, training accounts for 40 percent of the total AI model power consumption over its lifecycle. The remaining 60 percent is inference, where power consumption figures are less spectacular but add up over time. Trimming AI models down The increasing power consumption has pushed the computer science community to think about how to keep memory and computing requirements down without sacrificing performance too much. "One way to go about it is reducing the amount of computation," said Jae-Won Chung, a researcher at the University of Michigan and a member of the ML Energy Initiative. One of the first things researchers tried was a technique called pruning, which aimed to reduce the number of parameters. Yann LeCun, now the chief AI scientist at Meta, proposed this approach back in 1989, terming it (somewhat menacingly) "the optimal brain damage." You take a trained model and remove some of its parameters, usually targeting the ones with a value of zero, which add nothing to the overall performance. "You take a large model and distill it into a smaller model trying to preserve the quality," Chung explained. You can also make those remaining parameters leaner with a trick called quantization. Parameters in neural nets are usually represented as a single-precision floating point number, occupying 32 bits of computer memory. "But you can change the format of parameters to a smaller one that reduces the amount of needed memory and makes the computation faster," Chung said. Shrinking an individual parameter has a minor effect, but when there are billions of them, it adds up. It's also possible to do quantization-aware training, which performs quantization at the training stage. According to Nvidia, which implemented quantization training in its AI model optimization toolkit, this should cut the memory requirements by 29 to 51 percent. Pruning and quantization belong to a category of optimization techniques that rely on tweaking the way AI models work internally -- how many parameters they use and how memory-intensive their storage is. These techniques are like tuning an engine in a car to make it go faster and use less fuel. But there's another category of techniques that focus on the processes computers use to run those AI models instead of the models themselves -- akin to speeding a car up by timing the traffic lights better. Finishing first Apart from optimizing the AI models themselves, we could also optimize the way data centers run them. Splitting the training phase workload evenly among 25 thousand GPUs introduces inefficiencies. "When you split the model into 100,000 GPUs, you end up slicing and dicing it in multiple dimensions, and it is very difficult to make every piece exactly the same size," Chung said. GPUs that have been given significantly larger workloads have increased power consumption that is not necessarily balanced out by those with smaller loads. Chung figured that if GPUs with smaller workloads ran slower, consuming much less power, they would finish roughly at the same time as GPUs processing larger workloads operating at full speed. The trick was to pace each GPU in such a way that the whole cluster would finish at the same time. To make that happen, Chung built a software tool called Perseus that identified the scope of the workloads assigned to each GPU in a cluster. Perseus takes the estimated time needed to complete the largest workload on a GPU running at full. It then estimates how much computation must be done on each of the remaining GPUs and determines what speed to run them so they finish at the same. "Perseus precisely slows some of the GPUs down, and slowing down means less energy. But the end-to-end speed is the same," Chung said. The team tested Perseus by training the publicly available GPT-3, as well as other large language models and a computer vision AI. The results were promising. "Perseus could cut up to 30 percent of energy for the whole thing," Chung said. He said the team is talking about deploying Perseus at Meta, "but it takes a long time to deploy something at a large company." Are all those optimizations to the models and the way data centers run them enough to keep us in the green? It takes roughly a year or two to plan and build a data center, but it can take longer than that to build a power plant. So are we winning this race or losing? It's a bit hard to say. Back of the envelope As the increasing power consumption of data centers became apparent, research groups tried to quantify the problem. A Lawerence Berkley Laboratory team estimated that data centers' annual energy draw in 2028 would be between 325 and 580 TWh in the US -- that's between 6.7 and 12 percent of the total US electricity consumption. The International Energy Agency thinks it will be around 6 percent by 2026. Goldman Sachs Research says 8 percent by 2030, while EPRI claims between 4.6 and 9.1 percent by 2030. EPRI also warns that the impact will be even worse because data centers tend to be concentrated at locations investors think are advantageous, like Virginia, which already sends 25 percent of its electricity to data centers. In Ireland, data centers are expected to consume one-third of the electricity produced in the entire country in the near future. And that's just the beginning. Running huge AI models like ChatGPT is one of the most power-intensive things that data centers do, but it accounts for roughly 12 percent of their operations, according to Nvidia. That is expected to change if companies like Google start to weave conversational LLMs into their most popular services. The EPRI report estimates that a single Google search today uses around 0.3 watts of power, while a single Chat GPT query bumps that up to 2.9 watts. Based on those values, the report estimates that an AI-powered Google search would require Google to deploy 400,000 new servers that would consume 22.8 TWh per year. "AI searches take 10x the electricity of a non-AI search," Christie, the FERC commissioner, said at a FERC-organized conference. When FERC commissioners are using those numbers, you'd think there would be rock-solid science backing them up. But when Ars asked Chowdhury and Chung about their thoughts on these estimates, they exchanged looks... and smiled. Closed AI problem Chowdhury and Chung don't think those numbers are particularly credible. They feel we know nothing about what's going on inside commercial AI systems like ChatGPT or Gemini, because OpenAI and Google have never released actual power-consumption figures. "They didn't publish any real numbers, any academic papers. The only number, 0.3 watts per Google search, appeared in some blog post or other PR-related thingy," Chodwhury said. We don't know how this power consumption was measured, on what hardware, or under what conditions, he said. But at least it came directly from Google. "When you take that 10x Google vs ChatGPT equation or whatever -- one part is half-known, the other part is unknown, and then the division is done by some third party that has no relationship with Google nor with Open AI," Chowdhury said. Google's "PR-related thingy" was published back in 2009, while the 2.9-watts-per-ChatGPT-query figure was probably based on a comment about the number of GPUs needed to train GPT-4 made by Jensen Huang, Nvidia's CEO, in 2024. That means the "10x AI versus non-AI search" claim was actually based on power consumption achieved on entirely different generations of hardware separated by 15 years. "But the number seemed plausible, so people keep repeating it," Chowdhury said. All reports we have today were done by third parties that are not affiliated with the companies building big AIs, and yet they arrive at weirdly specific numbers. "They take numbers that are just estimates, then multiply those by a whole lot of other numbers and get back with statements like 'AI consumes more energy than Britain, or more than Africa, or something like that.' The truth is they don't know that," Chowdhury said. He argues that better numbers would require benchmarking AI models using a formal testing procedure that could be verified through the peer-review process. As it turns out, the ML Energy Initiative defined just such a testing procedure and ran the benchmarks on any AI models they could get ahold of. The group then posted the results online on their ML.ENERGY Leaderboard. AI-efficiency leaderboard To get good numbers, the first thing the ML Energy Initiative got rid of was the idea of estimating how power-hungry GPU chips are by using their thermal design power (TDP), which is basically their maximum power consumption. Using TDP was a bit like rating a car's efficiency based on how much fuel it burned running at full speed. That's not how people usually drive, and