Nvidia Pushes 800V Power for AI Data Centers
The artificial intelligence boom is pushing data centers toward a major change in power infrastructure. Nvidia, Google and Microsoft are among the companies backing 800-volt direct-current architectures for next-generation AI facilities. However, the industry has not settled on one identical design. Instead, different approaches are converging around the 800V power class.
Nvidia has developed its own 800 VDC architecture for AI factories. The company says the design can support future systems reaching 1 megawatt per rack and beyond. Meanwhile, Google and Microsoft have contributed to Open Compute Project work on high-voltage DC distribution. Their efforts include the Diablo 400 specification, which uses a bipolar ±400V design.
Why AI Data Centers Need Higher Voltage
AI racks are consuming far more electricity than conventional servers. Therefore, traditional 54V rack distribution creates increasing challenges as computing density rises.
Higher voltage allows the same amount of power to travel with less current. As a result, operators can reduce resistive losses, copper requirements and cable bulk. Nvidia says its architecture can also reduce conversion stages across the power chain.
The proposed architecture moves much of the conversion process away from individual compute racks. Instead, facilities can convert incoming AC power to high-voltage DC and distribute it closer to the compute systems. Consequently, racks can dedicate more physical space to GPUs and cooling equipment.
Industry Moves Toward 800V Infrastructure
The Open Compute Project has become an important forum for developing these power systems. At its 2025 summit, Google and Microsoft engineers participated in work covering advanced AC/DC distribution. The event also featured Nvidia’s 800VDC rack initiatives and Microsoft’s work on AI data-center power stabilization.
Meanwhile, the 2026 OCP EMEA Summit featured sessions focused on low-voltage DC distribution and 1MW-per-rack infrastructure. Therefore, the shift now extends beyond Nvidia’s proprietary roadmap into broader industry standardization efforts.
Still, the transition will take time. Safety requirements, electrical standards, equipment availability and different architectural choices remain significant challenges. In addition, 48V and 54V systems will continue operating alongside newer designs during the transition.
Nvidia expects its 800 VDC architecture to support the next generation of AI infrastructure. Meanwhile, hyperscalers are developing related approaches through open standards. Together, these efforts signal that higher-voltage DC power is becoming increasingly important as AI data centers move toward megawatt-scale racks.
