Could AI ignite a power crisis?

It’s hard to escape conversations about how we will power artificial intelligence (AI) in the industry right now, but which will be more disruptive: energy for AI or how AI will transform energy markets?

Powering AI is a “right here, right now” problem. The big tech companies are in a race to build AI data centers but they also have ambitious net zero targets. Meanwhile, there are early indications that AI will profoundly change not just our energy systems but potentially some of the underlying technologies that will power the planet.

Longer term, I believe AI will accelerate efforts to make energy cheaper and cleaner.

Utilities and energy-intensive businesses, together with policy makers, have some tough choices to make. They need to establish what role they want to play in the data center gold rush.

In the meantime, utilities and energy-intensive businesses (especially industrials, automotive and big tech companies), together with policy makers, have some tough choices to make. They need to establish what role they want to play in the data center gold rush. They also need to determine how AI will transform their core business and find the right partners to collaborate with to meet this once-in-a-lifetime opportunity. Or is that challenge?

AI for energy: where’s the value?

AI’s ability to analyze vast amounts of data, generate accurate models and more is?revolutionizing the way that the energy business operates — from AI-driven, end-to-end workflows to how power is generated, distributed and consumed.

Because of its ability to process massive amounts of data, AI has the potential to help solve some of the most complex challenges facing the industry. In the last two years, EY research shows a fivefold rise in the number of energy companies mentioning AI on their earnings calls and a threefold rise for industrial companies. Here’s where they are seeing value from AI:

• Energy companies are using AI to transform upstream, midstream and downstream operations, from optimizing drilling to analyzing seismic data, as well as improving production forecasting, safety, predictive maintenance, grid management and integrating renewable energy sources.
• Automotive companies are harnessing AI beyond vehicle autonomy for?predictive maintenance and process optimization in manufacturing, as well as optimizing battery management, energy storage and vehicle design to not only improve energy efficiency but utilize vehicles as energy storage.
• Industrial companies are using AI for smart energy management solutions for buildings and facilities, and?optimizing energy use in industrial automation, robotics, smart grids and process control systems.

AI also has huge potential for product innovation. One example is fusion. I remember in 1993, when I visited the European fusion program as an overly optimistic undergraduate and was persuaded that tritium powered fusion was less than a decade away. There it remains, due to considerable technical challenges. Will AI help unlock fusion? Maybe. One university and a tech company have had encouraging results using AI to predict plasma instabilities, so watch this space.?

Energy for AI: what’s the problem?

AI applications require high-speed processing of data – this is what is driving energy consumption. And those demands are on the rise.

AI models, particularly large language models, consume massive amounts of power. Amounts it’s hard to comprehend. In many markets, this is putting a strain on already stressed grids and limiting further enhancement. The challenge is on to find ways to make AI more efficient.

In many markets, AI is putting a strain on already stressed grids and limiting further enhancement. The challenge is on to make AI more efficient.

The high energy consumption from training and operating AI models is due to these models’ reliance on huge databases, and the cost of moving data between computing and memory, in and between chips. When training a large AI model, up to 90% of the energy is spent accessing memory[1], driven by AI web-based searches. A typical request to generative AI chatbots consumes 10 kilojoules — roughly 10 times as much as a conventional search.

Could a lack of power shape the AI revolution?

Big tech companies are rushing to build data centers to power AI. While the International Energy Authority (IEA) says data centers account for around 2% today, the growth rate could skyrocket between 35% to 128% by 2026[2]. This would be the equivalent of adding the annual energy consumption of Sweden or Germany at the top end. However, sceptics will be aware that the industry has been here before with growth projections for data centers that never materialized: in 2007 with the internet boom, and 2017 with cryptocurrency mining[3].

What’s clear is that a new scale of data centers is needed for AI. While traditional centers consume 5-10 megawatts (MW) of electricity, some of the new AI data centers are 100-150MW[4] — ?as much electricity as 400,000 electric vehicles or 6 steel mills. The most ambitious are planning mega data centers of 5GW[5].

The race to power data centers 24/7 is also a local one: in hubs like Virginia and Ireland, for example, they absorb 20-25% of local power consumption. It’s not unlike the first industrial revolution, where small towns of 5000 people had dozens of coal plants to power the factories, mills and foundries.

