Bloom Energy Introduced 800V DC Data Center Power
The new architecture aims to cut data center capital costs by bypassing traditional grid infrastructure.
Updated on Sept. 20, 2026 in Data Centers
Live Poll
Will efficient DC power architectures lower the long-term operational costs of AI data centers?
Bloom Energy has released a report detailing an 800V DC-native power architecture that generates on-site electricity for AI data centers. By utilizing solid oxide fuel cells to supply continuous DC power directly to AI accelerators, the system intends to eliminate traditional AC grid conversion steps.
Why it matters
As AI workloads demand higher density, this transition to direct DC power addresses the critical bottleneck of power distribution efficiency in massive-scale computing. The shift seeks to bypass the increasing constraints of aging electrical grids by moving generation directly to the server rack.
The architecture utilizes an 800V voltage rating to power AI hardware, with projections indicating a $3.6 billion reduction in non-compute capital expenditures for a 1 GW facility. This represents a 27% cost reduction compared to standard AC-based distribution methods.
The players
Bloom Energy
A provider of solid oxide fuel cell technology focused on on-site power generation for industrial and data center applications.
NVIDIA
The dominant designer of graphics processing units and AI accelerators that set the infrastructure standards for modern computing racks.
Nebius
A cloud platform and infrastructure company currently deploying high-performance computing power systems.
The details
The system employs solid oxide fuel cells — electrochemical devices that convert fuel into electricity without combustion — to generate power on-site using natural gas, biogas, or hydrogen. By delivering DC output directly to the server racks, the design avoids the energy loss associated with converting power from the alternating current (AC) grid to the direct current (DC) required by compute hardware. This process integrates directly with equipment racks, as demonstrated by the project contracted with Nebius for 328 MW of capacity.
Timeline
2027: NVIDIA is scheduled to adopt 800V DC standards for rack designs.
2030: DC-based power architectures are projected to reach 58% of new data center deployments.
The Tech Race
This development follows the technical transition path set by NVIDIA to shift high-performance data centers away from conventional AC power infrastructure. It competes against legacy grid integration methods by prioritizing direct, on-site DC delivery to meet the surging power demands of AI silicon.
The architecture primarily impacts large-scale data center operators and cloud providers, such as Nebius, by reducing the capital expenditure required to bring new AI capacity online. Future hardware deployments will likely require alignment with these 800V standards as the industry scales toward DC-native rack designs.
The takeaway
This architecture signals a broad industry shift toward on-site, DC-native power generation to sustain massive AI workloads. Watch for the 2027 rollout of rack designs to see if the promised 27% capital expenditure reduction holds at scale.
Further reading
For more on evolving infrastructure requirements, see Data Centers.
Live Poll
Will efficient DC power architectures lower the long-term operational costs of AI data centers?
