The artificial intelligence revolution has a power problem.
As AI adoption accelerates across every industry, the data centers powering these innovations face an unprecedented challenge: securing enough electricity to keep pace with skyrocketing computational demands.
Bloom Energy’s 2025 Data Center Power Report, which surveyed hyperscalers, colocation developers, and utilities, revealed a striking shift in priorities: power availability has decisively emerged as the leading factor for data center site selection. In fact, 84% of respondents ranked power availability among their top three considerations for site selection.
As recently as one year ago, proximity to fiber optic networks competed equally with power for the top spot, explained Bala Naidu, Bloom’s vice president of energy transition solutions. Today, power availability outweighs all other considerations — a clear reflection of the new reality for AI infrastructure.
AI is driving unprecedented power demand
Some data center campuses now require over a gigawatt of power, enough to supply more than 800,000 average American homes. By 2028, data centers are projected to consume 12% of nationwide electricity, up from just 4% in 2023. In Virginia alone, 25% of electricity already flows to data centers.
Median facility sizes are also growing rapidly, from approximately 175 megawatts today to an expected 275 MW by 2030 and 375 MW by 2035. To put that in context, today’s AI chips are so power-hungry that a single server rack can consume up to one megawatt — enough electricity to power 750 homes.
Underestimated timelines for grid power
To add another layer, data center developers are significantly more optimistic than utility providers about power delivery schedules. In high-growth markets like Northern Virginia, Dallas-Fort Worth, and Phoenix, utilities report timelines that are typically one to two years beyond what hyperscalers expect.
The implications of this misalignment are substantial. “In the AI race, waiting for power means billions of dollars in capital investment are depreciating rapidly,” Naidu said. “Fuel cells can help data centers avoid this by accelerating training model development and delivering better ROI compared to waiting years for power.”
Onsite power generation moves from backup to backbone
In response, data centers are rapidly pivoting to onsite power generation. Bloom’s survey data shows that 38% of data centers expect to incorporate onsite generation by 2030. Even more striking: 27% plan to be fully powered by onsite generation by 2030 — up from just 1% last year.
This rapid shift is being driven by the limitations of traditional centralized power generation, which requires extensive infrastructure expansions that can take years to build. At the same time, supply chains for turbines, engines, and high-voltage equipment remain strained, causing additional delays.
Why fuel cells are ideal for the AI era
Faced with these compounding challenges, data center operators recognize they can no longer rely solely on conventional power sources. Fuel cells are well-suited to bridge this growing gap. They match the efficiency of large combined-cycle gas turbines at the point of use and eliminate transmission and distribution losses.
Reliability — ranked as “non-negotiable” by survey respondents — becomes achievable with minimal redundancy. Bloom Energy’s fuel cell systems are based on 65-kilowatt building blocks and need just 10% overbuild capacity to maintain 99.9% availability, while turbines and engines require 20-50% more capacity for similar reliability
Fuel cells also excel at load flexibility, critical for AI workloads that create highly variable power demand. They maintain flat efficiency across load levels, adapting seamlessly to dynamic requirements while minimizing emissions and costs.
Cleaner power without the permitting delays
For projects with combustion-based power generation, air quality regulations add further complexity and potential for delays. Nearly half (46%) of planned data center sites are located in non-attainment zones, where stricter emissions permitting applies. In some cases, according to Naidu, “permitting can take up to two years, especially in air districts with severe non-attainment status.”
Fuel cells offer a clear advantage. They emit virtually no nitrogen oxide (NOx) — a major contributor to air pollution — and in many regions, are exempt from air quality permitting. “Why take permitting risk when you can utilize natural gas to deliver high-efficiency electricity without environmental impact?” Naidu said.
A future-proof solution
Natural gas remains critical to meeting data center power needs today. Bloom Energy’s solid oxide fuel cells serve as a rapidly deployable electricity transition technology, using natural gas in the cleanest way possible — through a non-combustion process that operates at significantly higher efficiency than traditional combustion-based resources. “Fuel cells powered by natural gas produce 30% lower greenhouse gas emissions versus turbines or reciprocating engines,” explained Naidu.
Beyond immediate emissions benefits, the platform’s fuel flexibility and modular design ensure it won’t become a stranded asset. These fuel cells are carbon capture-ready and capable of running on hydrogen, positioning them as a bridge technology that can evolve with decarbonization goals — offering developers a path to make progress on sustainability while maintaining the reliability their operations demand.
Onsite power as a long-term strategy for data centers
For executives investing in AI power infrastructure, the path forward is clear. Onsite power has emerged as an attractive solution to power large scale data centers, with fuel cells being able to deliver reliable power with lower emissions and fewer regulatory hurdles than combustion engines.The economics have also favorably shifted: Fuel cells are cost competitive with gas turbines and reciprocating engines. Upfront costs can look higher than conventional combustion alternatives — until you factor in the total cost to deliver the same level of reliability and cost of fuel, which for fuel cells, can be 30% lower due to higher operational efficiency. Companies that move quickly to adopt onsite solutions will lead the next wave of digital transformation, while those waiting for the grid to catch up risk being left behind.
This is partner content, brought to you by Bloom Energy.