that's not how GPUs work when running AI models. So Chung built ZeusMonitor, an all-in-one solution that measured GPU power consumption on the fly. For the tests, his team used setups with Nvidia's A100 and H100 GPUs, the ones most commonly used at data centers today, and measured how much energy they used running various large language models (LLMs), diffusion models that generate pictures or videos based on text input, and many other types of AI systems. The largest LLM included in the leaderboard was Meta's Llama 3.1 405B, an open-source chat-based AI with 405 billion parameters. It consumed 3352.92 joules of energy per request running on two H100 GPUs. That's around 0.93 watt-hours -- significantly less than 2.9 watt-hours quoted for ChatGPT queries. These measurements confirmed the improvements in the energy efficiency of hardware. Mixtral 8x22B was the largest LLM the team managed to run on both Ampere and Hopper platforms. Running the model on two Ampere GPUs resulted in 0.32 watt-hours per request, compared to just 0.15 watt-hours on one Hopper GPU. What remains unknown, however, is the performance of proprietary models like GPT-4, Gemini, or Grok. The ML Energy Initiative team says it's very hard for the research community to start coming up with solutions to the energy efficiency problems when we don't even know what exactly we're facing. We can make estimates, but Chung insists they need to be accompanied by error-bound analysis. We don't have anything like that today. The most pressing issue, according to Chung and Chowdhury, is the lack of transparency. "Companies like Google or Open AI have no incentive to talk about power consumption. If anything, releasing actual numbers would harm them," Chowdhury said. "But people should understand what is actually happening, so maybe we should somehow coax them into releasing some of those numbers." Where rubber meets the road "Energy efficiency in data centers follows the trend similar to Moore's law -- only working at a very large scale, instead of on a single chip," Nvidia's Harris said. The power consumption per rack, a unit used in data centers housing between 10 and 14 Nvidia GPUs, is going up, he said, but the performance-per-watt is getting better. "When you consider all the innovations going on in software optimization, cooling systems, MEP (mechanical, electrical, and plumbing), and GPUs themselves, we have a lot of headroom," Harris said. He expects this large-scale variant of Moore's law to keep going for quite some time, even without any radical changes in technology. There are also more revolutionary technologies looming on the horizon. The idea that drove companies like Nvidia to their current market status was the concept that you could offload certain tasks from the CPU to dedicated, purpose-built hardware. But now, even GPUs will probably use their own accelerators in the future. Neural nets and other parallel computation tasks could be implemented on photonic chips that use light instead of electrons to process information. Photonic computing devices are orders of magnitude more energy-efficient than the GPUs we have today and can run neural networks literally at the speed of light. Another innovation to look forward to is 2D semiconductors, which enable building incredibly small transistors and stacking them vertically, vastly improving the computation density possible within a given chip area. "We are looking at a lot of these technologies, trying to assess where we can take them," Harris said. "But where rubber really meets the road is how you deploy them at scale. It's probably a bit early to say where the future bang for buck will be." The problem is when we are making a resource more efficient, we simply end up using it more. "It is a Jevons paradox, known since the beginnings of the industrial age. But will AI energy consumption increase so much that it causes an apocalypse? Chung doesn't think so. According to Chowdhury, if we run out of energy to power up our progress, we will simply slow down. "But people have always been very good at finding the way," Chowdhury added.

AI Needs to Be More Energy-Efficient

Artificial Intelligence uses too much energy. Developers need to find better ways to power it or risk adding to the climate crisis Artificial intelligence is everywhere: it's designing new proteins, answering Internet search questions, even running barbecues. Investors are captivated by it -- and so is the U.S. president. Just after taking office, President Donald Trump announced his support for Stargate, a company worth up to $500 billion, bankrolled by some of the biggest players in this space, to facilitate AI development in the U.S. But the data centers and other infrastructure needed to develop and run the technology are incredible electricity hogs. And with Trump's declaration of a "national energy emergency" -- an undisguised ploy to increase fossil-fuel production -- AI's energy needs are poised to make climate change even worse. The technology is already responsible for massive greenhouse gas emissions that cause climate change. If Stargate and the many other companies developing AI platforms do not insist on cleaner and more efficient energy, they will only aid in the destruction of our planet. This technology's many flavors include the buzzy generative AI, the basis of ChatGPT and Google's year-old search-answer system. During its operation, generative AI guzzles electricity in two stages, requiring warehouse-size data centers to house the necessary computing. On supporting science journalism If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today. We all will have to pay when this exorbitant energy use inflates the cost of the kilowatt, regardless of our personal interaction with the technology. Developers must first train the AI model on vast stores of data, which takes countless hours and requires enormous computing capabilities. Training one ChatGPT precursor consumed enough electricity to power 120 average U.S. homes for a year. Every time a model is upgraded, it must be retrained. The sudden release of the DeepSeek chatbot out of China -- reportedly trained for a fraction of the price of ChatGPT and similar U.S. systems -- may lead to less energy-intensive processes, but it's too soon to know for sure. And the demand doesn't stop once a model is trained. Each query the AI receives requires it to consider everything it has been fed, then synthesize an answer from scratch in a process called inference, which also requires energy. Compared with search engines, text-generating systems can easily use 10 times as much energy to address a query, and sometimes they use dozens of times more. Image generation requires even more energy -- as much as 50 percent of the amount needed to fully charge a smartphone, one study found. Many analyses interpret this energy use for the training and large-scale operation of AI as an increased cost to the system's owner. For example, one estimate suggests that if Google uses generative AI to produce 50 words of text per answer in response to just half of the queries it receives, it will cost the company some $6 billion. But the truth is, we all will have to pay when this exorbitant energy use inflates the cost of the kilowatt, regardless of our personal interaction with the technology. The scale of consumption is simply too large, and as AI sneaks into ever more aspects of daily life, its energy use is projected to skyrocket. At the industry scale, it's difficult to isolate AI from other computing demands, but data centers serve as a convenient proxy, given that the rise of the technology has led to their boom. The numbers are staggering: In the mid-2010s U.S. data centers used about 60 terawatt-hours per year. (One terawatt-hour is the equivalent of one billion kilowatt-hours , the unit used to measure electricity consumption in most U.S. homes.) By 2023, a recent report from Lawrence Berkeley National Laboratory found, that number had nearly tripled to 176 terawatt-hours; demand is expected to rise to between 325 and 580 terawatt-hours by 2028. At that level, data-center energy use would potentially account for between 6 and 12 percent of total U.S. energy consumption, up from 4 percent in 2023. Even as commercial energy demand continues to grow, people are already seeing higher residential energy prices in some regions where thirsty technologies such as AI are taxing the grid. Amid this skyrocketing energy demand, work to decarbonize energy production is progressing too slowly both in the U.S. and globally. Climate change is already unfolding around us, worsening disasters ranging from the Los Angeles fires to Hurricane Helene to extreme heat and causing surprising and long-lasting