The number that started the argument is not a revenue forecast. On Oct 23, 2026, Reuters reported that Rystad Energy expects data-center fuel cell investment to reach about $30 billion by 2030 from roughly $2.8 billion in 2025, and that the visible order book already stands near 9 GW.
One day earlier, Bloom Energy reported the quarter that forecast rests on: revenue of $1,235.8 million, up 138.1% year over year, a 33.8% gross margin, and full-year guidance raised to $4.4 billion–$4.6 billion. The stock fell 11.24% to $193.30 that day. The reason sits in the cash flow statement, not the income statement.
Data-center fuel cell spend, 2030
~$30B
Rystad Energy forecast, up from ~$2.8B in 2025; reported Oct 23, 2026
Cumulative fuel cell demand, 2026–2030
10.4 GW
Rystad Energy research published Jun 25, 2026
Visible order book
~9 GW
Rystad Energy, June 2026 — mostly framework agreements, not firm orders
Bloom Energy Q3 2026 operating cash flow
-$27.1M
Quarter ended Sep 30, 2026; versus +$226.5M in Q2 2026
Two customers' share of Bloom Energy Q3 revenue
43% + 26%
Form 10-Q for the quarter ended Sep 30, 2026;97% of revenue from the U.S.
The Forecast
A $30 Billion Market Where Most of the Book Is Still Optional
Rystad Energy's research, published in June 2026 and reported again on Oct 23, 2026, is the anchor: roughly 10.4 GW of cumulative data-center fuel cell demand between 2026 and 2030, market revenue rising from about $2.8 billion in 2025 to roughly $30 billion in 2030, and about 40% of projected U.S. data-center capacity in 2030 modeled as pursuing dedicated on-site power rather than a grid connection.
| Metric | Value | Basis |
|---|---|---|
| Data-center fuel cell revenue, 2025 | ~$2.8B | Rystad Energy, June 2026 |
| Data-center fuel cell revenue, 2030 | ~$30B | Rystad Energy, June 2026 |
| Cumulative demand, 2026–2030 | 10.4 GW | Rystad Energy, June 2026 |
| Visible order book | ~9 GW, framework-weighted | Rystad Energy, June 2026 |
| Order book as a share of five-year demand | ~87% | Calculated from Rystad figures |
| Implied value per MW installed | ~$2.9M | Calculated from Rystad figures |
| Annual industry manufacturing capacity | ~1.8 GW now, ~4 GW by 2030 | Rystad Energy, June 2026 |
| Share of cumulative stationary deliveries from solid-oxide | ~53% | Rystad Energy, June 2026 |
Two lines in that table decide the investment case. The order book is a menu, not a purchase order: the ~9 GW includes framework agreements with Oracle, American Electric Power, Equinix and Brookfield, and Rystad itself cautions that framework agreements do not necessarily mean firm orders for the full stated capacity. The second line is the harder constraint. Industry output has to roughly double from about 1.8 GW a year to about 4 GW by 2030, which caps the whole industry near 4 GW a year against 10.4 GW of projected demand.
What Operators Buy
Operators Are Buying a Schedule, Not a Megawatt
The hardware is not cheap. Dividing Rystad's $30 billion of 2030 revenue by 10.4 GW of cumulative demand implies roughly $2.9 million per installed MW — multiples of gas turbine cost. Buyers pay it because of the calendar, not the kilowatt. In April 2026, Oracle signed a master services agreement for up to 2.8 GW of Bloom Energy fuel cells, with 1.2 GW already contracted. Last year Bloom Energy delivered a working system in 55 days against a 90-day plan. In May 2026, Nebius Group said fuel cells will let its first U.S. campus — 328 MW in Independence, Missouri — operate in 2026 and remove the reciprocating engines planned for the site.
- Permitting: fuel-cell systems face a lighter approval burden than combustion generation, which shortens the run from site selection to energized capacity (Nebius Group, May 20, 2026).
- Grid independence: behind-the-meter supply removes transmission build and interconnection queue time from the critical path (Bloom Energy and Nebius Group, May 2026).
