Bloom Energy sells fuel cells: stationary, on site power plants that turn natural gas into electricity without combustion. That technology is now a $20 billion AI infrastructure bet, and the order book is no longer dominated by Oracle.
The first thing to know about Bloom Energy's latest quarter is that the headline number is not the most interesting one. Fuel cell product revenue nearly tripled, and the order book crossed $20 billion. But the line item that should change how investors, hyperscalers, and regulators read the AI buildout is a quieter disclosure from management: more than half of the data-center backlog is now committed by customers other than Oracle.
That matters because Bloom Energy is not a household name, and it is on the short list of companies that decide whether AI capacity actually gets built in the near term. A fuel cell is a stationary power plant that turns natural gas, or hydrogen in some configurations, into electricity through an electrochemical reaction rather than combustion. Crucially, it can be sited on the same campus as the data center it feeds, which sidesteps a different bottleneck: the U.S. transmission grid, where interconnection queues now run years long and utility-scale projects routinely miss energization dates. Bloom, headquartered in San Jose, has become the dominant merchant supplier of that on-site power technology for data centers, and the April 28 first-quarter results make the scale clearer than the company's own marketing usually does. Product revenue hit $653.3 million, up 208.4% from $211.9 million a year earlier. Full-year 2026 guidance was raised to a range of $3.4 billion to $3.8 billion.
The Oracle relationship, the seed of the story, is also still expanding. On April 13, Bloom and Oracle announced an extension of their master services agreement for up to 2.8 gigawatts of fuel cell capacity, of which 1.2 gigawatts are already deploying across U.S. Oracle projects. CNBC noted the deal came days after a $400 million stock warrant. Two weeks later, Oracle, BorderPlex Digital, and Bloom said Project Jupiter in Doña Ana County, New Mexico, an AI data center campus, would be fully powered on site by Bloom fuel cells.
That is the half of the story anyone can find on a quote page. The half worth rereading is the rest of the backlog. According to the Motley Fool recap of the Q1 call, management told analysts more than half the data-center backlog is now non-Oracle, spanning cloud and colocation operators that have not previously been named in press releases. In effect, the same constraint that drove Oracle into Bloom's arms is now driving Oracle's competitors into the same arms. On-site fuel cells are not an Oracle-specific procurement choice; they are an industry-wide workaround for a grid that cannot expand at the speed AI customers want.
KR Sridhar, Bloom's chief executive, took that argument one step further on a recent 20VC episode with Harry Stebbings. Per the BigGo Finance recap of that conversation, Sridhar's claim is blunt: electricity is the input into this factory, and the winners of the AI race will be decided by who can plug in megawatts, not by who publishes the next model card. That is Sridhar's thesis and Bloom's commercial interest, not a market consensus, and it deserves to be read as an argument from a man whose company sells the units being plugged in. Treating it as established fact would be sloppy; treating it as a credible industry read from a major equipment supplier is fair.
The capital markets are voting similarly. Per the Harry Stebbings promotional post on X and the Apple Podcasts episode description, Bloom Energy is Leo Aschenbrenner's largest single holding inside Situational Awareness LP, his multibillion-dollar fund concentrated on the AI infrastructure trade, at roughly 16% of the portfolio. The fund's most recent 13F disclosed about 6.5 million BE shares, and per the March Fortune profile AUM sits near $13.7 billion. Bloom's stock has moved in lockstep, from around $105 in October 2025 per ECIKS and Yahoo Finance to roughly $290.81 by May 2026 per the same sources, an approximate 176% gain. The $90 billion market-cap figure still appearing in some episode titles and headlines reflects Aschenbrenner's own promotional framing; per stockanalysis.com, current live market cap was approximately $64.85 billion as of April 24, 2026.
Aschenbrenner and Sridhar have framed the moment as the opposite of an AI capex bubble, with the buildout pulled forward by a physical constraint grid planners cannot ease in months. That argument is theirs, not established consensus, and it hinges on whether on-site generation can keep up with permit-to-energize timelines for the next several years. Bloom's own 2026 Data Center Power Report: Mid-Year Pulse, published in June, makes the constraint more candidly than the IR materials. A Vested Finance summary citing Bloom's own disclosures puts industry-wide fuel cell data-center agreements between October 2025 and January 2026 at about $7.65 billion, an order of magnitude beyond prior years.
The watch items are also concrete. The Q2 earnings call later this summer is the first public check on whether the non-Oracle share of backlog kept climbing or stabilized. So is any disclosure from one of the unnamed cloud or colocation customers in that backlog on how many megawatts of permitted AI capacity they have lost to interconnection delays. A single concrete number from a major buyer on stranded capacity would do more to validate or puncture the Sridhar thesis than another quarter of headline revenue.