consequences. Reducing the harm of climate change requires ending fossil-fuel use as quickly as possible. Sudden, huge demand from any industry makes that more difficult. Sure, large technology companies could offer valuable resources to support the energy transition. The Stargate investment is expected to rely in part on solar power. Before leaving office, President Joe Biden opened public lands to data centers running on clean energy as a way to encourage its use for computing. But because solar, wind and hydropower production rates can vary with weather and other factors, nuclear energy is particularly appealing to ever thirsty AI technology companies, raising fears of nuclear waste contamination. Most notably, Microsoft has a deal to restart the infamous Three Mile Island fission facility that was the site of the worst nuclear accident in the U.S. Meanwhile OpenAI CEO Sam Altman is throwing his support behind, among other things, nuclear fusion, a technology that looks unlikely to provide energy at any significant scale until 2050 at the earliest. Even if AI companies lean heavily on clean power and don't worsen the climate crisis, the technology's seemingly insatiable need for energy remains concerning. And efficiency improvements, though vital, may not be enough. The so-called Jevons paradox, which posits that making a resource cheaper or more efficient can increase its use rather than shrinking its footprint, may be a factor. Wider highways invite more cars, and the Internet has led to doomscrolling as a time-consuming preoccupation that encourages more energy use. While technology companies push AI, we need to push them for not just small innovations in efficiency but big ones that keep the energy footprint of the U.S. reined in. The alternative may be an AI-enabled barbecue that chars the world.

Research Using AI in Energy Applications at CMU Showcases the Frontier of Opportunities

Pioneered at Carnegie Mellon University(opens in new window), artificial intelligence (AI) holds tremendous promise while also invoking challenges in its applications and use. AI's capabilities to synthesize mammoth amounts of data are being harnessed across every industry. However, running and iterating on the algorithms that compute new, innovative solutions faster also means energy to power them is needed in greater magnitudes. The pace at which AI is now advancing also requires consideration of how to efficiently and sustainably power the technology, and long-term, thoughtful strategies, said Daniel Tkacik(opens in new window), executive director of the Wilton E. Scott Institute for Energy Innovation(opens in new window). "At Carnegie Mellon, we are leading AI for the world, but we're also leading energy for the world," he said. "We've been leading work at the intersection of AI and energy for decades." As part of that mission, the Scott Institute is hosting CMU Energy Week(opens in new window), this year bringing together energy and sustainability leaders to combine forces and exchange ideas at the intersection of AI and energy. "There are good reasons why AI is being pursued in the way it is," Tkacik said. "It's going to make our lives better in a myriad of ways, and there are a lot of smart people trying to make sure that this is done in an ethical and responsible way."." Emma Strubell(opens in new window), Raj Reddy Assistant Professor in the Language Technologies Institute(opens in new window) in the School of Computer Science(opens in new window), who was recently named one of the most powerful people in artificial intelligence(opens in new window), said solutions regarding AI efficiency should start with agreeing upon what defines something as artificial intelligence, then measuring and reporting the energy usage. Strubell contributes to a nationwide project(opens in new window) funded by the U.S. National Science Foundation's (NSF) Expeditions in Computing Awards program hoping to lay this type of groundwork for sustainable computing. "In order to make informed decisions and policies -- for example, around energy use and the relationship between AI and future energy use in the U.S., due to data centers -- we need a much better understanding of the actual drivers of that energy use," they said. Strubell is among the Carnegie Mellon researchers examining these frontiers related to AI in energy and climate solutions set to contribute to discussions as part of Energy Week(opens in new window). "I've been thinking about the foundational work," they said. "There's a need for data. And there's a lot of analysis that we can do in academia with the information that's available to us, but there's also a lot that we can't do because there's not enough data about things like what the workloads actually are in data centers." Efficiency in computing also stems from more efficient processing systems, including computer chips, which has been the focus of work by Brandon Lucia(opens in new window), Kavčić-Moura Professor of Electrical and Computer Engineering in the College of Engineering(opens in new window), for the past 10 years. "We need to make computing more energy efficient, because if we don't do that, we can't continue to add functionality to these kinds of energy-constrained devices," he said. "We can't continue to push AI and computing forward in general without energy efficiency. Going forward, for the next five to 10 years, energy is the only thing that matters." Lucia and his research team are working on batteryless computer systems(opens in new window) that use energy-harvesting devices and intermittent computing, a term that his team established. The devices collect energy from the environment, such as solar energy, radio waves or vibrational mechanical energy, which is stored in a chargeable capacitor that is simpler than one that uses a battery. "There's no environmental impact of having to produce, distribute and dispose of batteries, and there's no maintenance in having to replace batteries," Lucia said of the project. "You have these devices which have all these other benefits, but they have an unfortunate side effect -- they turn off, because you don't always have power in the environment. And when they turn off, your system goes haywire." That's where intermittent computing comes in, Lucia said. Lucia began to push forward intermittent computing, and he and Nathan Beckmann(opens in new window), associate professor in the School of Computer Science, co-advised then-doctoral student Graham Gobieski on work on spatial dataflow architectures that were well-suited for intermittent computing. The trio founded Efficient Computer(opens in new window) to commercialize the spatial dataflow architecture. "With that little tiny bit of energy, the biggest impediment to the progress of those batteryless devices was the inefficiency of computing to begin with," Lucia said. "This spatial dataflow architecture is what unlocks all that efficiency, and so we've been developing those efficient architectures." Devices using these technologies could be used in complex environments that are difficult to access, where those with batteries would otherwise require changing or charging by humans, such as space exploration, disaster response, and construction or industrial sites. Fewer batteries would also lessen the environmental impact of otherwise disposing of or recycling used batteries. "With these batteryless devices, you can extend the lifetime of the device essentially indefinitely, until the energy transducer, like a solar panel, starts to break down," Lucia said. "So that might be a five- to 10-year lifetime, up to 30 years, in some cases." Ultimately, energy is a cornerstone on which future innovation rests, Lucia said, especially when it comes to applications such as personal devices for each of the seven and a half billion people on the planet. "We need to think about energy," he said. "The defining problem for humanity for the next several decades is how to match the need for computing with the energy required, and that includes AI, but not just AI. ... Energy is the biggest problem facing humanity." Using AI could help unlock a new potential source of energy to solve that problem, including work by Jeff Schneider(opens in new window), research professor in the School of Computer Science, and his research team studying nuclear fusion. The reaction, where atoms collide -- distinct from the splitting atoms of nuclear fission already used in nuclear power plants -- are created in a tokamak machine. The billion-dollar reactor heats hydrogen until it becomes plasma, which is then formed into a donut-shape, wrapped with magnets and confined within a magnetic field. The system simultaneously controls the injection of hydrogen particles, the