- Fuel flexibility: Bloom Energy's solid-oxide platform can run on natural gas supplied on site or 100% pure hydrogen, which removes the hydrogen plant from the critical path; FuelCell Energy's molten carbonate design is capped at 40% hydrogen blends (Rystad Energy).
- Emissions: fuel cells avoid combustion inside the stack and carry lower local emissions than conventional gas-fired generation (Rystad Energy).
- Scale: the marginal megawatt an AI campus needs in 2027–2029 is the one a turbine cannot deliver on schedule (Rystad Energy; GE Vernova Q2 2026).
The Displaced Layer
Fuel Cells Are Not Taking Turbine Orders — They Are Taking the Megawatts Turbines Cannot Deliver
It is tempting to read the fuel cell boom as an attack on the gas turbine order book. The numbers say otherwise. GE Vernova closed the second quarter of 2026 with 116 GW of gas power equipment backlog and slot reservations, up from 100 GW one quarter earlier, and told investors it expects at least 125 GW by year-end 2026. It shipped 3 GW of equipment in the quarter. At that run rate, the queue on GE Vernova's books is worth roughly a decade of shipments — and the fuel cell industry, at 1.8 GW a year today, is a rounding error against it.
Gas turbine output dwarfs fuel cell manufacturing capacity
GE Vernova shipped 3 GW of gas equipment in Q2 2026, annualized here for comparability. Rystad Energy puts fuel cell industry capacity at about 1.8 GW a year today, rising to about 4 GW by 2030; Bloom Energy plans to lift factory capacity from 1 GW to 2 GW by the end of 2026.
Unit: GW per year
GE Vernova gas equipment shipped, Q2 2026 annualized
3 GW shipped in the quarter
12
Fuel cell industry capacity, 2030
Rystad Energy projection
4
Bloom Energy factory capacity target, end-2026
Company plan in the FY2025 Form 10-K
2
Fuel cell industry capacity, 2026
Rystad Energy estimate
1.8
| Layer | What the evidence shows | Read-through |
|---|---|---|
| Gas turbine builders | GE Vernova Q2 2026: 116 GW of backlog plus slot reservations, 3 GW shipped, at least 125 GW targeted by year-end 2026 | Not displaced; fuel cells are too small to dent the queue |
| Reciprocating gensets | Nebius Group, May 2026: fuel cells removed the planned reciprocating engines at the 328 MW Independence campus | Displaced directly at the primary-load layer |
| Grid interconnection | Bloom Energy and Nebius Group: behind-the-meter supply removes transmission build and queue time | Bypassed by choice, at a hardware cost premium |
| On-site utilities | American Electric Power ordered 100 MW with a path to 1 GW to serve AI data centers inside its own territory (Nov 14, 2024) | Utilities resell on-site power rather than lose the load |
The genset layer is not standing still. Caterpillar grew second-quarter 2026 revenue to $20,543 million from $16,569 million a year earlier and earned $4,295 million of operating income; Cummins grew second-quarter revenue to $9,457 million and raised its 2026 revenue outlook to 10%–13% growth. Fuel cells take the primary-load slice and leave the backup slice alone — which is exactly what the Nebius Group procurement decision shows.
Where the Margin Goes
One Vendor Earns the Margin; Two Announce and Burn Cash
Scale is the whole margin story in this chain, and it is not close. Over the trailing twelve months, Bloom Energy ran a 32.3% gross margin on $3,829.9 million of revenue. Plug Power ran a -18.5% gross margin on $744.1 million of revenue. FuelCell Energy ran a -32.4% gross margin on $154.1 million of revenue, having announced a 380 MW data-center agreement with Fit Energy in June 2026. Announcements and margin are not the same thing.
Trailing-twelve-month gross margin across three fuel cell vendors
Trailing twelve months to each company's most recent reported quarter: Bloom Energy to Sep 30, 2026, FuelCell Energy to Jul 31, 2026, Plug Power to Jun 30, 2026.
