Oracle Reaffirmation Bought Fuel Cells One Day; Monday Already Erased Half of It

On Thursday evening Sept. 24, Oracle Corp sent a force majeure notice to STACK Infrastructure, the Blue Owl Capital unit developing the 1,400-acre Project Jupiter AI campus in Doña Ana County, New Mexico, after state regulators rejected a natural-gas pipeline extension and pushed the in-service date from August 2026 to February 2027. The filing lets Oracle Corp defer payments to the developer if the campus misses its 2028 operational target — financial commitments from Oracle Corp and Blue Owl were unchanged, but the headline was toxic for the fuel-cell complex. The contract at issue is up to 2.45 GW of Bloom Energy Corp fuel cell capacity, the largest single deployment in the company's history, and any slip in Oracle Corp's build-out ripples through the AI-power thesis.

Bloom Energy Corp had dropped into a 5.4% premarket hole on Friday before posting on X that it had spoken with Oracle Corp and that the company \"remains committed to Project Jupiter and its contract with Bloom to deliver 2.4 GW of fuel cell capacity.\" Both Oracle Corp and Blue Owl said the project remains on its planned schedule. The stock reversed sharply: intraday, Bloom Energy Corp touched a roughly 13% gain at 2:55 p.m. PDT per Trefis, then settled to an 8.27% close at $288.70. Plug Power tagged along, closing up 6.45% at $1.98, while FuelCell Energy — which has its own smaller data center deal — barely moved, up 0.61% at $16.49.

Friday's bounce came against the backdrop of a multi-month unwinding that was unusually broad for a sector that had become a clean-trade proxy. From its June 25 high of $351.28, Bloom Energy Corp lost more than 47% by July 24, when Hunterbrook published a short report alleging that Bloom Energy Corp understated its dependence on Chinese scandium oxide — a rare earth that sits at the core of each solid-oxide fuel cell. Bloom Energy Corp categorically rejected the claim in an 8-K filed July 9, telling the SEC it has \"sufficient supply of scandium oxide to meet our current fuel cell demand and backlog, and our supply is not dependent on China,\" but the stock needed six weeks and a clean Q2 print to claw back half the loss.

The macro backdrop compounded the company-specific wounds. Treasury yields hit multi-year highs in August on oil inflation, dragging the entire AI-infrastructure complex lower; on September 14 a coalition of leading AI CEOs called publicly for a slowdown in the development of advanced models, hitting the tech tape broadly. Even with the Friday bounce, Bloom Energy Corp closed the week at $288.70 — still 18% below its June high — and is now down roughly 8% since the index inclusion on Sept. 21, because the mechanical buying from S&P 500 passive flows has already passed.

The sector prints in unison on the green-trade tape, but the underlying businesses have almost nothing in common. Bloom Energy Corp is on operating leverage: Q2 product revenue of $935.4M grew 215.4% year-over-year, gross margin expanded 668 basis points to 33.4%, and management has now raised 2026 guidance twice since the start of the year, from an initial $3.1B–$3.3B to the current $3.9B–$4.2B. FuelCell Energy, by contrast, just signed its first data-center Capacity Reservation Agreement — a 75 MW project in Texas backed by an upfront reservation payment — but the existing module business shrank enough to drag fiscal Q3 revenue down 29%, and the gross loss widened to -$24.5M on roughly $17M of charges tied to the Fit Energy option agreement. Plug Power is somewhere in between: a Q2 revenue beat, gross margin reset to roughly break-even for the first time, and net cash usage of $61M (down 58% from Q1) — but the share count has expanded 131% over three years to fund the turnaround.

For Bloom Energy Corp, the supply chain is a single mineral at the input and a single hyperscaler at the output. The Hunterbrook short report in July zeroed in on scandium oxide sourcing — Bloom Energy Corp rebutted it point by point in a July 9 8-K and the Q2 earnings call, and shares recovered; but the underlying dependency is now a litigated fact in a securities class-action filing. On the demand side, the Oracle Corp contract is doing the heavy lifting: Oracle Corp alone is contracted for up to 2.45 GW of Bloom Energy Corp capacity, and the company's own filings show Oracle Corp took a stock warrant as part of the April 2026 expansion, tying the customer relationship tighter. Brookfield Asset Management's $25B financing framework (up from $5B at the October 2025 launch) is what makes the buildout credit-financeable — without it, Bloom Energy Corp would need to carry the capex itself or watch the backlog slip.

FuelCell Energy and Plug Power sit further down the same value chain but on weaker economics. FuelCell Energy's Fit Energy agreement adds an option for up to 380 MW of fuel cell systems and now accounts for most of the $3.65B \"committed plus awarded\" backlog the company disclosed Sept. 2 — a backlog that is more than double FuelCell Energy's $1.29B market cap, but is contingent on FuelCell Energy actually shipping at scale. Plug Power's supply chain is a hydrogen-production network spanning material handling, stationary power, and electrolyzers; its Airbus and Edison Motors partnerships are long-dated options, not near-term revenue. The shared exposure across all three is AI data center capex: if hyperscalers slow, the multi-gigawatt backlog narrative compresses fast.

Friday's move had the shape of a relief bounce, not a sector inflection. The Monday Bloom Energy Corp drop of 7.8% — with the index demand from the Sept. 21 S&P 500 addition exhausted and 26 insider open-market sales against zero insider buys over the prior six months — argues that holders are using any strength to de-risk. None of the three businesses turned in a Friday session that resolved the open questions: Oracle Corp's gas-pipeline delay still has to be re-routed or renegotiated, Bloom Energy Corp's scandium supply story still depends on a multi-source contracting strategy that has yet to be disclosed line-by-line, and FuelCell Energy's Q3 revenue miss shows the existing module business is contracting even as the data-center pipeline grows.

The cleanest read for an investor is to separate the three trades. Bloom Energy Corp is an execution story priced for near-perfection — 25x sales, 337 trailing P/E, an S&P 500 tailwind that has now passed — where any slip in the Oracle Corp schedule or the scandium story hits the multiple directly. FuelCell Energy is a pre-revenue data-center option priced at a fraction of its backlog, but the operating business is shrinking and gross margin is still deeply negative. Plug Power is the cheapest on forward earnings (13x) and has the cleanest turnaround math, but the share count has nearly tripled in three years to fund that turn. Friday bought all three a session; Monday's tape showed which one the market is willing to underwrite at scale.