shape of the plasma, and its current and density. "We just don't know how to keep that plasma in place at high enough temperatures and pressures for long enough periods of time so that it can be used in a power plant," said Schneider, who will also be speaking as part of Energy Week. "That's basically the one thing that's standing between us and unlimited clean energy." Machine learning is helping the team synthesize decades of data from past experiments to better understand exactly how each of the "shots" taken inside the machine at the DIII-D National Fusion Facility in San Diego. "Over the decades there have been tens of thousands of these shots that have happened, and all of that you can feed to a machine learning model to learn to predict those steps from one state to the next to use that to make a simulator," Schneider said. The simulator can then perform millions of shots, "more than have ever been run in real time," he said. "It just keeps practicing until it finds a control policy, and now what we can do is get to the real bottleneck to progress, the limited time -- only a few hours per year -- available to run experiments on the tokamak." Schneider said with machine learning, the team has been able to double the number of shots that could happen without disruption as a result of their experiments, and because of that, collaborators asked to run the same algorithm at the KSTAR tokamak in South Korea. "These methods are getting us these results faster than we otherwise would be able to get them," he said. "These are both examples of things that physicists have known about and been interested in for years, that you know they just hadn't been able to reach yet. Now, we've proven that we have the tools to solve them, and so what we're really trying to do now is to get the resources to roll this out at scale." Making progress on the science behind nuclear fusion will lead to progress toward power plants that can produce considerable energy from a renewable, emission-free source. "If you think about the world's grand challenges, many of them are just energy problems," Schneider said, citing global warming and food and water accessibility. "All these things are just clean energy problems, so that's why I'm really excited about solving the problems with clean energy, specifically with fusion." When it comes to research aiding systems that are already in place, Azadeh O. Sawyer(opens in new window), assistant professor in building technology with the School of Architecture(opens in new window), is using AI to make building design more efficient. Setting a benchmark standard for buildings' energy usage of a typical size in a specific location allows for future designs to use those figures to continue to improve efficiency, but many cities don't have those comprehensive energy benchmarking data to compare past designs with those under development, Sawyer said. "We realized we could actually use the power of AI and machine learning to analyze a lot of data from other similar cities and similar environments to predict what it would be for a city that's missing those benchmarks right now," she said. Sawyer received a seed grant from the Scott Institute(opens in new window) with her Ph.D. student Tian Li for this work to address the need to reduce carbon emissions and energy use in buildings across the country, then identify ambitious and achievable decarbonization targets. Previously, building scientists would apply statistical methods, she said. Now, using AI, researchers can augment those approaches by uncovering nonlinear patterns to make even better predictions and classifications based on larger and more complex datasets. "AI really changes how we do research," she said. "It gives you the opportunity to really do true exploration: Here's an idea area, I don't necessarily know everything that's out there, but I have this amount of data and can ask it to find patterns that our human eyes don't see." In another project, she is working with doctoral student Niloofar Nikookar to develop a dynamic lighting system based on AI. Static lighting systems remain the same regardless of the amount of daylight filtered through a building's windows or the color of the light, which could affect the moods and productivity of the people inside. Adjusting lighting through smart systems can not only help people feel better, but also improve energy efficiency, Sawyer said. "Our hope is that once we have the datasets of how people respond to different colors of lighting and different sky conditions, then we could actually train the model to create a dynamic lighting system using AI that responds to how someone feels and what kind of space they're in," she said. Building design could also benefit from predicting occupant behavior, which can also impact energy usage, Sawyer said, such as how people react to blinds that adjust on their own to create shade and reduce glare. "We're designing for people," she said. "We want to design responsibly so it doesn't harm the environment and so it doesn't negatively impact the people that are using our spaces." Mario Bergés(opens in new window), professor in the Department of Civil and Environmental Engineering(opens in new window), examines the way existing buildings monitor and use energy in order to make them more efficient. His research involves what is known as non-intrusive load monitoring, which analyzes smart meter data, identifying appliance usage and predicting malfunctions. "If you can be smart about how to analyze the data that's coming from your smart meter, then you are going to be able to fingerprint individual appliances and also get to know a lot about the behavior of people in the home through their usage of devices that consume electricity," Bergés said. Then, computer systems can be trained to analyze the data from the meter and provide feedback on how to improve energy consumption. "Say your washing machine is breaking down. There are signatures of that motor failing that could be detected from the smart meter itself," he said. "If you're careful about what you're paying attention to, you can know not only that these things are on or off, but also what they could be doing, and whether they are malfunctioning." Even though all the possible signatures from every appliance would be difficult to identify, systems could use AI to recognize anomalies closely enough to monitor them and make the data accessible. His team has used artificial intelligence to develop learning-based controls for heating, ventilation and air condition systems. AI can monitor the building, then emulate how to manage the temperatures before making it more efficient. Instead of managing only one building at a time, a set of buildings could be managed together and the energy usage could be coordinated using an algorithm. The energy can then be stored or released as needed. "Then you are essentially allowing all these buildings to act as a very big thermal battery," he said. "All of them together are creating this storage and they are allowing for the excess production of electricity to be stored." Using artificial intelligence in each of these ways requires human innovation and ingenuity, including the interdisciplinary collaboration encouraged by Carnegie Mellon and facilitated by the Scott Institute. The faculty members agreed that the students and colleagues at CMU are what makes considering and working toward solutions to these energy-related challenges possible. "We have the best students to work on these projects, and it really takes the best folks in AI and machine learning to tackle such hard problems," said Schneider. The future benefits of artificial intelligence will depend on how the research and innovation being pursued now also balance sustainability and the environment. "You can have both innovation and sustainability. We just need to sort of think about things differently," Strubell said. "Now is a critical time to be thinking about this, because I think we are about to build out a ton of data center infrastructure and the supporting energy systems to power those data centers, and I do believe that we can do that in a way that is compatible with sustainability in various ways." Processes using AI that improve energy usage, storage and reliance developed now through research at CMU will continue to transform and establish sustainable systems well into the future. "I'm optimistic, and the Scott Institute is optimistic, which is why we're pursuing this line of research," Tkacik said. "There are many applications of AI for energy and climate yet to be discovered, and these discoveries are being pursued in a very responsible way by a number of scholars, both across the country and right here at Carnegie Mellon."

