Has anything upended utilities and power markets quite like AI has in 2024?
You’d have to look back to, say, 1890 and the rise of electric-powered aluminum smelting, when Alcoa had to build its own power plants because the grid lacked capacity. Or jump forward to the 1950s and the rise of air conditioning, which led to new power market dynamics that exist to this day like peak planning and “cooling degree days.”
But those transitions played out over decades, giving markets time to absorb their impacts. In contrast, AI has been adopted at an unprecedented pace. This is especially true of generative applications like ChatGPT, which reached over 300 million users this month.
Because these AI models benefit from being trained and housed in a single data center, it became quickly clear that each center would have to be much bigger than the norm. The most-requested size used to be in the 50-megawatt range, but now the industry is seeing proposals and plans for five- and even 10-gigawatt data centers. No one — not data center developers, infrastructure investors, utilities, or regulators — was ready for this.
In January, at the World Economic Forum’s meeting in Davos, OpenAI CEO Sam Altman said power demand from AI would be well beyond current forecasts, necessitating a breakthrough in energy supply. He pointed to both nuclear fission and fusion, along with solar and batteries, as critical solutions. (Altman is a lead investor and executive chairman at the fusion startup Helion Energy, and the small modular reactor startup Oklo went public in 2023 via Altman’s SPAC.)
But that picture of an AI boom powered by emerging clean technologies isn’t yet being realized. And while certain clean firm power options are still making their way to commercialization, the rush to build is complicated by concerns that fossil fuels will end up powering AI, as the gas industry remains all too ready to step up.
A new era of load growth
At the start of 2024, the prevailing forecasts pointed to 25 to 30 GW of demand from data centers over the next six years, and a 2% to 6% annual increase in overall power demand.
AI was the main driver, but industrialization and electrification also played a part. A narrative emerged that not only did the power industry lack the supply to meet demand, but also that the aging grid was not up to the challenge.
Some people didn’t quite buy the panic, though. “People just need to calm the heck down,” data center researcher Jonathan Koomey told Latitude Media in January. “They’re running around like chickens without heads when, honestly, we do not know what will happen.”
But as the year wore on, no one calmed down. Despite reassurances from the tech sector, the alarm only escalated as power providers reacted to the load growth. Many utilities made it clear the only way to meet demand in time was with fossil fuels. In April, for instance, Georgia Power included 1.4 GW of new natural gas-fired generation — as well as plans to purchase power from Mississippi, where much of the power is coal-fired — in its resource plan. Microsoft, a major Georgia Power customer, pushed back on that plan, arguing the utility wasn’t valuing renewables properly and doing a poor job of forecasting demand.
But there is no escaping the industry’s reliance on gas. Duke, Dominion, and several others added new gas power to their resource plans, even as they planned to add more nuclear and other clean firm sources as well. While many utilities are still framing gas as a “bridge fuel” to net zero, those new gas plant builds are each 40-50 year investments. It’s a stretch to think they will all be quickly relegated to emergency backup power as renewables become a bigger part of their resource mix.
And fossil fuel companies themselves are also beginning to operate at the AI-energy nexus. Recent announcements from Exxon and Chevron indicate their willingness to jump in with gas and carbon capture for on-site generation at data centers. But again, these would be their first builds of power plants like this, and may not present a clear path to sustainable low-carbon power generation.
An unprecedented opportunity for renewables
The hyperscalers, to their credit, have embraced the AI boom as an opportunity to push harder on all forms of clean firm energy. Speaking at Latitude Media’s Transition-AI 2024 conference earlier this month, Google’s head of data center energy Amanda Peterson Corio said that this demand is what the industry has “been waiting for.”
“Independent of policy, [it] can really drive innovation and change,” she said. “We think…3% of electricity load is data center load, but it contributes to the digital economy’s 10% of the U.S. GDP.”
This may be the most important takeaway for the year: the economic and strategic imperative to “win” the AI race will catalyze unprecedented investment in renewables, storage, and novel solutions to rapid power demand.
The range of announcements in 2024 has been startling:
- Microsoft signed a PPA with Constellation to support the restart of a retired nuclear reactor at Three Mile Island;
- Amazon signed three agreements to support the development of nuclear power projects, and acquired a data center campus from Talen Energy, powered in part by an adjacent nuclear power station;
- Google debuted the concept of the “clean transition tariff” with geothermal company Fervo Energy and the utility NV Energy, a new rate model that would accelerate the deployment of novel solutions; Duke debuted its own Accelerating Clean Energy tariff with Amazon, Google, Microsoft, and Nucor, and other utilities like Xcel Energy seem poised to embrace the concept as well;
- Meta signed a deal with Sage Geosystems for up to 150 MW of baseload power, starting in 2027;
- Oracle’s Larry Ellison said the company plans to power a new data center with up to a gigawatt of nuclear capacity;
- Dominion Energy and Commonwealth Fusion Systems announced a plan to bring the world’s first 400-MW fusion facility online in the next decade;
- AEP agreed to secure up to a gigawatt of solid oxide fuel cells from Bloom Energy for data centers and large energy users; and
And it’s clear that the money needed to power the boom is there. 2024 also saw a slew of partnerships in data center finance.
Blackstone, for instance, announced it will spend over $8 billion developing data centers in Spain; and, in its largest investment in the Asian market, Blackstone acquired data center company AirTrunk for $24 billion. In September, BlackRock, Global Infrastructure Partners, Microsoft, and MGX announced the formation of the Global AI Infrastructure Investment Partnership, with a plan to raise $30 billion.
And as the year came to a close, Google detailed a strategic partnership with Intersect Power and TPG Rise Climate to develop data centers. In a reversal of how data centers have typically been sited, they plan to bring fiber and other resources to those sites where grid-connected zero-carbon energy is already available — rather than bring power to places where fiber is already present.
There’s no end in sight
Everyone agrees that load will keep growing, but estimates of by how much vary. Earlier this month, the research firm Grid Strategies determined that electricity demand in the United States could more than quadruple in the next five years, to 128 GW. Georgia Power forecasts that electricity demand will triple by the mid-2030s, and a report from Lawrence Berkeley National Lab said pretty much the same.
It’s clear now that power availability will continue to be the primary bottleneck to data center development — not chips, water, finance or fiber. But AI itself may offer solutions to its own impact. AI-enabled digital twins can help utilities better design and plan for load growth. AI can also underpin grid-enhancing technologies like dynamic line rating, unlocking significant capacity on transmission grids. Grid flexibility can also enhance the operation of grids with large data loads, and AI has been demonstrated as a means of modeling and supporting new flexibility, from the distribution system to larger regional transmission networks.
In November, the utility giant PG&E released its R&D strategy report, which for the first time elaborated its vision to become an “AI-enabled utility.” The plan is to harness the technology to significantly improve grid planning and operational efficiency — and may be the clearest example yet of how AI can become central to a utility’s future, even as it weathers load growth.
