The machine that made Bloom Energy famous is a power plant disguised as a row of tidy metal cabinets. It sits beside a data center, hospital or warehouse, accepts fuel and air, and quietly makes electricity where the electricity will be used. No spinning turbine. No combustion. No long transmission line from a distant plant. Inside, ceramic squares work at furnace-like temperatures, passing oxygen ions through a solid electrolyte and releasing electrons into a circuit. Stack enough squares together and the result is the Bloom Energy Server - the “Bloom Box” that arrived on television in 2010 looking like a prop from a cheerful future.
The future took longer than the launch party implied. Fuel cells were expensive, grid electricity was usually available, and Bloom's finances attracted scrutiny. Then the American power system acquired a queue. New factories and data centers began asking utilities for amounts of electricity once associated with cities, while transformers, substations and transmission projects remained slow to permit and build. Artificial intelligence compressed the customer's timetable further. The new question was no longer simply, “What is the cheapest kilowatt-hour?” It was, “How soon can we turn the building on?”
A ceramic shortcut around the grid
Bloom's solid oxide fuel cells turn chemical energy directly into electricity. Warmed air reaches one side of a ceramic cell; fuel and steam reach the other. Oxygen ions cross the electrolyte, react with the fuel and produce electricity, water, heat and, when the fuel contains carbon, carbon dioxide. Because the reaction is electrochemical, not a flame, the system avoids the nitrogen oxides, sulfur oxides and particulate pollution associated with combustion. It also uses no water during normal operation, a meaningful detail for large loads in water-stressed places.
One Energy Server occupies roughly a parking space and supplies about 200 to 250 kilowatts. Cells become stacks, stacks become modules, and modules become installations that can run from hundreds of kilowatts to multiple megawatts. The architecture is less like commissioning one enormous power station and more like adding server racks. Customers can place generation beside demand, run in parallel with the utility, form an islandable microgrid or operate off-grid.
This modularity is Bloom's immediate advantage. A gas turbine may offer excellent economics at enormous scale, but it has a long equipment queue and substantial site work. Batteries respond instantly but store rather than create energy; without generation, they eventually empty. Diesel engines arrive quickly but bring noise, combustion pollution and fuel logistics. Bloom sits in the middle: continuous, relatively compact generation with less local pollution and a deployment schedule designed for campuses rather than utility territories.
Bloom's most valuable product may not be a fuel cell. It may be the months or years a customer does not spend waiting for power.The new time-to-power economy
Who buys a power plant in a box?
The customer list explains the product better than a chemistry lesson. Google hosted Bloom's first commercial deployment in 2008. Retailers including Walmart and Home Depot wanted predictable energy and stores that could remain useful during outages. FedEx, manufacturers and semiconductor plants value uptime because a brief interruption can spoil inventory or halt an expensive process. Hospitals and college campuses need critical loads to survive grid failures. Equinix and other data-center operators need dense, uninterrupted power.
By 2026, the data-center case had become dominant. Oracle said it intended to procure as much as 2.8 gigawatts under an expanded master agreement, with an initial 1.2 GW deploying across US projects. That is not a boutique backup installation. It is utility-scale capacity assembled beside computing infrastructure. Bloom reported $1.065 billion of revenue in the second quarter of 2026, up 165.5 percent from a year earlier, and lifted full-year guidance to $3.9 billion to $4.2 billion.
What source gives us the lowest cost of electricity over decades?
What can deliver reliable megawatts at this site before our chips and building sit idle?
Bloom also sells to utilities, distributors and financiers. That last group is essential. Customers often do not want to own fuel-cell equipment any more than an office tenant wants to own a substation. A project-finance partner can buy the system and sell its output to the end user under a power agreement. Brookfield's framework with Bloom, first announced at $5 billion and expanded in June 2026 to as much as $25 billion, is meant to put institutional capital behind qualifying AI power projects.
Hardware, service and financial plumbing
Bloom is a manufacturer, but the income statement contains four businesses: product, installation, service and electricity. Product sales remain the engine. Installation turns delivered cabinets into an operating plant. Long-term service agreements pay Bloom to maintain availability and performance, creating recurring revenue as the installed base grows. In some structures, Bloom or its partners sell electricity rather than equipment.