Empowering the Energy Workforce for an AI-Driven Future

As artificial intelligence continues to make its way into the energy sector, it brings with it a wave of innovation and efficiency. AI is optimizing grid operations by predicting and preventing blackouts, enhancing energy efficiency by analyzing data to reduce waste, and seamlessly integrating renewable energy sources like solar and wind. It is also improving system safety and reliability through predictive maintenance and helping to protect the environment by reducing reliance on fossil fuels. As the industry embraces this digital transformation, AI is also reshaping the workforce. Understandably, a technological advancement of this magnitude can stir apprehension among workers who fear AI might replace their jobs or disrupt their routines. However, experts at Carnegie Mellon University emphasize that the role of AI in the energy workforce is to unlock the full potential of human workers. Collaborating across disciplines in the very place where AI was born(opens in new window), CMU's experts are continuing to develop the tools and resources to help society seamlessly adapt, ensuring that everyone benefits in an AI-driven future. "CMU's deep expertise on AI research across science, policy, engineering, computer science, humanities, arts and business allows for a systems-wide view of how to maximize the benefits and minimize the challenges of AI for society," said Costa Samaras(opens in new window), director of the Wilton E. Scott Institute for Energy Innovation(opens in new window). "This includes understanding the needs and skills for a future workforce and enabling opportunities for everyone to participate." AI systems, especially those that rely on remote data centers for processing, consume significant energy due to their high computational demands. But AI is not only a significant consumer of energy; it is also a potential source of innovative solutions to the challenges that the energy sector faces. Burcu Akinci(opens in new window), department head and Hamerschlag University Professor of Civil and Environmental Engineering(opens in new window) at CMU, explained that buildings are the biggest energy users, and within buildings, HVAC (heating, ventilation and air conditioning) systems consume the most energy. These systems are complex and customized for each building, making it difficult to monitor them reliably. Faults in HVAC systems are significant since HVAC systems contribute approximately 40% of total energy use in buildings and buildings are responsible for 40% of total energy use in the U.S. For example, a malfunctioning component or an error in the control logic might cause a system to heat and cool the air simultaneously, wasting energy. To address this, Akinci and her team developed an AI agent that continuously monitors the vast amounts of sensor data collected from buildings. By combining AI with digital twins (digital replicas of the physical assets), workers can diagnose faults and perform root cause analyses to determine where the fault is and how much money is being wasted. "As a result, engineers and managers are empowered because they can see what the issues are, fix them and reduce waste," she said. While AI tools are powerful, they can't provide all the answers because they only consider the data without looking at the context. For example, two air handling units serving the same building can behave differently depending on the areas they supply, such as an operating room versus office space. Akinci said human expertise will always be needed. "I do not think it's doom and gloom, that people are going to lose jobs. In fact, we need more skilled engineers who are experts in their domains equipped with the right tools. You really need that domain expertise working closely with AI tools to make the most effective decisions," she said. "AI systems will be like apprentices for engineers, freeing them up to use their knowledge and creativity in a more powerful way." The energy sector is transitioning both regionally and nationally, using existing carbon-based sources like natural gas while also exploring opportunities in geothermal, nuclear, solar, wind and other clean technologies. Companies like Microsoft and Google are investing in these new energy sources to power AI data centers, such as Microsoft's stake in Three Mile Island for nuclear energy(opens in new window) and Google's geothermal project in Nevada(opens in new window). "In other words, both traditional and new sources of energy are being explored to power not only existing sections of the economy but also what AI is demanding," said Ramayya Krishnan(opens in new window), dean of CMU's Heinz College of Information Systems and Public Policy(opens in new window) and faculty director of the university's Block Center for Technology and Society(opens in new window). "Because of that, there is a need for a workforce that has the appropriate skills to contribute to this build out in many ways." CMU is using AI to help solve workforce issues like this. The university's interdisciplinary nature brought together the Scott Institute and The Block Center to create the Workforce Supply Chains Initiative(opens in new window), which is helping workers, employers and policymakers navigate the evolving job market. By analyzing data, the initiative identifies which jobs are becoming more popular and which ones are declining. This information is crucial for workers to understand what skills they might need to learn for future job opportunities. In addition to identifying job trends, the initiative provides personalized advice to workers about new roles that match their skills and interests. It offers guidance on what new skills they might need, how to acquire them and the potential benefits of upskilling, ensuring workers are well-prepared for the changing demands of the job market. For employers, the initiative offers valuable insights into how to best utilize their employees' skills. It provides guidance on creating new job roles, training workers for these roles and employee retention, helping companies stay competitive and ensuring their workforce is equipped to handle future challenges. Krishnan said CMU's ability to look at issues from a systems standpoint is what makes the university uniquely positioned to address the complex challenges and opportunities that lie at the intersection of energy and AI. "That's our sweet spot," he said. "At CMU, we have strength in all the different layers. We have deep expertise in energy, deep expertise in AI and deep expertise in public policy, and we understand how these elements come together." Krishnan will moderate a panel discussion on AI and the energy workforce at CMU Energy Week 2025(opens in new window), where guests are expected to discuss data center expansion and the need for skilled workers to design, build and maintain electrical, mechanical and cooling systems. In an effort to equip society with the knowledge and skills needed to navigate and contribute to an AI-driven world, Krishnan has recommended through the National Artificial Intelligence Advisory Committee (NAIAC), on which he serves, the creation of a comprehensive AI curriculum for libraries, community colleges and high schools to enhance AI literacy across various educational levels and community settings. Key aspects of the recommendation include launching a national AI literacy campaign through a public/private partnership; integrating AI literacy into educational programs for K-12, community colleges and non-traditional learners; and ensuring libraries and community centers have access to AI educational materials to serve as local learning hubs. "People get worried or afraid of what they don't fully understand," he said. "If we can create a curriculum on AI and distribute it, that might be a great way for people to just get that base literacy. My sense is that people are going to experience AI through systems they already know, like Word or Excel or PowerPoint." Akinci recommends that people prepare to embrace the idea of continuous learning. "The world is changing so fast. Even students entering college today can't expect to learn all they will need to know for their career in four years," she said. "It's really important for all of us entering this brave new world to have an open heart, an open mind and an attitude of continuous learning and growth -- to embrace change rather than fear or resist it." Carnegie Mellon University pioneered artificial intelligence education, offering the first bachelor's degree in AI. Ranked #1 in artificial intelligence by U.S. News & World Report, CMU provides a comprehensive range of AI programs. These include undergraduate, master's and doctoral degrees in AI and machine learning, as well as specialized programs such as AI engineering, AI-centric business degrees and graduate certificates. CMU's interdisciplinary approach ensures that students from diverse backgrounds can engage with AI in various fields, from computer science to business and beyond.

