FCEL Q2 Earnings Call Highlights AI Push, Capacity Expansion

FuelCell Energy reported Q2 fiscal 2026 revenue of $35.59 million, missing the $41 million consensus by 13.4%, while adjusted EPS loss of 53 cents was wider than the 44-cent estimate. Operating loss widened to $77.9 million from $35.8 million due largely to a $42.6 million impairment charge, though adjusted EBITDA improved to negative $17.1 million from negative $19.3 million. Management’s offsetting message was a larger AI/data-center pipeline that reached 4 GW, a new 12.5 MW modular product, and an expanded Torrington capacity plan to 500 MW with $200 million to $275 million of expected cost.

Analysis

FCEL is trying to re-rate itself from a utility-adjacent equipment story into a picks-and-shovels AI power enabler, but the market will likely treat that as an execution story until backlog converts. The key second-order dynamic is that the 12.5 MW block reframes the sales process: smaller standardized units can reduce customer decision friction and shorten the path from pilot to purchase order, but they also expose FCEL to a more commoditized procurement process where pricing and delivery reliability matter more than narrative.

The real near-term winners are not FCEL shareholders yet, but data-center developers and hyperscalers that can use FCEL as an optionality provider in grid-constrained markets. If the behind-the-meter thesis gains traction, it pressures alternative distributed generation providers and may also pull demand forward for gas turbine OEMs, microgrids, and EPCs that can offer faster time-to-power. The manufacturing expansion is a double-edged sword: it improves credibility, but any ramp before backlog is locked increases working-capital risk and raises the probability of future dilution.

The main catalyst window is 3-9 months, when investors should see whether the 4 GW pipeline turns into signed orders or fades as long-cycle diligence resets expectations. The tail risk is that AI power demand proves real, but FCEL loses share to cheaper, simpler, or faster-to-deploy solutions; in that case the stock can remain structurally cheap despite a strong end-market. Conversely, if even a modest portion of the data-center pipeline converts, the equity could squeeze sharply because the market is still anchored to balance-sheet skepticism and low-quality earnings.

Consensus appears to be underestimating how much operating leverage can work in reverse: a successful commercial ramp may not mean immediate earnings quality improvement if management has to pre-invest in capacity and inventory. The more interesting contrarian angle is that FCEL may be a better trading vehicle than a fundamental long right now — sentiment is negative, the story is improving, and any backlog print could force revisions. That said, the cleanest setup remains event-driven, not long-duration ownership.

Key Decisions for Investors

• Trade the catalyst: buy FCEL common or call spreads into the next 1-2 quarters only if management begins showing signed backlog conversion from the data-center pipeline; target 20-30% upside on a credible order print, with 15-20% downside if conversion stalls.
• Pair trade: long FCEL / short an incumbent distributed power or fuel-cell peer with less AI exposure over the next 3-6 months; the trade works if AI power narrative drives multiple expansion faster than fundamentals, but cap the short if FCEL fails to convert pipeline.
• Avoid adding to FCEL on the manufacturing expansion alone; wait for backlog coverage metrics or customer name disclosure because pre-building capacity before contracts is the most likely dilution trigger over the next 12 months.
• Use FCEL as a volatility event trade: purchase short-dated calls only around quarterly updates or investor days where management can validate the 12.5 MW block adoption; risk/reward is favorable only when implied vol is elevated but order commentary is underappreciated.
• Monitor for a short squeeze setup if revenue guidance and backlog both improve materially; if that happens, cover any tactical shorts quickly because a thin-float name with a data-center AI narrative can rerate violently on incremental evidence.