AI Data Centers Pivot to 800V Power to Support Megawatt Racks
Borrowing innovations from the electric vehicle industry, data centers are adopting high-voltage DC power to eliminate copper bottlenecks.
AI data centers are transitioning to 800V Direct Current (VDC) power architectures to sustain the massive energy demands of next-generation training models. This shift is essential to prevent power delivery bottlenecks as rack-level energy requirements scale toward the megawatt level.
The transition is driven by an explosion in power density. Rack power requirements have surged from approximately 20kW in the pre-ChatGPT era to 100kW today, with projections suggesting they will exceed 1MW per rack by 2030. Traditional 480V and 54V architectures cannot scale to meet these needs; supporting a 1MW rack with legacy systems would require over 200kg of copper and dedicate an unsustainable amount of physical rack space to power conversion.
The EV Connection
This infrastructure overhaul is heavily inspired by the electric vehicle (EV) industry, which spent the last decade commercializing 800V platforms to enable faster charging and higher efficiency. As EV market growth fluctuates, power electronics providers are pivoting their expertise toward the data center.
Central to this move are Wide Bandgap (WBG) semiconductors. Silicon Carbide (SiC) is being deployed for high-voltage conversion stages and power factor correction, while Gallium Nitride (GaN) is utilized for low-voltage conversion due to its superior switching speeds. Just as WBG semiconductors enabled the 800V shift in EVs, they now facilitate the transition to high-voltage DC in the data center.
Industry Implications
Moving to 800VDC represents a fundamental redesign of the data center. By increasing the voltage, operators can significantly reduce cabling weight and minimize I^2R power losses—the energy wasted as heat when electricity travels through conductors. This efficiency is critical for the viability of future AI hardware, allowing massive amounts of power to be delivered to a small physical footprint.
This shift is already gaining institutional momentum. NVIDIA is actively leading the transition, collaborating with silicon providers such as Texas Instruments and Infineon to support 800VDC infrastructure starting in 2027.
What to Watch
The industry is now monitoring how quickly these high-voltage standards can be implemented across existing facilities. While the technical path is clear, the transition requires a coordinated effort between chip designers and facility operators to ensure that the power grid can support the projected 1MW-per-rack density by the end of the decade. This evolution marks a critical juncture where the physical limits of electrical engineering must evolve to keep pace with the exponential growth of AI compute.