Artificial Intelligence Makes Energy Demand More Complex -- And More Achievable

Artificial intelligence, a field known for its expanding uses across society, is also increasingly notorious for the massive amount of energy it needs to function. In a 2024 paper(opens in new window), researchers from Carnegie Mellon University and machine learning development corporation Hugging Face found that generative AI systems could use as much as 33 times more energy to complete a task than task-specific software would. "The climate and sustainability challenge can be overwhelming in the amount of new clean technology that we have to deploy and develop, and the ways that the energy system has to evolve," said Costa Samaras(opens in new window), head of the university-wide Wilton E. Scott Institute for Energy Innovation(opens in new window). "The scale of the challenge alone can be overwhelming to folks." However, Carnegie Mellon University's standing commitment to the United Nations' Sustainable Development Goals and its position as a nationally recognized leader in technologies like artificial intelligence mean that it is uniquely positioned to address growing concerns around energy demand, climate resilience and social good. The research in sustainability and climate solutions that happens across the university's seven schools and colleges -- as exemplified by the countless and ongoing contributions of researchers like Harry Kresja, Destenie Nock, Azadeh Sawyer(opens in new window), Emma Strubell(opens in new window), and many others -- reflects the expertise and innovative spirit that makes Carnegie Mellon uniquely suited to address modern energy and sustainability challenges. On March 25, Carnegie Mellon University kicked off its Energy Week(opens in new window) event, a gathering which brings together national leaders from industry, government and research to Carnegie Mellon's campus to learn about and share advancements being made around the country. The Scott Institute proudly champions efforts at the intersection of energy, climate, technology and public policy, and the conference serves as its flagship annual event. But Energy Week is also a chance for the university to spotlight the growing need -- and ability -- of institutions and individuals to solve problems at the intersection of AI and energy. Samaras, director of the Scott Institute, has brought a wealth of experience in engineering and climate action to CMU. Having left the university to serve in the Biden-Harris Administration(opens in new window) from 2021-24, he has seen the landscape of energy innovation from one of the highest offices in the land. Before returning to Carnegie Mellon, Samaras helped to craft an executive order on artificial intelligence, penning a section that defined how the technology intersects with America's energy system and the climate. Now leading the Scott Institute, Samaras said Energy Week serves as a chance for individuals and institutions who interface with energy systems to take a similar approach to the problem. Among them are several from Carnegie Mellon's own campus, such as The Block Center for Technology and Society(opens in new window). "We hold Energy Week every year, and it was clear to us even last summer that the intersection of AI and energy was going to have broad interest to the community as this year's topic, but also serve as a way to showcase all the different work that's been happening at this intersection at Carnegie Mellon." "What I've seen at Carnegie Mellon is that people here understand the scale, and they welcome the challenge," Samaras said. One of the ways the university has sought to take on the task is through the expertise of The Block Center, which seeks to address the question of how technologies like AI can be harnessed for social good and quality of life improvements. In 2024, the center published a transition memorandum(opens in new window) in order to help guide incoming policymakers, industry leaders and society as a whole on issues of technological change and its demands. Among the energy-based challenges the memorandum seeks to address are the difficulty in coordination of industry resources, a lack of transparency in energy and emissions reporting, and the inadequacy of existing metrics for knowing AI's environmental impact. The relationship between sustainability and artificial intelligence isn't one that simply stops at making AI more energy efficient, Samaras said. "I think a lot of people look at this challenge of, 'How do we get to net-zero climate pollution? How do we ensure that we increase energy resilience? How do we ensure that we increase energy affordability?' And they look at these challenges as individual silos that need to be tackled by individual disciplines." Samaras said. "Carnegie Mellon doesn't think like that. Carnegie Mellon says, 'Let's get together and solve this problem and we'll bring together all the different skills and perspectives that we have in the research in the university and beyond.'" It also requires looking at how AI tools can be used to bolster existing infrastructure and amplify ongoing efforts to solve problems holistically. The work of individuals like Rayid Ghani(opens in new window), Distinguished Career Professor in the Machine Learning Department(opens in new window) and the Heinz College of Information Systems and Public Policy(opens in new window), is one example of how this cross-disciplinary approach can look. Ghani often looks at the applications of machine learning and artificial intelligence not exclusively in a climate context, but concerning a wide range of social and economic potential. His research primarily focuses on how to use the technology to promote social good in areas such as public health, economic development and urban infrastructure. But Ghani said the impact of AI on energy efficiency and sustainability has implications for many of the other areas of research and application that happen at CMU -- and taking an approach that centers socioeconomically disadvantaged and impacted people is one of the most important ways the university works to make AI sustainable. "Working on these problems throughout my career, I realized that they are all connected. A lot of the work I initially did required being opportunistic and seeing where I could make a difference," he said. "For instance, why do we care about something like clean air? It's not for the sake of the air -- it's about the people who are breathing the air. It's not about just any single thing, and so you can't focus on energy, or the environment and not worry about all the other things that impact people's health." Many experts at Carnegie Mellon approach their chosen subject matter with a similar mindset, and leveraging interdisciplinary expertise is one of the main ways that the Scott Institute has been able carry out its mission.

