18 November 2026

AI and Energy: A Paradox That Will Define the Next Decade

At the Smart Energy 2026 event in Sion, Katrin J. Yuan, director of the Swiss Future Institute and chair of the AI Future Council, made a blunt observation: artificial intelligence is neither the problem nor the solution for the energy system. It acts as an amplifier, for better or for worse, depending on how we choose to deploy it. After all, AI is already capable of saving energy or driving those savings.

To open her presentation, Katrin J. Yuan compared two figures. Training a model like GPT-4 consumed about 50 gigawatt-hours of electricity—equivalent to the annual electricity consumption of roughly 11,000 Swiss households. But today, a single query sent to Gemini consumes only 0.24 watt-hours—less than nine seconds of television, according to figures published by Google. “Same technology, two realities. Which one will prevail?”

Why AI Is an Energy Issue

Every response generated by artificial intelligence comes at a very real physical cost, Katrin J. Yuan pointed out. The computation involves tens of thousands of GPUs and specialized chips running continuously. Cooling can account for up to 40% of a data center’s energy consumption. Data centers themselves consumed 415 terawatt-hours worldwide in 2024, a figure that the International Energy Agency (IEA) predicts will more than double by 2030, reaching approximately 945 terawatt-hours (roughly equivalent to Japan’s total electricity consumption today). Finally, new AI campuses are now negotiating directly with grid operators for connections of several gigawatts.

One contrast struck her in particular: the Chinese model DeepSeek V3, which is comparable in class to Western models, required only about 2 gigawatt-hours and $5.6 million to train—roughly 25 times less than GPT-4. “The race for AI is therefore not just a race for power; it’s also a race for efficiency.”

The Opposing View: AI as a Driver of Solutions

But artificial intelligence also has largely untapped potential, the chair of the AI Future Council emphasized. DeepMind’s AlphaFold system has accelerated the mapping of protein structures by a factor of 45,000, paving the way for new materials for batteries, catalysts, and even fusion research. Yet, according to the IEA, only 2% of capital invested in energy startups currently funds AI-related projects. “The technology that is accused of being the problem is also our fastest tool for solving it,” she stated.

For Katrin J. Yuan, the question is therefore not whether AI threatens or safeguards the energy supply. “It’s not AI versus energy. AI is an amplifier—of the problem or the solution—depending on how we govern it.”

Transparency regarding consumption, infrastructure choices, and political will will, in her view, determine which way the scales tip.

Already Measurable Gains

Several concrete examples across multiple continents show that AI is already improving the energy system today. At Google, DeepMind has reduced the energy used for cooling its data centers by 40%. In the United Kingdom, National Grid has improved the accuracy of its solar generation forecasts by 33%. In China, AI-driven virtual power plants are already coordinating 3.5 gigawatts of peak demand in the Shanghai, Jiangsu, and Guangdong regions. Globally, the IEA estimates that AI-based analytical tools could unlock up to 175 gigawatts of additional transmission capacity without building a single new transmission line.

Katrin J. Yuan focused in particular on the case of Google, which she considers emblematic. “Google’s engineers didn’t add any servers. A neural network simply learned to control the cooling system better than a human operator, in real time, at the lowest PUE ever recorded at that site.” A silent, invisible, yet massive gain at scale—which, she says, is typical of what AI can bring to the energy sector.

A Look Toward 2050

According to the director of the Swiss Future Institute, three trends will shape the energy landscape in 2050: self-balancing grids, controlled by millions of AI agents that match local supply and demand in real time; disruptive technologies, with computing-accelerated breakthroughs in next-generation fusion and energy storage; and Switzerland as a trusted hub, an ideal testing ground for AI-energy pilot projects and governance standards. “None of this will happen automatically. It depends on the choices we make today,” she cautioned.

Katrin J. Yuan concluded her presentation with three key points. AI’s energy footprint is real and growing, but it is only part of the story. The same intelligence that consumes energy is already learning to conserve it and produce it more efficiently. And the bottom line is not a technological issue, but a matter of governance.

She also called for energy to be viewed as a strategic asset amid growing geopolitical rivalries, advocating for greater cooperation rather than isolated initiatives. “In the future, it won’t be raw intelligence that makes the difference, but the ability to ask the right questions and interpret the results.” She concluded with a warning that resonated throughout the room. “The real risk isn’t that AI will become smarter than us, but that we’ll lose our ability to think critically.”

Interview conducted on August 28, 2026, during the Smart Energy Event