Unit: % of revenue
$1,238.1M gross profit on $3,829.9M revenue
32.3%
-$138.0M gross profit on $744.1M revenue
-18.5%
-$49.9M gross profit on $154.1M revenue
-32.4%
| Vendor | Trailing revenue | Gross margin | Net income | Output scale |
|---|---|---|---|---|
| Bloom Energy | $3,829.9M | 32.3% | $518.3M | 2 GW/yr factory target by end-2026 |
| Plug Power | $744.1M | -18.5% | -$1,641.4M | Building out PEM; $132.5M deal with Stream Data Centers (Feb 26, 2026) |
| FuelCell Energy | $154.1M | -32.4% | -$175.6M | 37.1 MW annualized rate; 500 MW at Torrington by June 2028 |
The manufacturing gap explains the margin gap. FuelCell Energy operated at an annualized production rate of about 37.1 MW in its September quarter, targeting 100 MW in October 2026, while Bloom Energy plans to run 2 GW — a gap of about 54 to one. The newest entrant to the chain is not American. Doosan Fuel Cell signed an approximately 501.4 billion won contract with HyAxiom to supply phosphoric acid fuel cell systems to U.S. AI data centers, phased from the second half of 2026 into the first half of 2028, off roughly 275 MW of annual production capacity, and in August 2026 booked a roughly 108.7 billion won solid-oxide stack export to Reverion in Germany.
The Cash Tell
The Reason the Stock Fell Is in the Cash Flow Statement
The income statement says the fuel cell business works. The cash flow statement says it does not work yet. Bloom Energy earned $225.4 million of operating income and $250.3 million of net income in the third quarter of 2026 and still burned $27.1 million of operating cash, because working capital consumed $356.0 million — receivables up $175.2 million, inventory up $170.4 million — on top of $153.5 million of capital spending. Free cash flow was negative $180.6 million.
| Line item | Sep 30, 2026 | Dec 31, 2025 | What it says |
|---|---|---|---|
| Operating cash flow, quarter | -$27.1M | +$226.5M (Q2 2026) | Growth consumes cash before it returns it |
| Free cash flow, quarter | -$180.6M | +$174.8M (Q2 2026) | Capex is accelerating with the ramp |
| Contract assets | $858.6M | $241.2M | A $617.5M build of delivered, unbilled work |
| Accounts receivable | $634.7M | $371.8M | Customers taking longer to pay |
| Customer deposits | $661.4M | $78.2M | Buyers are pre-funding the equipment |
| Unsatisfied performance obligations | $661.2M | $394.4M | The only firm, contracted backlog figure in the filing |
| Total assets | $6,571.6M | $5,628.4M (Jun 30, 2026) | $943.2M added to the balance sheet in one quarter |
- The firm number is small next to the headline. The 10-Q discloses $661.2 million of unsatisfied performance obligations, against a market order book being quoted in gigawatts — and management did not quantify backlog on the third-quarter call.
- Concentration is extreme: two customers, one of them a related party, took about 43% and 26% of third-quarter revenue, and 97% of revenue came from the United States.
- The customer of last resort was equity. To secure the Oracle agreement, Bloom Energy issued a warrant over up to 3,531,073 shares struck at $113.28, valued at $251.6 million on the April 9, 2026 grant date, accounted for as consideration payable to a customer's customer, which reduces reported revenue as systems ship — $17.3 million of revenue reduction recognized through Sep 30, 2026.
- The balance sheet is being funded to match. Bloom Energy raised $1,439.9 million of net debt over the trailing twelve months and ended the third quarter with $2,486.7 million of cash and $295.3 million of net debt.
The Chain
Who Gets Paid, Layer by Layer
- Stacks and modules: Bloom Energy builds and ships its own solid-oxide platform; FuelCell Energy is scaling Torrington to 500 MW by June 2028 with a 30 MW committed delivery and warrants tied to up to 380 MW under the Fit Energy agreement; Doosan Fuel Cell is exporting phosphoric acid systems and solid-oxide stacks into the U.S. on-site build.
- Balance of plant: Vertiv sells the power conversion, switchgear and thermal kit every fuel cell farm needs, and has already integrated fuel cell modules into its UPS line with Ballard — its Power Module H2 pairs two 200 kW Ballard cabinets with a Liebert EXL S1 UPS and hydrogen storage in a 1 MW microgrid. In March 2026, Vertiv and Generate Capital launched a bring-your-own-power-and-cooling model for U.S. data centers.
