Renewables Fall Short of Meeting AI’s Massive Energy Needs

In the realm of artificial intelligence (AI), where data crunching and machine-learning algorithms reign supreme, the demand for energy has emerged as a critical concern. Mark P. Mills, the executive director of the National Center for Energy Analytics (an initiative I oversee at the Texas Public Policy Foundation), argues that the energy requirements for AI systems are far more substantial than most of us know. His insights paint a sobering picture of the energy landscape that awaits us as AI continues its relentless advance into every facet of modern life.

Mills contends that the computational intensity of AI applications, such as deep learning and real-time data processing, is driving an unprecedented surge in energy consumption. According to the International Energy Agency, the global electricity consumption by AI alone could reach 1,000 terawatt-hours (TWh) annually by 2026, slightly more than the total electricity consumption of Japan. The appetite will be formidable as it becomes integral to industries ranging from health care to finance, and transportation to agriculture.

At the heart of the debate lies a fundamental question: Can renewable energy sources adequately power the AI revolution? Silicon Valley, home to tech giants like Google, Facebook, and Tesla, has been a vocal advocate for renewable energy solutions. Many of these companies have committed to ambitious sustainability goals, including achieving carbon neutrality or even operating entirely on renewable energy. Most of these promises are hollow, at best, in that they rely on periodic renewable energy contracts to make the claim that they’re 100 percent renewable while connected to a grid stabilized and made reliable largely by traditional dispatchable thermal power — nuclear, natural gas, and even coal.

California is the nation’s test case for renewables. It’s the state with the most aggressive greenhouse gas reduction agenda. I voted against AB 32, the “California Global Warming Solutions Act of 2006,” which kicked off this effort. Back then, California’s electricity prices were the eighth most expensive in the nation and 44 percent above the national average. Today, after installing all that “cheap” solar and wind energy, California’s electricity prices are the second most expensive in the United States, trailing only Hawaii, with consumers paying almost double the national average.

Yet, while a grid dominated by renewables isn’t affordable, it’s also not reliable. Mills argues that though renewable sources like solar and wind have made significant strides, they face inherent limitations in meeting the continuous and predictable energy demands of AI systems.

The reality is stark: AI operations require uninterrupted power to function optimally. Unlike conventional electricity generation, where output can be adjusted to meet fluctuating demand, renewable sources depend on weather conditions and geographic location. This intermittency poses challenges for maintaining the stability and reliability of the power supply necessary for AI’s computational tasks, which often operate around the clock. The same can be said of chip fabrication as well as other industrial processes.

Moreover, the infrastructure needed to support AI’s energy demands goes beyond generating capacity. Mills points out that the transmission and storage of electricity — key components in ensuring a reliable energy supply — are critical bottlenecks that must be addressed to accommodate AI’s voracious appetite for power. Without substantial advancements in grid technology and energy storage solutions, the scalability of renewable energy in meeting AI’s needs remains a mirage — an expensive mirage.

A promising solution lies in the adoption of modular nuclear reactors and nuclear power in general. These technologies offer the continuous and reliable energy necessary for AI operations, providing a stable base load that complements intermittent renewable sources. Nuclear power, with its low carbon emissions and high energy density, is uniquely positioned to support the energy-intensive demands of AI.

Unfortunately, the regulatory process for permitting new nuclear power plants resembles a plate of spaghetti, with environmental lawsuits as the sauce on top. Only two new nuclear reactors have come online in the United States in the past three decades — Vogtle Units 3 and 4, which connected to the grid in July 2023 and April 2024 and “produce enough electricity to power 1 million homes.” China, on the other hand, has 55 nuclear reactors with 23 under construction, while India has more than 20 with seven more under construction. Rather than reduce the red tape, Congress has sought to pour subsidies on the problem — meaning that if nuclear power does get built here, it will take too long and cost too much.

Silicon Valley’s techno-optimism — and business plans — must be fueled by reliable power. But green tech advocates remain steadfast in their belief that renewables can and should power the AI future. However, the gap between aspiration and practicality is widening, sparking interesting political frictions in what used to be a close alliance.

The political and policy implications of this debate are profound. Germany is a cautionary example of a nation that grappled with decarbonization goals and commitments under the Paris Agreement, voluntarily starting the process of deindustrialization in service of green goals — something envisioned by the Morgenthau Plan in the aftermath of World War II as a punishment and a means of preventing the Germans from having the capacity to start another world war. Now Germany is faced with making a costly volte-face on energy if it is to avoid being entirely dependent on China, much less even try to participate in the AI space.

Furthermore, the economic dimensions of AI’s energy demands cannot be overlooked. Mills warns that overlooking the scale of energy consumption by AI could lead to supply constraints and price volatility in global energy markets. For industries reliant on AI technologies — from autonomous vehicles to smart grids — ensuring stable and affordable energy sources is essential for long-term viability and growth.

AI is coming, whether policymakers understand the energy implications or not. Since politicians aren’t likely to move fast enough, the fascinating thing to watch will be the necessity-driven transformation of Silicon Valley into a major energy-producing powerhouse.