As tech giants race to build increasingly powerful artificial intelligence models, their data centers are demanding electricity at a scale that’s straining traditional power infrastructure. A single modern AI training facility can require over a gigawatt of power — enough to supply almost 1 million homes — and the grid simply can’t keep up.
With trillions of dollars in investment in AI infrastructure facing a critical power bottleneck, the industry is turning to fuel cells as part of the solution. After undergoing a quiet revolution in cost and scalability, fuel cells have emerged as a proven technology for onsite power generation.
The grid gap and industry response
In the United States, the Department of Energy reports that data center electricity usage climbed from 58 terawatt-hours in 2014 to 176 TWh in 2023; it estimates a further increase to between 325 TWh and 580 TWh by 2028.
This explosive growth has created a new reality for data center developers. Bloom Energy’s recent data center power survey found that power availability now decisively outranks all other considerations for site selection, with 84% of data center operators ranking it among their top three factors.
The response to planning for this growth has been dramatic: 27% of data centers now plan to run entirely on onsite power generation by 2030, up from just 1% last year, according to the Bloom survey. This shift represents more than a temporary workaround; it’s a complete reimagining of how to power data centers.
The no-combustion advantage
Unlike traditional generators that burn fuel to create electricity through combustion and mechanical motion, fuel cells use an electrochemical process to convert natural gas directly into power. The key distinction: Traditional generators rely on burning fuel and spinning turbines (with all the mechanical complexity and efficiency losses that entails), while fuel cells generate electricity through a direct chemical reaction, no combustion or moving parts required.
DOE notes that fuel cells can convert chemical energy directly to electrical energy with efficiencies capable of exceeding 60% — significantly higher than combustion engines. They also operate quietly, with few moving parts, and produce lower emissions compared to conventional generators.
This fundamental difference delivers critical advantages for AI workloads. Solid oxide fuel cells like Bloom’s achieve 54% fuel efficiency at the point of use. Their higher efficiency can translate into savings of between $70 million and $100 million in fuel costs for the median data center (175 MW) over five years, assuming $5 per MMBtu gas costs, according to a recent Bloom white paper.
Rapid deployment and scalability
Perhaps the most compelling advantage in today’s data center market is deployment speed. For example, Bloom Energy reports they can deliver 100 megawatts of onsite power in as little as 90 days. The company’s modular systems arrive prefabricated and can scale from single megawatts to hundreds, growing incrementally with compute demand, which can help both hyperscalers and smaller data centers expand rapidly.
In a recent Latitude Media and Bloom Energy webinar, Tom Hellstern, a McKinsey partner, noted that “in some very severely power constrained markets, like Ohio or Northern Virginia, we’ve actually seen some of the smaller inferencing data centers that are on the order of 100 to 200 MW come with their own onsite power as well.”
Razvan Panati, Bloom’s VP of product strategy, explained why fuel cells are particularly well-suited for these deployments: “Because of the modular design, fuel cells can achieve 99.99% uptime with minimal redundancy and just 10% additional capacity. With traditional turbines or engines, you’d need to double or triple your installed capacity to reach the same reliability levels.”
According to Bloom’s specifications, fuel cells can be installed at ground level delivering 30 MW per acre, or stacked vertically to achieve 100 MW per acre, which is critical for space-constrained data centers in urban areas.
Fuel cells also produce virtually no nitrogen oxide emissions, often exempting them from air quality permits that can delay combustion-based projects. This becomes particularly relevant as nearly half of planned data center sites are located in nonattainment zones where stricter emissions permitting applies, according to Bloom’s analysis.
Solving the power volatility problem
One underdiscussed challenge in AI infrastructure is power volatility. Testing has shown that AI power loads can swing from 20% to more than 150% of provisioned power in milliseconds — and tens of swings that can happen every minute, according to the Bloom white paper. Traditional combustion generators, with their mechanical lag, struggle to respond to these rapid fluctuations.
Because fuel cells have no moving parts and generate electricity through a direct electrochemical process, they respond at least twice as fast as rotating generators when stepping up, and instantly when stepping down. When combined with supercapacitors, they reach 100% in milliseconds.
Fuel cells also maintain high efficiency at partial load, making them uniquely suited to handle the volatility and density of AI workloads.
Cost competitiveness and ROI
What was once considered prohibitively expensive technology has started to come down the cost curve. Over the past decade, Bloom Energy has continuously innovated and relentlessly improved the technology, reducing costs by double-digit percentages year after year.
While capital expenditure for Bloom Energy fuel cells is about 10% to 15% higher on a nameplate power capacity (kW or MW rating) basis versus small gas turbines and engines, this cost difference shrinks when factoring in the overbuild required to achieve grid-parity availability in an islanded microgrid, i.e. 99.9% uptime. The amount of overbuild needed depends on the type of power generation technology used.
Meeting data center power needs, today
For data center operators facing immediate power needs, fuel cells are a compelling solution: cleaner than traditional generators, fast to deploy, and built for efficiency and reliability. Solving the power bottleneck is central to winning the AI race. For an industry prioritizing speed to power, fuel cells are emerging as a near-term solution to include in procurement strategies.
This is partner content, brought to you by Bloom Energy.