Can energy-hungry AI help cut our energy use?

by Anne-Muriel Brouet, Ecole Polytechnique Federale de Lausanne It takes 10 times more electricity for ChatGPT to respond to a prompt than for Google to carry out a standard search. Still, researchers are struggling to get a grasp on the energy implications of generative artificial intelligence both now and going forward. Few people realize that the carbon footprint of digital technology is on par with that of the aerospace industry, accounting for between 2% and 4% of global carbon emissions. And this digital carbon footprint is expanding at a rapid pace. When it comes to power use, the approximately 11,000 data centers in operation today consume just as much energy as the entire country of France did in 2022, or around 460 TWh. Will the widespread adoption of generative AI send those figures soaring? The new technology will clearly affect the amount of energy that's consumed worldwide, but exactly how is hard to quantify. "We need to know the total cost of generative AI systems to be able to use them as efficiently as possible," says Manuel Cubero-Castan, the project manager on Sustainable IT at EPFL. He believes we should consider the entire life cycle of generative AI technology, from the extraction of minerals and the assembly of components -- activities whose impact concerns not only energy -- to the disposal of the tons of electronic waste that are generated, which often gets dumped illegally. From this perspective, the environmental ramifications of generative AI go well beyond the power and water consumption of data centers alone. The cost of training For now, most of the data available on digital technology power use relates only to data centers. According to the International Energy Agency (IEA), these centers (excluding data networks and cryptocurrency mining) consumed between 240 TWh and 340 TWh of power in 2022, or 1% to 1.3% of the global total. Yet even though the number of centers is growing by 4% per year, their overall power use didn't change much between 2010 and 2020, thanks to energy-efficiency improvements. With generative AI set to be adopted on a massive scale, that will certainly change. Generative AI technology is based on large language models (LLMs) that use power in two ways. First, while they're being trained -- a step that involves running terabytes of data through algorithms so that they learn to predict words and sentences in a given context. Until recently, this was the most energy-intensive step. Second, while they're processing data in response to a prompt. Now that LLMs are being implemented on a large scale, this is the step requiring the most energy. Recent data from Meta and Google suggest that this step now accounts for 60% to 70% of the power used by generative AI systems, against 30% to 40% for training. ChatGPT query vs. conventional Google search A ChatGPT query consumes around 3 Wh of power, while a conventional Google search uses 0.3 Wh, according to the IEA. If all of the approximately 9 billion Google searches performed daily were switched to ChatGPT, that would increase the total power requirement by 10 TWh per year. Goldman Sachs Research (GSR) estimates that the amount of electricity used by data centers will swell by 160% over the next five years, and that they will account for 3% to 4% of global electricity use. In addition, their carbon emissions will likely double between 2022 and 2030. According to IEA figures, total power demand in Europe decreased for three years in a row but picked up in 2024 and should return to 2021 levels -- some 2,560 TWh per year -- by 2026. Nearly a third of this increase will be due to data centers. GSR estimates that the AI-related power demand at data centers will grow by approximately 200 TWh per year between 2023 and 2030. By 2028, AI should account for nearly 19% of data centers' energy consumption. However, the rapid expansion of generative AI could wrong-foot these forecasts. Chinese company DeepSeek is already shaking things up -- it introduced a generative AI program in late January that uses less energy than its US counterparts for both training algorithms and responding to prompts. Another factor that could stem the growth in AI power demand is the limited amount of mining resources available for producing chips. Nvidia currently dominates the market for AI chips, with a 95% market share. The three million Nvidia H100 chips installed around the world used 13.8 TWh of power in 2024 -- the same amount as Guatemala. By 2027, Nvidia chips could burn through 85 to 134 TWh of power. But will the company be able to produce them at that scale? Not always a sustainable choice Another factor to consider is whether our aging power grids will be able to support the additional load. Many of them, both nationally and locally, are already being pushed to the limit to meet current demand. And the fact that data centers are often concentrated geographically complicates things further. For example, data centers make up 20% of the power consumption in Ireland and over 25% in the U.S. state of Virginia. "Building data centers in regions where water and power supplies are already strained may not be the most sustainable choice," says Cubero-Castan. There's also the cost issue. If Google wanted to be able to process generative AI queries, it would need to set up 400,000 additional servers -- at a price tag of some 100 billion dollars, which would shrink its operating margin to zero. An unlikely scenario. Untapped benefits Some of the increase in power consumption caused by generative AI could be offset by the benefits of AI in general. Although training algorithms requires an investment, it could pay off in terms of energy savings or climate benefits. For instance, AI could speed the pace of innovation in the energy sector. That could help users to better predict and reduce their power use; enable utilities to manage their power grids more effectively; improve resource management; and allow engineers to run simulations and drive advances at the leading edge of modeling, climate economics, education and basic research. Whether we're able to leverage the benefits of this kind of innovation will depend on its impacts, how extensively the new technology is adopted by consumers, and how well policymakers understand it and draft laws to govern it. The next-generation data centers being built today are more energy efficient and allow for greater flexibility in how their capacity is used. By the same token, Nvidia is working to improve the performance of its chips while lowering their power requirement. And we shouldn't forget the potential of quantum computing. When it comes to data centers, the IEA calculates that 40% of the electricity they use goes to cooling, 40% to running servers and 20% to other system components including data storage and communication. At EPFL, Prof. Mario Paolone is heading up the Heating Bits initiative to build a demonstrator for testing new cooling methods. Five research groups and the EcoCloud Center have teamed up for the initiative, with the goal of developing new processes for heat recovery, cogeneration, incorporating renewable energy and optimizing server use. Keeping the bigger picture in mind Another (painless and free) way to cut data centers' power use is to clear out the clutter. Every day, companies worldwide generate 1.3 trillion gigabytes of data, most of which ends up as dark data, or data that are collected and stored but never used. Reseadrchers at Loughborough Business School estimate that 60% of the data kept today are dark data, and storing them emits just as much carbon as three million London-New York flights. This year's Digital Cleanup Day was held on 15 March, but you don't have to wait until spring to do your cleaning! Cubero-Castan warns us, however, to keep the bigger picture in mind: "If we begin using generative AI technology on a massive scale, with ever-bigger LLMs, the resulting energy gains will be far from enough to achieve a reduction in overall carbon emissions. Lowering our usage and increasing the lifespan and efficiency of our infrastructure remain essential." The energy impact of generative AI mustn't be overlooked, but for now it's only marginal at the global level -- it's simply adding to the already hefty power consumption of digital technology in general. Videos currently account for 70% to 80% of data traffic around the world, while other major contributors are multiplayer online games and cryptocurrency. The main drivers of power demand today are economic growth, electric vehicles, air-conditioning and manufacturing. And most of that power still comes from fossil fuels.

