Headlines imploring utilities to build new natural gas infrastructure to power AI and data centers are everywhere – they could also be the next AI bubble to burst.
Late last year, Meta announced plans for a $5 billion data center project in Louisiana, powered by 2,300 megawatts (MW) of new natural gas generation, enough to power approximately 390,000 homes. Morgan Stanley reports data center developers "will increasingly gravitate toward large complexes powered by natural gas-fired turbines and fuel cells.” NextEra Energy, the highest valued utility in the U.S., just signed a big deal to build more gas plants to meet rising power demand from AI.
But what if these assumptions are wrong? What if utilities and tech companies have talked themselves into a race to the bottom, building billions of dollars of infrastructure at risk of becoming stranded assets long before their retirement dates? What if, as one observer noted, this mad dash to build gas in service to data centers is entering “irrational exuberance territory?” And what if we can’t even build new gas in time to meet rising demand?
It wouldn’t be the first time forecasts for increased power demand were way off the mark.
Serious financial risks loom for utilities, regulators, and grid planners taking a simplified, short-term approach to this question by building huge amounts of new natural gas infrastructure. Unfortunately, consumers and the climate will bear the costs if things go wrong.
And a new development in AI undermines the calculus for how much energy AI actually needs.
The AI-Energy Demand Bubble May Have Just Burst
In the rush to build new gas for AI, utilities and tech companies are overlooking a fundamental premise—AI currently requires a tremendous amount of energy, all that power is incredibly expensive, and any innovator that finds a way to run AI using less energy will face an enormous competitive advantage.
Now, such an innovator may have burst the AI-energy demand bubble.
AI startup DeepSeek recently released a model with roughly the same capabilities as other large language models such as ChatGPT, but at a fraction of the cost. “DeepSeek’s emergence may offer a counterpoint to the widespread belief that the future of AI will require ever-increasing amounts of power and energy to develop,” reports Bloomberg.
Indeed, DeepSeek’s arrival has caused tech stocks to plummet, along with the stocks of power producers anticipating AI’s high energy demand.
Because the exact innovations DeepSeek provides are still unclear, uncertainty swirls around a new technology we’re still trying to understand, and innovation is clearly far from dried up, spending billions on new gas infrastructure looks like unwise long-term planning. DeepSeek seems to have caught the world by surprise, and the reality is no one knows what’s coming next, which means flexibility should be the priority right now, not locking us into expensive fossil fuel infrastructure that we already know will worsen the climate crisis and could very well end up being irrelevant to AI.
We’ve been here before on false energy demand predictions
Strong precedent exists for markets to find ways to improve efficiency, upending energy forecasts. In 2007, the U.S. Energy Information Administration (EIA) predicted demand would grow 21% over the ensuing 15 years.
That didn't happen.
785 terawatt hours of demand from EIA’s forecast never materialized. For context, that’s equivalent to roughly 19% of 2023 U.S. electricity generation. Energy efficiency strategies were the biggest contributors to lessening that thirst for power– between 2006 and 2021, utility efficiency programs avoided roughly 30% of projected demand growth over the same period.
A new era of rising power demand
Forecasts from pundits, market analysts, and utilities all project data centers will drive a sharp increase in electricity demand over the next half decade, as their thirst for power could double by 2030. Some have even wondered if “America is running out of power.”
It’s true that after years of stagnant growth, electricity demand is surging. But new data centers are gobbling up much of that power, and increasing electrification of buildings, transportation, and industry, as well as growing domestic manufacturing will all soon contribute to rising electricity demand.
But just how much and when remain an open question. This uncertainty means it’s crucial to make sure that utilities pursue least-regrets solutions, especially since the last time forecasters predicted a huge spike in demand, a huge chunk of it never materialized.
Addressing the problem rather than repeating the past
Faced with rising electricity demand for the first time in two decades, many utilities seem to be reverting to an overly simple solution—build more stuff, in this case gas. That would be a disaster for the climate and could lead to an “extraordinary’ rise in US electricity bills.”
It also runs counter to the findings of the tech industry and utilities. NextEra CEO John Ketchum says new wind and solar projects can be developed in 12 to 18 months whereas new gas generation is unlikely to be developed before 2030, it will only be possible to build it in certain parts of the country, and it will be expensive.
“This is an industry that really hasn’t seen any active development or construction in years,” Ketchum said.
Utilities have other solutions to meet rising electricity demand without locking themselves into expensive, time-intensive, polluting gas infrastructure. Consider energy parks, which integrate multiple renewable energy and storage solutions like batteries, potentially co-located with electricity consumers such as factories or data centers, all connected to the grid at a single point. Energy Innovation research shows energy parks can bring new power online faster than building new thermal generation or waiting in the interconnection queue since the project is co-located with large customers, which also reduces the need to build time-intensive transmission lines. Other research highlights solar microgrids as a fast, affordable solution.
The truth is we just don’t know what future energy demand will look like. Could AI cause the huge spikes many forecasts suggest? Sure! Or, as recent developments may indicate, perhaps it’s not the power hungry beast we all thought it would be. Given that uncertainty, consumer costs, and the climate crisis, we should demand flexible solutions right now, not huge gas buildouts. What utilities shouldn’t do, and regulators shouldn’t approve, is adding new gas without first considering all the available tools they can use to meet the oncoming challenge. Doing so creates new reliability issues while worsening climate pollution and public health, threatening higher consumer costs.