The Paradox of Limitations: How Constraints Drive Innovation in AI-Driven Energy Systems
Key Highlights
- Understanding the behavior of loads, not just their magnitude, is crucial for maintaining grid stability amid rising AI demands.
- AI enables real-time data analysis, predictive maintenance, and grid optimization, helping to address the challenges of fluctuating power requirements.
- The future of energy infrastructure depends on flexible power generation, advanced sensor integration, and smarter management strategies driven by AI.
Constraints impose limitations, but why’s that such a bad thing? Limitations can be good and unlock innovation.
Think about it. This seems like a paradox, of course, but one that makes nearly perfect sense throughout history and now this impending age of artificial intelligence.
AI cannot solve all problems, including the energy it sucks up like a gigawatt vacuum. But AI offers a world of potential for optimizing power generation, improving efficiency and delivery even as it demands unprecedented amounts.
The grid is in trouble as data center rack density expands and future AI factories will rise to 1 GW and beyond. This means trouble for substations, transformers and other connective electrical tissue in the digital ecosystem.
Key is understanding behavior of the system, not just magnitude of the load
“These assets were not meant for these kinds of loads we’re seeing right now,” Asim Akram, CEO of digital platform MultiSensor AI, said in an exclusive interview with EnergyTech.com.
“What is the price we pay in the end, is that these things go out of business,” Akram added. “It’s not the magnitude of the load, it’s the behavior of the load. The transient load goes up and down very quickly.”
Transient load is a growing challenge for both data center operators, hyperscalers and the energy infrastructure providers working to meet AI demand. As Rev Lebaredian, vice president of Omniverse at AI chip giant NVIDIA has put it, a traditional data center is essentially a warehouse or library which can be accessed with relatively small amounts of computer power.
An AI factory, however, is all that virtual warehoused material sometimes instantaneously activated into power-hungry manufacturing and refining operations. This creates transient load, which depends on the AI activity in the moment and can swing hundreds of megawatts or even gigawatts within milliseconds.
“You put into a factory all the raw materials and energy, and all the raw materials are reconfigured into it and out comes a refined production that is better than all of its parts,” Lebaredian said at RE+ late last year. “With the factory you want to maximize the density as much as possible.”
If that imagery doesn’t work, how about this: Try pushing your SUV zero to 60 in 10 seconds a couple of times per day and see how long the engine and transmission survive.
Therein lies the challenge for both the grid and distributed energy resources. Power generation must be flexible, but that’s a daunting task against transient load swings.
Meet the new solutions, same as the new challenges
“There is where AI comes in,” Akram pointed out. “What’s changed is the capacity and power to take data and make sense of it. The complexity of multi-sensors coming help us correlate. AI makes sense of it.”
That’s predictive maintenance, grid optimization, dynamic line rating, as well as virtual power plant aggregation. But it all begins with the transmission or distribution lines spitting out raw data and being digested by the one tool programmed to filter and discern the critical details in real time. Make that many tools in one, kind of like a digital Swiss army knife.
“No single type of sensor is sufficient for that environment,” Akram said.
Analysis paralysis is a real thing in human terms. Training an AI model to plunge head-first into information overload and bring up the cream of the data; that’s what it’s made for. To err is human and to AI hopefully can be divine or at least more human than human.
“It’s continuous visibility without the need for a human to be there,” Akram said.
Of course, us humans will be needed to ensure that AI discernment is on target.
Suman Kanuganti, CEO of Personal AI, has argued that a more distributed AI grid will offset some of the constraints. AI growth is reshaping conversations around power consumption and infrastructural readiness, he noted.
Getting smaller, more distributed and precise will help pave the way to the next generation of AI and power interaction.
“Let’s push token demand closer and closer to the user,” Kanuganti said. “The more distributed it is, the less burdening for the power system.”
Mastering Requirements, Standards and Risk
in a Digitally Transformer Energy Sector
Join us for July 21 EnergyTech Webinar with Siemens
About the Author
Rod Walton, EnergyTech Managing Editor
Managing Editor
For EnergyTech editorial inquiries, please contact Managing Editor Rod Walton at [email protected].
Rod Walton has spent 17 years covering the energy industry as a newspaper and trade journalist. He formerly was energy writer and business editor at the Tulsa World. Later, he spent six years covering the electricity power sector for Pennwell and Clarion Events. He joined Endeavor and EnergyTech in November 2021.
Walton earned his Bachelors degree in journalism from the University of Oklahoma. His career stops include the Moore American, Bartlesville Examiner-Enterprise, Wagoner Tribune and Tulsa World.
EnergyTech is focused on the mission critical and large-scale energy users and their sustainability and resiliency goals. These include the commercial and industrial sectors, as well as the military, universities, data centers and microgrids. The C&I sectors together account for close to 30 percent of greenhouse gas emissions in the U.S.
He was named Managing Editor for Microgrid Knowledge and EnergyTech starting July 1, 2023
Many large-scale energy users such as Fortune 500 companies, and mission-critical users such as military bases, universities, healthcare facilities, public safety and data centers, shifting their energy priorities to reach net-zero carbon goals within the coming decades. These include plans for renewable energy power purchase agreements, but also on-site resiliency projects such as microgrids, combined heat and power, rooftop solar, energy storage, digitalization and building efficiency upgrades.