The conversation around artificial intelligence and energy is becoming more sophisticated as it shifts from an initial focus on US fossil fuel supply as a way to power data centers to the potential to advance energy efficiency — potentially saving more energy than it uses. AI could also facilitate faster electrification of the energy system while easing some of the problems this creates.
- AI has the potential to unleash huge savings in energy demand.
Schneider Electric CEO Jean-Pascal Tricoire believes that AI can generate immediate gains of “roughly 20%” in most applications — including “every building, most of the manufacturing and certainly most of the infrastructure in cities” — at a modest energy cost of less than 1% in extra generation to power data centers.
AI will make “smart” buildings truly smart instead of just pretending they are, agreed Holly Paeper, in charge of commercial applications at heating, ventilation and air-conditioning company Trane Technologies, at the recent Climate Week in New York. “AI can sit on top of a building’s thermal management and continuously optimize it,” she said. “What we’re finding is that buildings can save 20%, 30%, 40% in what they’re using, it’s tremendous.”
These savings would feed through into a net positive impact on CO2 emissions. While AI could add around 300 million tons of CO2 emissions through 2035, its use could reduce emissions by five times as much, or 1.5 billion tons over that period, said Bank of America’s global head of sustainable and infrastructure finance, Karen Fang. Energy efficiency, which is often overlooked, plays an important role in all long-term climate plans. Primary energy consumption per unit of global GDP falls by 2.2% per year over 2023-50 in the International Energy Agency’s (IEA) base Stated Policies Scenario, for example, and by 3.1% per year in its Net-Zero Emissions Scenario, compared with an actual 1.6%/yr over 2010-23.
AI would also help power grids run more efficiently. Grids are an ideal use case, said GE Vernova’s head of grid automation, Marco Simiano, at the recent Reuters Energy Transition conference in New York. AI allows “nowcasting” — or real-time data analysis — to identify the need for immediate action, such as very fast load-shedding to avoid incidents, for example. The next step will be to switch from smart grids, where human decisions are assisted by technology, to fully autonomous systems, said Jim Taylor, head of grid software and smart infrastructure at Siemens USA.
- While it’s hard to predict future AI applications, many point to accelerated electrification.
Beyond pure efficiency and grid management, examples of potential AI applications in the energy sector include supply and demand forecasting, distributed generation and storage management, intelligent charging and battery management, and autonomous trading. AI will also facilitate predictive maintenance and truly autonomous operation of equipment.
AI could even help expedite licensing of new equipment, from rooftop solar panels to nuclear reactors. Zackary Rad, chief regulatory officer of energy consultancy LucidCatalyst, said AI could track complex data through the application process and adjust it as needed, then be used later, for example, when an audit occurs. Similarly, an AI system could help identify which rooftops in a city are best suited for solar panels, possibly in combination with batteries and electric vehicle (EV) charging locations, determine acceptable technical solutions and automatically deliver permits, said think tank Ember’s Kingsmill Bond.
The main common point between most potential AI applications is that they involve electrification. This is essentially because AI communicates with the world via electrical signals. AI can, of course, interact with every kind of system or equipment, but it is easier and more efficient when done directly. EVs, for example, are much easier to automate than conventional vehicles. In addition to being simpler and more efficient, their core systems, such as acceleration, regenerative braking and battery management, are electrical and controlled electrically.
Electrification, in turn, inherently brings efficiency because it involves less energy wastage in combustion — EVs and heat pumps are more efficient than internal combustion engine vehicles and boilers. But electrification also facilitates digitization and connectivity, said Schneider’s Tricoire. “If you look at the next two decades, nobody has a doubt that they’re going to be the decades of AI,” he said. “But you can’t dissociate AI from electrification: Whoever gets an advantage on electrification will get an advantage on AI.”
AI could also help ease some of the challenges thrown up by rapid electrification. Bond, also noting the “close affinity” between AI and electricity, predicted that AI would turn a current problem, renewable power intermittency, into an opportunity. UK energy company Octopus has started to offer vehicle-to-grid bundles, where drivers get free electricity in exchange for turning their EVs into grid assets providing flexibility to the system, he noted. “We’re moving from central sources of supply and largely passive demand to a system where you’re going to have millions of sources of supply and potentially billions of sources of demand having to be balanced in real time,” he said. “Only AI has the capacity to handle that level of complexity.”
- Gas will still play an important role in the US over the medium term.
Most speakers during Climate Week agreed that natural gas will be the fuel of choice to power AI in the US for the next five years or so because it is cheap and readily available. Timing is of the essence, as AI companies are engaged in a race for deployment and market share. Utilities will maximize use of existing capacity, squeezing more from gas and nuclear plants, said Joseph Dominguez, CEO of Constellation Energy, a major US nuclear operator that earlier this year also bought independent power producer Calpine and its 26 gigawatts of gas-fired capacity.
Given time constraints, many hyperscalers may choose to build onsite gas turbines to initially power their data centers and later connect them to the grid, said data center operator Crusoe’s CFO, Matthew DeNezza. With time, as grids improve and capacity expands, those turbines could shift to backup generation while also offering grid services such as frequency and voltage regulation, other speakers said.
But gas has the “inherent disadvantage” of price volatility over the 20-plus years over which a new asset is financed, with the additional drawback of being a fossil fuel likely to be constrained by future regulation, even in the US, Dominguez said. “I think the call for sustainable technologies that don’t have either the risk of an underlying fuel price or the risk of a change in law is pretty significant.”
- China’s rapid electrification should position it well.
Power consumption in China grew by an average 6.8% per year over 2010-23, according to IEA data, compared with 0.3% in the US and minus 0.4% in Europe. China has been rapidly expanding its generating capacity to feed this growth, meaning that AI’s additional power consumption — even if it ends up matching that of the US — would more easily be served by the grid.
The other key difference between China and the West is its rate of electrification, or the share of energy end-use met by electricity. This has been largely stagnant in the US and Europe at around 20% but has quadrupled in China since 2010 to around 25%, and is expected to double again by 2050, according to consultancy DNV. With electrification and AI expected to feed off each other, this raises serious “competitiveness concerns” for the US and Europe, Tricoire warned.
More generally, he believes AI will help the Global South leapfrog fossil fuel-based centralized energy systems by moving straight to decentralized electrical systems, which are more complex and easier to manage with AI.
link
