Share this
This blog is a recap of Peter Kelly-Detwiler’s presentation at Yes Energy’s EMPOWER conference.
We’re standing on the precipice of change. Artificial intelligence (AI) will change the world in ways we can scarcely begin to imagine. For the utility sector, AI presents enormous short-term supply challenges for the grid but also holds significant promise for a more efficient, transparent, and abundant energy future.
When ChatGPT was announced in late 2022, the energy industry experienced a seismic shift. ChatGPT is a generative AI (gen AI), a type of artificial intelligence that learns from trillions of “training” data points to create new content like text, images, and even music.
Traditional artificial intelligence or machine learning is simpler, using basic algorithms and if/then statements. Generative AI goes further – it infers, is creative, and responds like the human brain. It has the potential to help humanity think better and become more efficient, drive our autonomous vehicles, and improve our agricultural processes.
For all its promise, gen AI is extremely power-hungry, and that’s why data centers, which house the machines behind AI, are the hot topic in energy circles today. AI systems can require thousands of graphics processing units (GPUs), each demanding up to 700 watts or more of power. With each new generation of GPU demanding more power and increased demand for AI-powered applications, experts predict data center power demand could double or even triple by 2030.
AI Presents Challenges for Utilities
The explosion of AI presents several challenges for utilities, not the least of which is the need for additional generation capacity. The utility industry will have to marshal a lot of capital to generate enough power to satisfy the demand of AI data centers.
For example, Entergy has proposed developing a $3.2 billion 1.5 GW natural gas plant to serve the load of a new Meta data center in Louisiana.
And then there’s the infrastructure problem. With the current transmission infrastructure near capacity, utilities will need to invest in transmission and distribution equipment to deliver power to new AI data centers.
Planning for these new loads is complicated because there are still many unanswered questions about the nascent industry.
- What will the AI landscape look like in 10 years?
Utility supply assets have a lifecycle of 30 to 40 years. But your typical data center supply contract may include a four-year ramp period to build and bring the asset online and then a 12-year contract with the utility, so there’s a big temporal mismatch. If the contract terminates, the utility is left with a lot of infrastructure it needs somebody else to pay for. Then there’s the question as to how many winners there will be in this new and highly competitive industry. Anyone who was around during the dot.com era knows that not every company playing in the AI space today will be a winner. The potential mismatch between forecasted and actual load once the winners and losers have been declared poses a significant risk for utilities.
Will the survivors gobble up the abandoned generation capacity, or will the utility have stranded assets? Woe to the utility that builds out the generation and transmission capacity only to have the customer fade into the annals of history.
- How much will AI really cost?
Earlier this year, DeepSeek, a Chinese gen AI company, said it had built a gen AI model for about $6 million using a relatively small number of older, less powerful chips, as well as less energy to train its model. Many of the company’s claims have been debunked in the months since the announcement, but the possibility that AI could be done with less power sent shockwaves through the industry.
If today’s cost estimates are overblown, will society use more gen AI because it’s cheaper (a phenomenon known as Jevon’s Paradox)? Some say yes, but no one knows yet how this will play out.
- How will technological advancements change power consumption?
Data center cooling accounts for roughly 38% of a data center’s total load; however, advancements in technologies like liquid cooling, where chips are immersed in a dielectric fluid, could cut that by 90%. The chips used in AI are also getting more energy efficient. Will efficiency drive down future demand? Time will tell. Today, we simply don't know what long-term equilibrium looks like.
- How many data centers will organizations build?
Experience has shown that only about 19% of projects in the supply interconnection queue actually make it through and eventually flow power. Developers frequently submit multiple power project applications in order to increase their options and chances of success. Recent reports indicate that data centers may be hedging their bets in a similar fashion, by submitting applications to interconnect at multiple locations, only to select the site that offers the quickest route to power. These so-called phantom load applications may be skewing the number of planned data centers. Recently passed Texas Senate Bill 6, if it passes the House and gets signed into law, would address this issue, requiring data center applicants to divulge where else they're seeking power.
How the Energy Industry Can Leverage AI
For all the challenges AI presents, it also has the potential to transform the energy sector. Utilities can employ AI-powered tools for everything from advanced wildfire detection to optimizing energy use, properly siting distributed energy resources (DERs), managing supply and demand, and making power markets more efficient.
AI is also well-suited to address supply-side applications in the utility and grid operator interconnection queues. It can ensure applications are correct and complete and rapidly assess the power flow implications of any changes or application withdrawals.
It can also model and manage virtual power plants and bi-directional power flows, providing insights across the entire transmission and distribution system.
What could the energy industry do if it could get accurate, real-time, locationally specific data? The answer is a lot. Today's grid is utilized at around the 42% capacity/load factor. AI can help the industry figure out ways to push down the peaks and pull up the valleys through efficiencies, DERs, and better dispatch, helping us create a more efficient grid than we have today.
Conclusion
Artificial intelligence is an incredibly powerful new tool – one that we still don't fully understand and haven't yet fully developed. However, it’s clear that AI has astronomically important implications for the future of energy.
Yes Energy’s EnCompass solution is a power flow modeling solution that helps you produce market price forecasts, analyze generation and transmission development, and make informed decisions in the transitioning power grid. It’s the software of choice for making optimal power supply decisions from short-term scheduling and trading to long-term capital investment.
In addition, Yes Energy’s Demand Forecasts enable you to make the most informed decisions possible when buying and selling in energy markets worldwide. They help you manage risk and gain a competitive edge with forecasts for electric load, solar, wind generation, gas consumption, or district heating and cooling. They’re highly accurate forecasts, especially on unpredictable days, using an econometric model that a team of data analysts constantly tune.
Contact Yes Energy to learn how our solutions can supercharge your participation in energy markets.
About the speaker: Peter Kelly-Detwiler is an energy industry thought leader, and he consults and speaks all over the world. He is the author of "The Energy Switch: How Companies and Customers Are Transforming the Electrical Grid and the Future of Power." Find Peter Kelly-Detwiler on LinkedIn.