'One Supertanker Could Fuel the Entire US For a Year'

Former Meta CTO Mike Schroepfer believes AI can solve critical climate problems. The constant availability of AI tools, ready to answer questions round the clock, has truly made life easier. In fact, with the advent of models like ChatGPT, Perplexity, Claude and DeepSeek, AI is already at the cusp of becoming an integral part of the modern human lifestyle. These models already have a profound impact, often in ways users may not fully realise. Beyond their day-to-day utility, each prompt processed by AI consumes energy, water and resources that are vital for survival. Mike Schroepfer, founder of Gigascale Capital and former CTO of Meta, argued that energy is the critical bottleneck hindering the widespread adoption and advancement of AI. In a recent podcast interview, Schroepfer highlighted the urgent need to scale energy production to democratise access to AI globally and explored the intricate relationship between AI and climate change. He proposed using nuclear fusion energy as a solution. "If you wanted to multiply the power grid in the United States by five times and power all of it with fusion, one supertanker could fuel the entire US for a year." 5x More Power Needed Even Without AI Schroepfer, who spent nine years leading technical teams at Facebook, now invests in companies using technology to combat climate change. He emphasised that the surge in AI development and deployment places demands on energy infrastructure. He pointed out that even without AI, the United States needs about five times the existing power grid by 2050 to reach all its goals. Considering the power consumption of individual AI agents and then extrapolating that across billions of users. The numbers quickly become concerning, underscoring the scale of the energy challenge that lies ahead. Similarly, in terms of water usage, a 2023 study found that ChatGPT consumed 0.5 litres of water during each lengthy conversation with a user. When applied to millions of daily users across all AI systems and LLMs, the total water consumption becomes significant. Another 2027 projection showed that the world's demand for AI would lead to large amounts of water withdrawal -- freshwater taken from the ground or surface water sources, either temporarily or permanently. This demand, however, presents a unique opportunity to accelerate the adoption of innovative and sustainable energy solutions. 80% of New Energy in the US is Solar Solar energy is already making significant contributions in various parts of the world, including the US and India. India's solar energy sector is motivated by PM Surya Ghar: Muft Bijli Yojana, the world's largest rooftop solar initiative, which aims to bring solar power to one crore households by March 2027. As installations rapidly increase, with over 10 lakh installations expected by the end of this month, the numbers are expected to double by October this year, reach 40 lakhs by March 2026, and ultimately achieve the target of one crore. According to Schroepfe, solar energy has been widely adopted in the US, which most people don't know about. "80% of the new energy on the grid in the US in 2024 is solar, utility-grade solar." Regardless, it still has its limitations. Schroepfe said solar works 25% of the time, and it doesn't work at night or in cloudy conditions, which causes a "time-balance mismatch". This makes it necessary to explore alternative solutions for continuous power supply. While Schroepfe expresses that solar power's cost-effectiveness makes it an attractive option for expanding energy capacity in the US, it still remains an expensive mode of energy generation in countries like India. Another technique to investigate is nuclear fusion -- the same reaction that happens at the centre of the Sun. Globally, government labs and companies are racing to generate power from fusion. As per reports, even China has joined the nuclear fusion race, with an estimated budget of $1.5 billion. Other interesting projects also include offshore compute platforms that harness wave energy while simultaneously providing cooling for data centres. Moreover, in an interview with AIM, Ann Dunkin, former chief information officer at the US energy department, shared insights into how India can balance AI growth with sustainable energy solutions using renewable energy. She spoke about the 'all-of-the-above energy policy' that revolves around wind, geothermal, solar, clean hydrogen, and nuclear power. Hyperscalers Move to Nuclear "Most of the hyperscalers have announced purchase agreements for power for either existing or new nuclear power plants," Schroepfer said. He believes that market forces are the most effective driver for scaling and deploying these technologies. He notes that the AI sector's demand for reliable and affordable energy creates strong incentives for energy companies to innovate and expand their capabilities. Hyperscalers, who recognise that energy is now as important as data, computation, and algorithms, are already entering into power purchase agreements with both existing and next-generation power plants. Just as the declining cost of computing power enabled the rise of the internet and mobile technologies, a similar revolution in energy could unlock a wave of innovation across various sectors. Furthermore, AI has the potential to reduce global greenhouse gas (GHG) emissions by 5-10%, which is equivalent to the annual emissions of the entire European Union. AI could help manage grid challenges by simulating scenarios and finding the best ways to balance renewable energy fluctuations. In an earlier interview with AIM, Shantanu Som, Asia executive engineering leader at GE Vernova, explained how cutting-edge tools could transform the way we keep power systems running reliably and efficiently. This included predictive maintenance, solving with parity, generative AI, diagnostics, and troubleshooting. Last year, it was proposed that AI could help solve the problem faced by the biggest magnetic fusion facility in the US just by using deep reinforcement learning, according to research published in February 2024. As the world chases a development in achieving artificial general intelligence (AGI), Schroepfer believes while AI can be a tool for solving climate change, relying solely on AGI being an answer to anything is not sufficient.