FuelCell Energy: The AI Data Center Pipeline Is Real. The Unit Economics Aren't.

FuelCell Energy's stock spent the past year being turned into an AI trade. Over roughly twelve months it ran from the low single digits to a $37.88 high — still up about 97% year to date after giving much of that back — on the story that its molten carbonate fuel cells, which burn natural gas electrochemically, could drop on-site power onto a data center campus and skip the five-to-seven-year interconnection queues throttling AI buildout. The pitch has engineering legs: no combustion means no air permits, and a standardized 12.5-megawatt "Energy Block" turns deployment into a repeatable product instead of a bespoke project.

Then, on September 2, the fiscal third quarter (ended July 31) forced investors to weigh that momentum against what the quarter actually produced. Revenue fell 29% to $33 million. Gross loss widened to $24.5 million. The stock fell about 10.7% the day of the report and has kept sliding since.

The reason wasn't just the headline numbers. It was a $17 million charge buried in gross margin, tied to Fit Energy — the first big data-center customer, signed for up to 380 megawatts. Management's own accounting recorded the charge because product costs and manufacturing overhead exceeded the contractual price the customer agreed to pay; the company's explanation was that it was operating below the volume required for cost alignment. In plain English: on the order leading its AI pivot, FuelCell loses money on every megawatt it ships because it doesn't yet build enough for its unit cost to come down to what the contract pays.

On its lead AI order, unit cost is above contract price, and the company's own filing admits it. That is the whole trade in one number.

The momentum is real as intent, not as business

FuelCell is not fabricating customer interest. In the second quarter it reported its sales pipeline had jumped 267% to 4 gigawatts, driven largely by data centers. It has since signed its first "capacity reservation agreement" with a major data center operator for a planned 75-megawatt project in Texas, with an upfront payment held ahead of a final contract, and it added $2.4 billion of "awarded capacity" from Fit Energy options. Combined committed and awarded backlog hit $3.6 billion.

But pipeline is not backlog, and backlog is not revenue. The pipeline, by the company's own definition, is "ongoing commercial discussions," not signed agreements. Nearly all of the Fit Energy "awarded" figure — the 350 megawatts of options out of the 380 total — is exercisable at Fit Energy's sole discretion and is not a firm order. The Texas agreement is a reservation, not a definitive purchase contract.

Watch what happened to the number that actually means contracted revenue while the pipeline inflated. In the second quarter, committed backlog shrank — down about 10% year over year to $1.14 billion — even as the 4-gigawatt pipeline appeared. Momentum measured in discussions went up; momentum measured in signed, payable business went down. The third-quarter backlog rebound to $1.3 billion came from booking Fit Energy's small firm tranche — the 30-megawatt Phase 0, a fraction of the headline deal. The intent is real; the contracted base barely moved.

The funding model is the tell

This is where economics beat specs, and the specs are not kind. FuelCell is expanding its Torrington, Connecticut plant to an annualized 500 megawatts — a serious, costly bet, estimated at $200 to $275 million and scheduled for completion in June 2028 — to stand ready for a pipeline it has not converted. That capacity is being paid for with newly issued stock.

The cash is there: $737.3 million as of July 31, roughly double the $341.8 million on hand at the start of the fiscal year. But it was built on dilution, not operations. Weighted-average shares outstanding roughly doubled over the past year, alongside at-the-market stock sales during and around the quarter — about $100 million in the second quarter and another $53 million right after — which is why a company with deeply negative margins is sitting on a growing cash pile.

The AI pivot is being funded by selling the very shares the AI narrative made expensive.

The gap FuelCell has to close is the one every prior quarter widened. Management targets positive adjusted EBITDA in the fourth quarter of fiscal 2027 — more than a year out — and says the path runs through an annualized production rate of 100 megawatts this October, with Torrington's cost structure aligning at that volume. The third-quarter charge is the observable test of that claim: at current volume, cost sits above price, and the company said so. The entire bullish case is a bet that scaling toward 100 megawatts bends the unit-cost curve down fast enough to make the Fit Energy price — and every data center price that follows — profitable. That is a hypothesis, not a result. Every revenue quarter so far has pushed gross margin further negative, not toward breakeven.

The stock spent a year being priced on the pipeline. The September report was the first forcing of that read back onto the economics, and at roughly 2.5x its level twelve months ago there is room for it to keep repricing until cost actually follows volume. What would prove the thesis: the October production milestone hit on schedule, Fit Energy converting options into firm orders, the Texas reservation maturing into a paid contract, and — the one that matters — gross margin turning positive as output climbs. Until that last number turns, the "AI data center momentum" is a headline describing intent, and the company funding it ships at a loss and pays for growth with shares. Watch the unit cost, not the press release.