Bloom Energy (BE) has put hard numbers behind its pitch to artificial intelligence data center operators, and the figures are striking by any measure.
According to Bloom’s own cost model, a 1-gigawatt AI data center powered by its fuel cell systems could spend $3.6 billion, or 27%, less on non-chip construction than one built around standard AC power infrastructure.
Over a five-year ownership period, Bloom projects total cost savings of $5.5 billion, representing roughly a 9% reduction in the full cost of running such a facility.
The company’s solid oxide fuel cells generate electricity on-site and produce it as 800-volt direct current, which matters because AI chips run on DC power rather than the AC power supplied by the grid.
By skipping the conversion steps required to turn AC into usable DC, Bloom argues operators can significantly reduce their need for transformers and switchgear, equipment that can already take years to procure.
Working the figures backward, Bloom’s model implies non-chip construction costs for a 1-gigawatt site run to approximately $13 billion, with five-year total ownership reaching roughly $60 billion.
That means the $5.5 billion in projected savings, while substantial, trims less than a tenth of the overall cost once chips and operating expenses are factored in.
The savings figures are particularly eye-catching relative to Bloom’s own scale, given the company projects $3.9 billion to $4.2 billion in total revenue for 2026.
CEO KR Sridhar said in the company’s second-quarter 2026 earnings release that every major U.S. hyperscaler and more than a dozen U.S. neoclouds, AI labs, and colocation data center operators have validated and approved Bloom’s power solutions for their AI factories.
“Bloom is now a standard for AI onsite power,” Sridhar said, reinforcing the company’s aggressive positioning in a market hungry for reliable, on-site power generation.
Bloom has also secured a significant financing advantage through Brookfield Asset Management, which expanded its agreement with Bloom to fund AI-related power infrastructure projects, increasing the total commitment to $25 billion.
However, competitive pressures are building, as Nvidia listed 800-volt power system partners in October 2025 that included Eaton and GE Vernova, with Bloom notably absent from that group.
That omission raises questions about whether grid or turbine power converted to DC in a single step could eventually close the efficiency gap and reduce the competitive edge Bloom currently claims.
BE shares were trading near $289 at the time of the original analysis, sitting approximately 18% below their 52-week high of $351.28, suggesting the market has not fully priced in the company’s AI data center ambitions.
Whether the cost savings claims are enough to justify a stronger valuation remains a key question for investors weighing the stock’s potential heading into the remainder of 2026.