That variety solves the customer's financing problem, but it adds complexity. Bloom's 2025 filing showed that three customers or distributors represented roughly 43, 13 and 12 percent of revenue. Big projects can make quarterly sales lumpy, and project-finance relationships can concentrate exposure. Fuel cells also need scheduled maintenance and stack replacement. The business is therefore not a clean handoff of metal boxes at a loading dock; it is a long relationship among manufacturer, owner, fuel supplier and power user.
Energy Servers and electrolyzers, manufactured at scale and sold directly or through a financier.
Electrical, mechanical and site work that turns modular equipment into a functioning plant.
Long-term operations and maintenance tied to system availability and performance.
Power sold from systems under contractual arrangements rather than a one-time equipment purchase.
The green claim, with the gas line visible
Bloom's environmental argument requires precision. When an Energy Server runs on fossil natural gas, it emits carbon dioxide. Non-combustion does not mean carbon-free. The advantages are high electrical efficiency, negligible combustion-related air pollutants, no normal operating water use and generation close to the load, which avoids transmission losses. Those are useful improvements, especially where diesel engines or an inefficient grid are the alternatives, but they do not turn methane into a zero-carbon fuel.
The company's longer transition story rests on fuel flexibility. The same platform can use biogas or hydrogen. Because its exhaust can contain carbon dioxide at about 95 percent purity, Bloom argues that capture is simpler than separating CO2 from a turbine's nitrogen-heavy flue gas. Run on biogas with carbon capture and the system could, under the right lifecycle accounting, produce net-negative power. Those pathways depend on clean-fuel supply and carbon infrastructure that are not yet universally available.
The second product, the Bloom Electrolyzer, reverses the central trick. Electricity and steam enter a high-temperature solid oxide cell; hydrogen and oxygen emerge. Heat supplies part of the energy that lower-temperature electrolyzers must take as electricity, so the design is attractive beside nuclear plants or industrial sites with available steam. Bloom's work with Idaho National Laboratory and SK ecoplant is aimed at proving that advantage beyond the lab.
A category between the meter and the power plant
Bloom competes on several maps at once. FuelCell Energy and Doosan sell stationary fuel-cell systems. Ceres Power, Sunfire and others develop solid oxide technology or electrolysis. Caterpillar, Cummins and Wärtsilä sell conventional engines. Turbine makers serve larger projects, while batteries paired with renewables handle short-duration resilience. The incumbent alternative is simply the utility grid.
Bloom's difference is the combination: a commercially deployed solid oxide platform, modular scale, a US manufacturing base, low water use, multiple fuel pathways and financing partners willing to own infrastructure. Its engineering organization included 62 PhDs at the end of 2025, reflecting the materials-science difficulty of repeatedly heating ceramic cells to about 800°C while preserving output and life. Bloom says cell life has increased more than two and a half times since its first generation.
The company was founded in 2001 as Ion America by KR Sridhar, who had led a University of Arizona lab working on technology for NASA's Mars program. The original assignment was to use Martian gases to make oxygen and fuel. When that mission ended, Sridhar redirected the solid oxide work toward Earth. Kleiner Perkins backed the company early; more than $1 billion in venture funding followed before a $270 million IPO in 2018.
It is a wonderfully Silicon Valley origin story, though Bloom's actual work looks less like software than patient industrial iteration: ceramic inks, thermal cycling, power electronics, manufacturing yield, field crews and maintenance contracts. The culture it describes is mission-led and technically deep. In late 2025, the company granted eligible employees below director level with sufficient tenure restricted stock units intended to make them shareholders in 2026. More than 2,000 people work there worldwide.
The Bloom Box began as a cleaner way to make power. Its breakout use may be more prosaic: making power appear on schedule.Twenty-five years from Ion America to AI infrastructure
Where does Bloom fit? It is not a utility, though it builds utility-scale fleets. It is not merely a generator vendor, because service and finance shape the sale. It is not a pure hydrogen company, because natural gas powers much of today's installed base. The cleanest description is distributed energy infrastructure: controllable power plants placed behind the meter, close enough to the load that the customer can choose speed and resilience rather than wait for the wider system.
That position carries a test. AI demand can make almost any available megawatt look attractive, but Bloom must manufacture reliably, manage customer concentration, service a rapidly growing fleet and show how today's gas-fed systems lead toward lower-carbon fuels rather than simply locking in another fossil asset. The ceramic square from the Mars lab has already survived one reinvention. The next will be judged in data-center construction schedules, service margins and tons of carbon, not television applause.