Connecting to power is the top constraint for new data centers, lengthening lead times from 1-2 years to 4-8 years. Grids remain the real bottleneck.

Connecting to power is the top constraint for new data centers, lengthening lead times from 1-2 years to 4-8 years. It’s not about generation — there will be more than enough supply, though it must be green given the tech players’ ambitious carbon commitments. Grids remain the real bottleneck, with transmission and permitting for desired sites an ongoing headache.?

This is evidenced by, for example, higher power outages in US cities, where nearly half of the almost 11,000 data center locations worldwide are located[6]. Trillions of dollars are needed to integrate renewable energy, modernize the grid and improve energy efficiency — and that was before AI.

?While 86% of new generation capacity in 2023 was renewable energy[7], it requires heavy investment in storage and power system flexibility to provide the reliability sought by the world’s most valuable companies. As they seek to establish dominance in AI, might the tech giants reach for the reliability and expediency of gas over renewables? That certainly was the case for one leading company, which is powering its AI data center largely with powerful mobile gas turbines and mega-battery packs until it can be connected to the grid.

Crunch time: how can we fulfil rising energy demand more sustainably and affordably??

Big Tech need to win the AI race but balance net zero commitments. Their recent environmental reports disclose increases in carbon emissions of 30-40% since 2020.?

There’s also a risk that the considerable costs of powering AI could limit accessibility for smaller organizations and effectively knock out second tier tech companies. The costs of the real estate, talent, infrastructure and power needed to train generative AI models could restrict this to a handful of players who can afford the price tag and wield influence with power companies.

Tech giants and industrial companies want access to low carbon, scalable power in locations close to data centers. And given the backlog of nearly 2600GW of generation and storage capacity actively seeking grid interconnection in the US[8], they’re not waiting for utilities to supply it.

Tech giants and industrial companies want access to low carbon, scalable power in locations close to data centers. And given the backlog of nearly 2600GW of generation and storage capacity actively seeking grid interconnection in the US[8], they’re not waiting for utilities to supply it. For example, Microsoft signed a 20-year PPA with Constellation Energy[9] to bring the decommissioned nuclear plant at Three Mile Island in the US back online. Small modular reactors are attracting new interest but these won’t be ready before 2030-2035. And new types of strategic partnerships are emerging to place new clean energy and storage plants next to planned data centers. The earliest these are expected to be operational is early 2028 for first movers.

What role do you want to play?

AI is shaping the path to autonomy across industries, revolutionizing how businesses process and act on vast amounts of data.

But there’s a lot we don’t know right now. Is the forecast power demand for data centers real or a bubble? Will data centers get massively more energy efficient? Will tech companies innovate processors beyond GPUs, e.g., with quantum computing? Will AI power demands be met by renewables, nuclear or gas?

For companies looking to unlock new avenues for growth, AI represents a major opportunity. Powering AI will unlock massive investment across the energy industry and beyond. So despite demand remaining an unknown, we need to think beyond the rush for data centers and consider the bigger picture. What role do you want to play in the energy transformation that’s already underway, and how should you position yourself today for the AI of tomorrow?

[1] Bourzac, Katherine, “Fixing AI’s Energy Crisis,” Nature, 17 October 2024.

[2] Electricity 2024: Analysis and Forecast to 2027, International Energy Authority, January 2024.

[3] Liebreich, Michael, “Generative AI: The Power and the Glory,” BloombergNEF, 24 December 2024.

[4] Spencer, Thomas, “What the data centre and AI boom could mean for the energy sector,” International Energy Authority website, 18 October 2024. www.iea.org

[5] Liebreich, “Generative AI.”

[6] Minnix, John, “115 Global Data Center Stats You Should Know in 2024,” Brightlio, 22 April 2024.

[7] Renewable capacity statistics 2024, International Renewable Energy Authority, March 2024.

[8] Rand, Joseph, “Grid connection backlog grows by 30% in 2023, dominated by requests for solar, wind, and energy storage,” Berkeley Lab Energy Markets & Policy, April 2024.

[9] “Constellation to Launch Crane Clean Energy Center, Restoring Jobs and Carbon-Free Power to The Grid,” 20 September 2024. www.constellationenergy.com