- Fuel: American Electric Power is buying the equipment to serve AI load inside its own territory, while Bloom Energy's solid-oxide design runs on natural gas supplied on site and is ready for 100% pure hydrogen.
- Financing: Brookfield took the vehicle from a $5 billion framework in October 2025 to $25 billion in June 2026, and Bloom Energy disclosed a $1.7 billion project investment supporting the Nebius Group deployment — the capital intensity is being moved off the operator's balance sheet and onto Bloom Energy's own, which still committed $71.6 million to those joint ventures at Sep 30, 2026.
- Demand owners: Oracle holds up to 2.8 GW, Nebius Group 328 MW at its first U.S. campus, Equinix more than 100 MW across 19 IBX sites with roughly 75 MW already operating, and American Electric Power up to 1 GW — together the framework agreements Rystad counts in the roughly 9 GW order book.
- Ceramics: Kyocera sells solid-oxide stacks and the fine ceramics that go into them. The ceramic layer captures volume without the premium, and for a group of Kyocera's scale the fuel cell line is immaterial to earnings.
Two concentrations matter more than the layer map. Rystad's assessment is that Bloom Energy holds virtually every primary-load solid-oxide contract in the visible order book, so an execution slip at one vendor is an industry slip. And the customer list is short: Oracle, Equinix, Nebius Group, American Electric Power and Brookfield account for the framework agreements that make up the market's headline number, and Bloom Energy's revenue is 97% American.
Horizons
Two Quarters to Prove It, Three Years to Scale It
- Vertiv reports third-quarter results on Oct 28, 2026; the on-site power and cooling line is the cleanest read on whether behind-the-meter generation is moving from pilot to standard specification.
- GE Vernova reports third-quarter results on Oct 28, 2026; the gas power backlog and slot reservation figure, and whether the at-least-125 GW year-end target is met, sets the price of the megawatts fuel cells compete against.
- Bloom Energy must show whether operating cash flow turns positive as customer deposits convert into revenue; two more quarters of negative cash with contract assets still climbing would confirm the working-capital trap.
- FuelCell Energy must convert its 75 MW Texas capacity reservation, with an upfront reservation payment already made, into recognized revenue on a business running a -32.4% trailing gross margin.
- Doosan Fuel Cell begins phased U.S. deliveries in the second half of 2026 — the first hard test of whether a non-American stack maker can hold a U.S. on-site power schedule.
- Industry output has to roughly double from about 1.8 GW a year to about 4 GW by 2030 to serve 10.4 GW of demand; the binding constraint is stack and balance-of-plant manufacturing, not orders.
- The ~9 GW framework order book has to convert into firm performance obligations. Bloom Energy's contracted figure was $661.2 million at Sep 30, 2026, and the gap between that and the gigawatt headlines is the single most important number to watch.
- FuelCell Energy's Torrington expansion to 500 MW by June 2028 has to arrive with positive gross margin, or the second-tier fuel cell market stays a concentration story for Bloom Energy alone.
- Concentration risk: with two customers at roughly 69% of Bloom Energy's third-quarter revenue and 97% of revenue from the United States, any pause in AI capital spending hits one supplier's income statement hardest.
- Policy risk sits outside the filings: the emissions case for fuel cells rests on avoiding combustion inside the stack, and any tightening of local air permitting standards changes the permitting advantage the sales argument depends on.
Synthesis
Own the Schedule, Not the Gigawatts
The facts: Rystad Energy has put data-center fuel cell spending on a path from about $2.8 billion in 2025 to roughly $30 billion in 2030, with about 40% of projected U.S. data-center capacity going on-site. Bloom Energy has proven the demand is real at a scale nobody else has reached, turning revenue up 138% in a quarter while holding a 33.8% gross margin. GE Vernova is not losing the queue; it is 116 GW deep in it and shipping 3 GW a quarter. FuelCell Energy and Plug Power have announcements and negative gross margins.
The inference: speed-to-power is a genuine moat, because the turbine queue is roughly a decade long and the fuel cell industry can only build about 1.8 GW a year — but a moat that requires constant working capital is a fragile one. The speculation worth pricing: if fuel cells become standard rather than exceptional in hyperscaler procurement, the ~$2.9 million per MW premium survives the walk down from scarcity.
