Bloom Energy Corporation BE

Key data

The business

Bloom makes solid-oxide fuel cells, boxes it calls Energy Servers that sit on a customer's site and turn natural gas, biogas or hydrogen into electricity through a chemical reaction rather than combustion. The customer gets power in about 90 days, against the multi-year wait to hook a large new load onto the grid. That speed is the whole pitch in 2026: AI data centers need hundreds of megawatts now, utilities can't deliver them for years, and Bloom fills the gap. Historically the biggest customer was Korea's SK ecoplant; the growth now comes from US data center developers and their power partners. The money is made twice, once selling the box, then again on a long-term contract to service the fleet, which is the recurring, higher-quality half. The moat is that installed base and those service agreements: once a customer standardizes on Bloom's stack and signs a decade of service, switching means ripping out working power. A short-seller report earlier this year questioned the economics; management answered it in the 10-Q and the stock kept climbing.

The numbers

The numbers say one thing: this company just crossed from perpetual cash-burn into profit.

Read the operating-income column: red through mid-2025, positive from Q3, then $72M in a single quarter, Q1 2026, which nearly matched all of 2025's operating profit. Gross margin held near 30% while revenue doubled quarter-over-quarter at the run-rate level (Q1 2026 vs Q1 2025), the signal that scale is not eroding the unit economics.

The annual view shows the same turn, slower: operating income went positive in 2024, yet the bottom line stayed red, a $0.37 loss per share in 2025, because $2.6 billion of debt and its interest sit between operating profit and net profit. In late 2025 Bloom raised roughly $1.5 billion, taking debt from $1.1B to $2.6B and cash from $0.8B to $2.5B, a war chest for the data-center build-out that also enlarged the interest bill.

The latest print, Q2 2026 reported July 28, showed $0.78 of earnings against a $0.39 estimate and $199 million of net income; this pack does not isolate the quarter's revenue line. Treat that $199M with care: it is nearly triple the best operating quarter on record, so much of it is a one-time, non-operating gain. The cleaner story is the beats, four straight and each roughly double the estimate, which means either the Street can't model the ramp or management guides low. Consensus has revenue reaching $9.9 billion by 2028, nearly five times 2025. the market credits the revenue ramp but not the margins under it, and the print that settles it is gross margin holding above 28% as volume triples.

Management

The record is one-sided: zero insider buys in twelve months against $160 million sold across 51 transactions. CEO KR Sridhar sold $34 million in February 2026, chairman John Chambers $16 million in May, executive Shawn Soderberg $10 million in April; plan status is not disclosed in the filings on hand, so planned and discretionary can't be split, and Sridhar's is a large single sale. Selling after a five-fold run is human, but nobody bought a share. Pay leans on stock, and dilution is real, with share-based compensation running about 2.7% of revenue. The guidance record is the bright mark: four consecutive quarters cleared by wide margins.

How it fails or surprises you

AI power scarcity keeps compounding (right tail). If interconnect queues stay multi-year and one or two more hyperscaler-scale deals land, the $9.9B 2028 estimate is a floor, not a target. The market pays for the ramp it can see, not for deals unsigned. First tell: a step-up in take-or-pay backlog in the next two 10-Qs.

Customer concentration and the capex cycle. Revenue is lumpy and leans on a handful of large buyers whose spending can turn fast. A slipped or renegotiated data-center contract shows up as a hard sequential revenue drop. First tell: backlog converting slower than the 2026 production forecast management says it can meet.

The profit is thinner than the headline. FY2025 was still a loss despite positive operating income, and Q2 2026's $199M net income looks padded by a one-time gain. This was the open question days ago and it stays open: strip non-operating items and the real run-rate profit is a fraction of what a 32x multiple implies. First tell: the operating-income-to-net-income bridge in the Q3 2026 10-Q.

Closing thoughts

This is an uncertainty a named print resolves, sitting on a valuation that already prices success. At 32 times a 2028 number and 24 times trailing sales, the demand story is largely paid for; what isn't paid for, either way, is whether the margins and the profits are durable. The fatter tail is arguably the upside on revenue, because the power scarcity is real and Bloom's speed advantage is real, but the risk that matters is not demand, it is execution, earnings quality, and $2.6 billion of debt if the ramp ever stalls. Break the concentration linchpin and a high multiple de-rates violently; land the right tail and today's estimates prove far too low.

The bet is still that America can't wire its data centers fast enough, so they buy Bloom's boxes, and that the boxes earn money running, not just selling. It breaks if a big customer's spending turns, or if the profit turns out to be one-time. Watch two numbers: quarterly gross margin staying above 28%, and net income that traces back to operations rather than to gains. If margin cracks while backlog stalls, the story is wrong no matter what the revenue headline says.