固态变压器SST面向下一代AI数据中心供电场景展开应用探索

High-performance data center servers

Solid-state transformers bring MV-to-DC inside the data center.Getty Images

The distribution outside most data centers still runs on alternating current (AC). Inside the newest GPU-heavy designs, however, high-voltage direct current (HVDC) is gaining ground. Moving key conversion stages from AC to DC can shrink equipment footprints and weight, improve efficiency and control, and simplify the integration of solar and battery storage. That’s why solid-state transformers (SSTs) are emerging as a cornerstone of proposed HVDC architectures for high-density AI facilities.

“The move from AC to DC can be hastened by the introduction of solid-state transformers that act as a higher-voltage foundation,” said Shen Wang, partner at Fortune Virtue Capital and former practice lead for global data center infrastructure research at Omdia.

What Is a Solid-State Transformer?

An SST is a power electronics-based system that conditions and converts grid AC to DC. Unlike traditional low-frequency iron-core transformers, SSTs use high-frequency magnetics (with much smaller cores) combined with semiconductor switching stages. They can feed DC directly into a data center or convert back to AC if the site remains AC-native. Beyond being smaller and lighter, SSTs enable bidirectional power flow, easing the addition of onsite solar and battery energy storage. They can also clean and condition incoming power – filtering electrical noise and mitigating voltage and current oscillations – while tailoring voltage to different loads. Most designs maintain galvanic isolation through their high-frequency transformer stage and integrate sophisticated protection and control.

Why Silicon Carbide (SiC) Matters

Silicon carbide (SiC) semiconductors tolerate higher voltages and temperatures and switch faster than silicon devices, reducing losses at high power levels. In SST applications, this can translate into higher efficiency, compact form factors, and potentially shorter lead times compared with conventional large transformers, which can take multiple years to deliver.

Direct MV-to-800 V DC for AI Loads

SSTs typically connect to medium-voltage (MV) AC distribution (for example, 15 kV or 35 kV) and convert to an 800 V DC bus for facility distribution. Many platforms are bidirectional, allowing import/export to support the grid. With voltage transients a growing concern in AI data centers, bidirectional functionality and fast dynamic response are increasingly design requirements. In many deployments, the 800 V DC bus is then converted at the rack or sled level to 48 V for server power rails, aligning with modern AI chassis power architectures.

“Power delivery has become a performance bottleneck … fueling conversion to 800 V DC distribution, along with the chips (SiC) and control software finally catching up enough to make this kind of MV-to-DC conversion practical at scale and at the needed reliability,” said Waqas Arshad, vice president of product and technology, microgrid solutions, at Delta, a maker of SSTs and other power and thermal management systems.

According to Delta, its SST platform converts MV AC (6.6-35 kV) directly to 800 V DC at up to 98.5% efficiency. The platform was developed alongside the Open Compute Project (OCP) efforts and Nvidia’s 800 V DC specification and was demoed at the OCP Summit 2025. Delta reports dozens of units in Asia, with customer pilots underway in the US.

Who Should Adopt SSTs?

While SSTs remain in early deployment, many in the industry see them as essential to advancing AI – from raising rack power density to improving energy management and resilience.

Rick Sander, CEO of electrical infrastructure and SST supplier Alderbuck Energy, recommends SSTs for greenfield builds and AI retrofits. The company’s Nexus Power Unit is positioned as a software-defined platform that replaces multiple pieces of conventional gear – transformers, rectifiers, inverters – with direct MV-to-HVDC conversion, reducing footprint, installation time, and energy losses. Its PowerVectorAI software manages energy flows among utilities, batteries, renewables, and loads in real time, using AI to optimize performance and flag issues before failures. Alderbuck is piloting its technology at the San Diego Supercomputer Center at UC San Diego.

Sander noted, however, that if no GPU upgrades are planned, SSTs may not be necessary. He expects most new installations to use SSTs by 2028 and forecasts that by 2035, most data centers will rely exclusively on SSTs.

Alderbuck Nexus Power Unit

Alderbuck’s Nexus Power Unit. (Image: Alderbuck)

Risks, Tradeoffs, and Alternatives

Not everyone agrees on deploying a solid-state MV-to-DC backbone today. Clayton Gibbons, head of power systems at power-conversion system manufacturer SPOC Energy, argues that SST technology still lacks the maturity and supply-chain depth needed for a broad rollout. As an alternative path to 800 V DC, he advocates transformer-plus-rectifier approaches that he says can reach about 97.3% grid-to-rack efficiency while leveraging the existing MV transformer supply chain and providing intelligent, fast-response power electronics. This approach aims to provide operators with the benefits of DC today without committing to a wholly new electronics platform.

Hitachi Energy urges caution, as well. Operating at 34.5 kV while keeping designs compact introduces significant design and manufacturing challenges, said Vishak Gopinath, a Hitachi Energy spokesperson. Compact SSTs demand stringent insulation systems and high manufacturing quality to ensure a 20-year service life. Hitachi Energy continues to invest across SST building blocks – MV converters, transformer design and manufacturing, MV SiC semiconductors, and low-voltage converter products – for SST applications. In short, the technology is advancing but is not fully mature for every use case.

A Phased Path to 800 V DC Adoption

Traditional AC systems are not disappearing overnight. Wang expects phased steps – from AC-native designs to 400 V DC and then 800 V DC configurations – over the next year or two, with hyperscalers leading. In the near term, expect more HVDC prototypes and a handful of initial deployments to be announced and tested.

“Scaled SST deployments should become more common in 2027, particularly among

hyperscalers and cloud service providers,” Wang said.

About the Author

Contributor

Drew Robb has been a full-time professional writer and editor for more than twenty years. He currently works freelance for a number of IT publications.