Can Bloom Energy build them? - by Ankit Gordhandas
The Load Growth
SubscribeSign in
Can Bloom Energy build them?<br>Hyperscalers will pay for fuel cells. The question is whether one ceramic factory in Fremont can scale from 1 GW to 4
Ankit Gordhandas<br>May 15, 2026
Share
Bloom Energy's backlog went from modest to absurd in 90 days. Oracle signed for 2.8 GW. AEP committed $2.65 billion for 1 GW. Brookfield put up a $5 billion financing framework. By the end of Q1 2026, the company reported a total backlog of roughly $20 billion, up 65% from a year earlier1. It had shipped about 1.5 GW in its entire 23-year history. The contracts on the books call for multiples of that.<br>The demand question is settled. Hyperscalers will pay a premium for fuel cells because the alternative is waiting five to seven years in the interconnection queue. That math was the subject of Piece 1. This piece is about the other side of the equation.<br>Can Bloom actually build them?
Thanks for reading The Load Growth! Subscribe for free to receive new posts and support my work.
Subscribe
That is the Bloom question. It is also Leopold Aschenbrenner’s largest bet. His position was worth roughly $2.2 billion after the Oracle deal. The stock has since roughly doubled again, riding the backlog higher. If Bloom delivers, the position keeps compounding. If it cannot scale manufacturing from 1 GW per year to the 4+ GW per year its contracts imply, the company is a $65 billion market cap built on a promise.<br>The answer is not obvious. Scaling a ceramic electrochemistry business by 4x is not like scaling a software product, and it is not like scaling a gas turbine factory. It requires a supply chain that barely exists at the volume needed, a rare metal that mostly comes from China, and field reliability data at a scale no one has ever tested.<br>Here is what we know.<br>What’s inside the box
Bloom energy server in Northeast US (courtesy: Nelson + Pope)<br>A Bloom Energy Server is a stack of solid oxide fuel cells. Each cell is a thin ceramic plate, roughly 100 by 100 millimeters, made primarily of scandia-stabilized zirconia (ScCeSZ). Scandium oxide, or scandia, is doped into the zirconia crystal structure because it provides higher ionic conductivity at operating temperatures than the more common yttria-stabilized alternative2. That conductivity is the entire mechanism.<br>Natural gas flows over one side of the plate. Air flows over the other. At roughly 1,000 degrees Celsius, oxygen molecules on the air side split into ions, migrate through the ceramic, and react with the methane on the fuel side. The reaction produces electricity, water, and carbon dioxide directly, without combustion. All without a flame, turbine, or spinning generator.
The result is 53 to 65% electrical efficiency, depending on conditions, roughly 50% better than a simple-cycle gas turbine and competitive with combined-cycle plants3. Capture the waste heat and total efficiency exceeds 90%. Because nothing burns, NOx and SOx emissions are near zero.<br>Each cell produces a small voltage. Bloom stacks cells into columns, coats them with a nickel-based anode on one side and a lanthanum strontium manganite cathode on the other, then assembles the columns into a “hot box” packaged inside a unit called an Energy Server. The current model, the Server 6.5, is rated at 300 kW. A 10 MW installation is 33 servers. A 100 MW installation is 333 servers. A gigawatt is 3,333 servers. The modularity is the point: you do not build a gigawatt plant, you deploy 3,333 identical units.<br>That sounds simple, but it’s not.<br>The ceramic problem
Every Bloom fuel cell starts as a powder. Scandia-stabilized zirconia is milled to a precise particle size, mixed with binders, tape-cast into thin sheets, and fired in a kiln at temperatures above 1,300 degrees Celsius. The firing process is where the physics gets unforgiving. Ceramic sintering is sensitive to temperature gradients, atmosphere composition, heating rate, and impurities at the parts-per-million level. A batch that fires 20 degrees too hot warps. A batch with slightly wrong humidity cracks. Quality control is not statistical sampling at the end of the line. It is control of every variable at every step.<br>Then each plate gets coated with electrode materials on both sides. These are applied as slurries, dried, and fired again. The interfaces between layers determine how efficiently ions transfer, which determines how much power the cell produces, which determines the economics of the entire system.<br>The cells go into hot boxes, the sealed high-temperature modules at the core of each Energy Server. Bloom does not make all of these in-house. MTAR Technologies, based in Hyderabad, India, manufactures the hot box assemblies and holds an estimated 60 to 70% wallet share of that critical component. MTAR is scaling from 8,000 to 12,000 hot box units per year. A single supplier in a different country making the core component of a product with a $20 billion backlog is a...