For investors, the trade is not fuel cells in the abstract. It is the vendor that converts time-to-power into cash-generative megawatts — and the balance-of-plant and financing layer that gets paid whichever vendor wins.
Power remains a key constraint for AI infrastructure build-outs. We chose Bloom because their fuel cells solve that directly.
The investable chain
- Q3 2026 revenue of $1,235.8 million grew 138.1% at a 33.8% gross margin, proving the order book converts to hardware.
- Operating cash flow of -$27.1 million shows growth funded by deposits and debt, not earnings.
- Two customers took roughly 69% of Q3 revenue and 97% of revenue was U.S., leaving no cushion for an AI capex pause.
- Watch the fourth-quarter 2026 report: positive operating cash flow is the catalyst for a valuation near 133 times trailing earnings.
- 116 GW of gas backlog and slot reservations at Q2 2026 with only 3 GW shipped leaves the queue roughly a decade long.
- Fuel cells add about 1.8 GW a year industry-wide, so the incumbent captures the megawatts fuel cells cannot build in time.
- Third-quarter results on Oct 28, 2026 and the at-least-125 GW year-end target are the near-term checks on that pricing power.
- Longer term, developing its own fuel cell technology hedges the on-site layer it does not currently own.
- Second-quarter 2026 revenue of $3,274.3 million at a 37.7% gross margin rides the same build without fuel cell execution risk.
- Every fuel cell farm still needs switchgear, power conversion and thermal, so Vertiv sells the balance of plant behind every on-site megawatt.
- Third-quarter results on Oct 28, 2026 are the cleanest read on whether behind-the-meter power has become standard specification.
- Longer term, the bring-your-own-power-and-cooling model with Generate Capital deepens its role in financing on-site generation.
- A 380 MW Fit Energy headline in June 2026 sits against $154.1 million of trailing revenue and a -32.4% gross margin.
- Committed backlog was only $1.296 billion at Jul 31, 2026, up 4.1% year over year; $2.35 billion of awarded capacity sits outside it.
- It announces gigawatts while selling dollars below cost, and a 37.1 MW production rate cannot absorb a 380 MW promise.
- The 75 MW Texas reservation with an upfront payment is the near-term test; Torrington's 500 MW by June 2028 is the long-term one.
- 328 MW of fuel cells is due online in 2026 at Independence, Missouri, with reciprocating engines removed from the site.
- Behind-the-meter power that avoids transmission build pulls cloud revenue forward months earlier, which matters more than cost per megawatt-hour.
- Trailing revenue of $1,355.1 million against a build of this size means financing partners, not operating cash, fund the campus.
- Longer term, firm on-site power becomes a sellable feature of the AI cloud rather than an infrastructure constraint.
- Up to 2.8 GW of fuel cells, with 1.2 GW contracted, is the single largest claim on the visible order book.
- The 55-day delivery record locks in the fastest path to energized AI capacity, at the cost of depending on one vendor.
- The warrant struck at $113.28 shows the price of that speed: equity rather than cash, booked as a reduction of revenue.
- Longer term, a 2.8 GW single-vendor dependency becomes a procurement risk as on-site alternatives multiply.
- Second-quarter 2026 revenue of $20,543 million, up from $16,569 million, with $4,295 million of operating income shows genset demand is not collapsing.
- Fuel cells take primary load, so the genset line keeps the backup slice while losing the primary-load install.
- Cummins expects data-center-related sales up 80% to $9 billion by 2030, and power generation revenue at Caterpillar rose 41% in the first quarter.
- Near term, order momentum drives the line; over one to three years, primary-load share is the risk.
- A 501.4 billion won HyAxiom contract puts phosphoric acid systems into U.S. AI data centers from late 2026 into 2027.
- Phased deliveries to 2028 export a stack maker into the U.S. on-site build with visibility no U.S. fuel cell peer matches.
- Annual production capacity near 275 MW exceeds FuelCell Energy's 37.1 MW run rate and trails Bloom Energy's 2 GW target.
- Watch the first U.S. delivery milestone in 2027; slippage would end the export story quickly.
