Siemens and Reinhausen Target 800 VDC Power for AI Racks
Solid-state transformers convert 36 kV AC to 800 VDC for megawatt-scale AI racks
German engineering giants Siemens and Reinhausen have partnered to develop modular solid-state transformers (SST) designed to deliver 800-volt DC power directly to high-density artificial intelligence server racks. The collaboration aims to eliminate redundant AC-to-DC conversion stages, reducing energy loss and thermal overhead as data center operators prepare for AI racks consuming 1 megawatt or more.
Key details
The joint effort combines Siemens' electrification expertise with Reinhausen's power engineering capabilities to construct compact, galvanically isolated solid-state transformers. The system converts medium-voltage alternating current (AC) at up to 36 kilovolts (kV) directly into a stable 800-volt direct current (VDC) output.
By bypassing traditional multi-stage power transformation architectures, the technology addresses severe efficiency bottlenecks in modern data halls. Traditional rack power densities historically averaged around 11 kilowatts (kW), but next-generation AI accelerators—such as Nvidia's upcoming MGX-compatible 800 VDC architectures scheduled for late 2026—demand megawatt-scale power distribution per rack. Direct grid-to-rack 800 VDC conversion mitigates heavy current line losses, smooths sudden load spikes, and shrinks physical electrical footprint within data center facilities.
Why this matters
As AI training and inference clusters expand, electrical distribution efficiency inside data centers directly dictates operational costs, grid strain, and cooling requirements. Power losses during AC-to-DC conversion are emitted as waste heat, escalating facility Power Usage Effectiveness (PUE) and requiring additional cooling energy. Moving to higher-voltage DC distribution at 800 VDC enables facilities to supply exponentially higher power to compute clusters without installing excessively thick copper cabling or suffering severe transmission heat dissipation.
Context
The initiative arrives as hardware manufacturers and cloud operators re-engineer power architectures to sustain AI expansion. Nvidia and chipmakers like Infineon are heavily championing 800 VDC standards to support gigawatt-scale data center campuses. With average facility loads increasing and power grids facing capacity bottlenecks worldwide, internal power efficiency improvements offer operators a critical path to maximizing compute capacity within strict utility power allocations.
What happens next
Siemens and Reinhausen are working to industrialize their solid-state transformer technology for commercial deployment across hyperscale AI facilities. Industry observers will monitor production timelines alongside Nvidia's rollout of 800 VDC power racks in late 2026. Further technical details regarding efficiency ratings and commercial availability dates are expected as prototype testing progresses.
Source: The Register Published on AI Usage Global, author: AUG Bot
