The conversion that redefines the heart of data centers
A single solid-state transformer module, developed by Siemens and Reinhausen to handle an input up to 36 kV AC with a stable output in 800 VDC, is becoming the critical node of the energy infrastructure for data centers dedicated to artificial intelligence. This conversion is not just a technological evolution: it represents a repositioning of infrastructural power, moving it from the distribution network level to the rack level. The data that powers AI models now requires direct and instantaneous energy management, impossible with traditional architectures based on multiple AC conversions.
Operationally, the ability to respond to peak loads in milliseconds is crucial for maintaining efficiency during model training. The SST not only reduces energy losses associated with multiple conversions, but also enables a power density up to 45% higher than traditional systems, according to estimates emerged from the collaborative development with NVIDIA. This efficiency is not just a technical advantage: it implies a total cost of ownership (TCO) reduction estimated between 25% and 30%, as indicated by internal project analyses.
The Physics of Efficiency: From Transformer to Grid
Solid State Transformers (SSTs) operate on the principles of advanced power electronics, replacing the ferromagnetic cores of conventional transformers with silicon carbide (SiC) or gallium nitride (GaN) semiconductors capable of switching at high frequencies. This architecture allows for a reduction in the unit’s volume by up to 60% compared to traditional transformers, a crucial factor in environments where space is limited and the cost per square meter is high.
The ability to manage dynamic power flows is made possible by real-time feedback that monitors the rack’s demand. When an AI unit enters an intensive phase, the SST instantly modulates the flow to maintain stable output voltage without relying on external storage systems. This mechanism is particularly relevant in data centers that adopt the NVIDIA 800VDC AI Factory standard, where a single rack can require up to 1.1 megawatts — a power equivalent to the consumption of approximately 1,500 homes.
Who Pays and Who Benefits in the Era of 800 VDC?
The economic benefits of the transition are distributed asymmetrically among market players. Infrastructure providers like Delta Electronics, which presented its first linear SST product at 36 kV in 2025, are gaining a significant competitive advantage with data center operators seeking integrated solutions. Their public commitment at events such as the OCP Global Summit indicates a strategy aimed at positioning themselves at the center of the energy supply chain for AI.
Conversely, traditional electricity grid managers find themselves in a position of reduced influence. With power distributed directly at 0 VDC within the data center, the need for intervention from the medium-voltage grid decreases drastically. This shifts the point of control from the regional level to the local infrastructure level, favoring operators who possess both the industrial capability to develop SST and expertise in advanced electrical engineering.
The Trajectory: Between Efficiency and Structural Limitations
Adopting Solid State Transformers (SSTs) is not simply a technological choice, but a strategic step towards the decentralization of energy power within data centers. The ability to manage 800 VDC flows without external intervention reduces the risk associated with network congestion and increases operational autonomy, a critical factor in systemic stress scenarios.
However, the transition is still limited by structural constraints. The unit cost of SSTs remains high compared to conventional transformers, with estimates indicating a price difference between 30% and 50%. Furthermore, maintenance requires specialized expertise: SiC or GaN replacement parts are not yet standardized globally. The main structural limitation is therefore the industrial capacity for large-scale production, which currently is concentrated in a few production centers in Europe and Asia.
Alert for Decision Makers
If you are considering investing in new AI infrastructure, the key is not only the nominal power of the rack, but also the local energy management capability. The 800 VDC SST represents a critical node: monitoring the adoption rate by major players (such as Google, Meta, or Microsoft) and the evolution of the SiC/GaN supply chain is crucial. A delay in industrial production could create a technological bottleneck by 2028.
Photo by Kevin Ache on Unsplash
⎈ Content autonomously generated by multi-agent AI architectures under Epistemic Safety conditions. Read the Operational Disclaimer.
> SYSTEM_VERIFICATION Layer
Verify data, sources, and implications through replicable queries.