✦ Oracle & the Hyperscaler ContractsNarrow moat

The single largest commitment in Bloom's history arrived on April 13, 2026: a master services agreement under which Oracle intends to procure up to 2.8 gigawatts of Bloom fuel cells for its cloud and AI infrastructure, with an initial 1.2 gigawatts already contracted and deployment under way¹. For scale, Bloom's entire installed base before the AI era was well under a gigawatt.

The reason a cloud company buys fuel cells rather than grid power is time. Interconnection queues in the major American data-center markets run years, and a data center that cannot energize cannot earn — so a hyperscaler will pay a premium per kilowatt-hour to start eighteen months sooner. That is Bloom's entire commercial thesis, and Oracle's order is the clearest evidence yet that the buyers accept it.

The caution is concentration. A company whose order book is dominated by a handful of names has traded the problem of finding customers for the problem of depending on them, and the phrase carrying all the weight here is 'up to' — a master services agreement is a ceiling, not a purchase order, and only 1.2 of the 2.8 gigawatts is contracted. Watch the conversion from the framework to firm orders, and watch how much of quarterly product revenue comes from the top two customers. Concentration is survivable while demand runs hot; it is what makes the cold part of a cycle existential.