Data centers turn to fuel cells: the value of earlier power and the costs that remain

Bloom Energy reported third-quarter company revenue of $1,235.8 million on October 22, up 138.1% year over year, not data-center revenue alone.¹ A Reuters explainer the following day revisited AI data centers' demand for fuel-cell power.² The developments bring an infrastructure question into focus: operators need more than bigger electricity commitments. They need to know when reliable power will become available.

Archive photograph: Bloom Energy fuel-cell power equipment on display, February 24, 2010. “Bloom Energy Servers”, photograph by © Jakub Mosur, published by Bloom Energy. Source: Wikimedia Commons; CC BY 2.0.

First establish what the large numbers measure

Rystad’s June 25 forecast puts annual fuel-cell market revenue at $2.8 billion for 2025 and $30 billion for 2030: yearly revenue, not cumulative investment or a new October forecast.³

Bloom’s April 13 announcement described Oracle’s intended procurement of up to 2.8 GW, with 1.2 GW contracted: a ceiling, not firm orders for the entire amount or proof of operational capacity.⁴

Understanding the market thus requires two sets of measures. Revenue is money over a period; generating capacity measures the power equipment can provide. Without equipment prices, service scope and revenue-recognition schedules, one cannot be converted directly into the other. Rapid company-wide revenue growth also cannot establish a particular customer category's contribution or demonstrate that every installation is delivering its promised performance.

Why earlier electricity can be valuable

The US Department of Energy explains that fuel cells convert fuel's chemical energy into electricity through an electrochemical process. They can keep generating while fuel is supplied, rather than following an ordinary battery's charge-then-discharge cycle.⁵ Placing generation near a load therefore provides another supply route, not an energy source requiring no inputs.

In project economics, earlier electricity could reduce idle time if servers and buildings are ready but grid access remains uncertain. The value comes from starting operations sooner, which is different from paying less per unit of electricity. A sound comparison places the operating value of that earlier start against equipment, fuel, maintenance and backup costs over a consistent period.

Different campuses can consequently reach different conclusions. A project whose construction is unfinished may not immediately realize the value of earlier power. One with grid access already available may not need to pay for another long-term generation system. Fuel cells' appeal depends on a project's critical schedule constraint, rather than a single electricity price shared by all AI facilities.

Generation modules still need a complete system

The department's system explanation treats stacks, fuel processing and power conditioning separately. Fuel impurities can affect efficiency and service life. Stacks produce direct current; equipment requiring alternating current also needs conversion and power-quality control.⁶ A procurement list's module rating therefore does not automatically equal the complete, stable power supply available to a data center.

Solid-oxide fuel cells' high operating temperatures support the processing of different fuels but also create requirements involving slower startup, insulation and material durability.⁷ These are characteristics of a technology category, not proof that a supplier's particular product faces identical limitations. Actual projects still require configuration checks and test results.

It follows that continuous generation and sudden load changes are distinct engineering tasks. Designers must explain how supply continues through maintenance or faults, how long backup arrangements can last and how power conditioning meets facility requirements. A description such as round-the-clock power cannot replace scrutiny of overall redundancy and switching capability.

Less grid dependence does not mean no dependence

On-site generation changes the dependencies rather than eliminating them. Fuel routes, equipment lead times, maintenance support and local construction and operating permits remain conditions to verify. If waiting for electricity becomes waiting for fuel or equipment, the anticipated time saving may shrink. These are due-diligence risks, not claims that the companies discussed have encountered these obstacles.

Emissions must also be distinguished by fuel. The department notes that pure-hydrogen fuel cells produce water and heat at the point of operation, while reforming carbon-containing fuels releases carbon dioxide.⁵⁶ Fuel-cell technology is therefore not itself a zero-carbon label. A natural-gas system cannot simply inherit pure hydrogen's point-of-use emissions description.

A fuller comparison must consider fuel production and transport and identify the alternative supply being compared. Operating emissions, supply-chain emissions and emissions savings relative to another option are three different questions. Without consistent boundaries and project data, the technology's name alone cannot establish an environmental benefit.

Reliability also needs site-level assessment, rather than only average equipment performance. If several modules share one fuel inlet, some failures could affect all of them simultaneously. More modules do not necessarily remove a common failure point. Comparisons should identify which risks are genuinely diversified and which remain concentrated, reflecting that boundary in operating design rather than using total capacity alone as a safety margin.

From expansion expectations to operating evidence

The next evidence to watch is intended capacity becoming contracted equipment, equipment becoming electricity delivered on time, and operating results supporting the original economics. For buyers, clear service responsibilities, maintenance windows and delivery conditions constrain project risk more effectively than broad market-size forecasts.

Fuel cells can be an important candidate within a data center's power mix, but their commercial value must pass tests of timing, cost and reliability together. If earlier operation creates enough value to cover the added system costs, on-site generation can deliver a real benefit. If it merely shifts an unresolved dependency elsewhere, large market numbers cannot substitute for usable electricity.

Source notes

1. 2026-10-22 · Bloom Energy · Bloom Energy Reports Record Third Quarter 2026 Financial Results and Raises Full Year 2026 Guidance. Reporting quarter ended 2026-09-30. ↩

2. 2026-10-23 · Arunima Kumar and Shivansh Tiwary / Reuters, republished by The Star · Explainer-Why AI data centers are powering a boom in fuel-cell demand. ↩

3. 2026-06-25 · Rystad Energy · Fuel cell investment by data centers set to grow tenfold, reaching $30 billion by 2030. ↩

4. 2026-04-13 · Bloom Energy · Bloom Energy and Oracle Expand Strategic Partnership to Deploy up to 2.8 GW to Accelerate AI Infrastructure Build-Out. ↩

5. Displayed date 2013-08-14; updated 2026-10-09 · US Department of Energy · Fuel Cells. ↩ ↩²

6. First published 2014-02-25; updated 2026-06-02 · US Department of Energy · Fuel Cell Systems. ↩ ↩²

7. First published 2014-02-25; updated 2026-06-02 · US Department of Energy · Types of Fuel Cells. ↩