Bloom Energy, a global leader in power solutions, today released a special report “The New Rules of AI Power,” showing that its unique 800V DC-native fuel cell solution can reduce non-compute capital expenditures for a 1 GW AI data center by $3.6 billion, or 27%, and five-year total cost of ownership by $5.5 billion, or 9%.
The conventional power system architecture was built over a century ago around the challenge of moving electricity over long distances safely. AC, or alternating current, became dominant because the remotely located centralized power plants could efficiently transport the power they produced at very high voltage over hundreds of miles, and then step it down with copper-consuming transformers to medium and low voltage AC for industrial and residential use. The electro-mechanical industrial age of the last century was built around this high voltage AC electric grid infrastructure catering to medium and low voltage AC power consumption.
The digital age operates differently. Much of the technology that defines modern life, from smartphones and laptops to servers and data centers, runs on low voltage DC power. For years, since the only prevalent source of abundant and continuous power was the AC electric grid, the digital world paid for and managed the wasteful conversion of AC to DC power because it required only modest amounts of equipment and infrastructure. Power-hungry AI chips are completely upending this practice and making it impractical. These GPU chips, and the density with which they are packed within a server rack, need enormous amounts of power concentrated in one place, and require an 800V DC power input.
NVIDIA is already moving toward an 800V DC architecture for next-generation AI infrastructure. Beginning with Rubin Ultra and Kyber rack architecture, NVIDIA has specified 800V DC for 2027 and all generations after it. It is impractical to deliver the amount of power needed in that rack using the old architecture of low-voltage DC.
The grid, onsite turbines, and onsite reciprocating engines all deliver AC power to the data center, which must then be converted to DC that compute ultimately consumes. Bloom generates that DC power from the start with no conversions. Its solid oxide fuel cells generate continuous 800V DC power electrochemically and onsite, removing layers of transport and conversion equipment between the power source and the rack.
This approach reduces the electrical infrastructure required between generation and compute, lowering capital cost, energy losses and dependence on constrained components such as transformers and switchgear, where lead times are already measured in years. The critical materials needed to build them, like copper, are in short supply and becoming increasingly costly. Bloom Energy’s analysis found that for a 1 GW AI data center, using its DC native fuel cells translates to $3.6 billion reduction in non-compute CAPEX and $5.5 billion lower total cost of ownership over five years.
“AI is not just increasing electricity demand. It is catalyzing a positive transformation of how power is generated and consumed by becoming the first scale adopter of both onsite and DC power,” said KR Sridhar, Founder, Chairman and CEO of Bloom Energy. “To move away from the AC paradigm that is over a century old requires the pain of not switching to be far greater than the benefit and comfort of accepting the status quo. AI, by presenting the challenge of consuming enormous amounts of DC power where compute happens, is enabling this transformation. Commercial, industrial, residential and electric vehicles will all be beneficiaries of this massive change. Bloom, by natively generating continuous, reliable and clean 800V DC at scale, is the industry leader in ushering in this new era for electric power.”
“We’re seeing the shift to 800V DC happen faster than many expected,” said Natalie Sunderland, Chief Marketing Officer at Bloom Energy. “Our 2026 Mid-Year Data Center Power Report found that data center leaders expect DC-based architectures to account for 58% of new deployments by 2030. Bloom’s ability to generate continuous 800V DC power natively puts us at the forefront of this transition, helping AI data centers reduce cost, complexity and power losses while accelerating deployment.”
Bloom’s economic analysis is based on its 800V DC Total Cost of Ownership model for a 1 GW compute data center using next-generation AI rack assumptions. Actual project economics will vary based on site design, energy prices, equipment costs and other project-specific factors.
