Siemens and Reinhausen have formed a partnership to develop and manufacture a solid-state transformer designed for AI data centers, with the device rated to handle grid voltages up to 36 kilovolts AC and deliver a stable 800-volt DC output directly to rack-level infrastructure.
The technical case centers on collapsing the conventional power-conversion chain. Traditional data center architecture moves power through multiple AC-to-DC stages before it reaches a server—each stage adding heat, cost and square footage. The solid-state transformer eliminates those intermediate steps, taking medium-voltage AC from the utility grid and converting it in a single pass to the 800 VDC that high-density AI servers now require.
Speed matters as much as efficiency. AI training workloads produce sharp, unpredictable swings in power draw as GPU clusters ramp up and down between compute bursts. The solid-state transformer responds to those load changes in milliseconds, a reaction time that conventional magnetic transformers cannot match and that directly affects whether a cluster finishes a training run without interruption.
The 800 VDC output figure is not arbitrary. Hyperscale rack designs from major server vendors have converged on 800-volt DC bus architectures because higher voltage at the same wattage means lower current, which in turn means thinner cop less resistive loss and smaller busbars inside the rack. A transformer that delivers 800 VDC natively removes the need for a separate DC-DC conversion stage inside the facility.
The partnership divides engineering work along each company's existing strengths. Reinhausen brings decades of manufacturing experience in voltage regulation components for power transformers and has built specific expertise in medium-voltage power electronics—the silicon carbide switching devices and associated passive components that sit at the core of any solid-state transformer architecture. Siemens contributes power conversion design, grid integration engineering, protection systems and automation controls.
The architecture is modular. The design supports full 36 kV operation but can be configured in smaller modules for sites connected at lower grid voltages, which extends the addressable market beyond the largest hyperscale campuses to colocation facilities and enterprise data centers that take power at 13.8 kV or lower. A single hardware platform that scales down reduces the engineering cost of adapting the product to different utility interconnect requirements across geographies.
The business logic reflects a supply-side problem the data center industry has not solved. Power electronics capable of handling medium-voltage AC efficiently at the scale AI workloads demand do not exist as off-the-shelf products. Siemens and Reinhausen are building toward industrialization—meaning volume manufacturing, not prototype demonstrations—which is the step that separates a technology development from a revenue line.
Reinhausen's position as the world's leading manufacturer of tap changers and voltage regulation hardware for power transformers gives the partnership credibility with utility engineers and data center electrical contractors, who are skeptical of novel power architectures until they come from vendors with long operating histories. Siemens's grid integration and protection portfolio addresses the interconnection review process that any new transformer technology must clear before a utility will allow it on its network.
The competitive landscape for solid-state transformer technology in data center applications is still forming. ABB has published research on solid-state transformer architectures, and several U.S. Department of Energy-funded projects have demonstrated medium-voltage solid-state transformer designs at smaller scale. None has reached volume production for commercial data center deployment. The Siemens-Reinhausen partnership is positioning to be first to industrialize a 36 kV-rated product for this application.
For hyperscale buyers—the companies spending tens of billions of dollars annually on data center construction—the power architecture question is not academic. Power density per rack has risen from roughly 10 kilowatts five years ago to 40 to 60 kilowatts for current AI clusters, with next-generation liquid-cooled GPU racks pushing toward 100 kilowatts and above. Every additional kilowatt that a more efficient power chain recovers from conversion losses goes directly back into compute capacity without requiring additional utility capacity—the binding constraint on AI infrastructure expansion in most major markets.
No production timeline or pricing has been disclosed. The companies are at the development and industrialization phase, which typically precedes commercial availability by 18 to 36 months for hardware of this complexity.
