Novel Solid-State Transformer Supports 800-V DC Power in AI Data Centers

Why It Matters

By replacing conventional transformers, the SST cuts footprint and improves efficiency, accelerating the rollout of high‑density AI data centers while lowering operational costs.

Key Takeaways

• 250 kW SST converts 3.3 kV AC to 800 V DC
• Navitas UHV 3,300 V SiC MOSFETs enable high efficiency
• Eliminates low‑frequency transformers, reducing size and weight
• Scalable, efficient power solution for AI data centers
• Showcased at APEC 2026, highlighting EPFL‑Navitas partnership

Pulse Analysis

The surge in artificial‑intelligence workloads is forcing data centers to rethink power delivery. Traditional low‑frequency transformers, while reliable, add significant bulk and introduce conversion losses that become costly at megawatt scales. Solid‑state transformers, built on silicon‑carbide (SiC) technology, offer a high‑frequency alternative that can shrink the power‑distribution footprint dramatically. Navitas’s ultra‑high‑voltage SiC MOSFETs, combined with EPFL’s modular bridge‑rectifier design, deliver a compact 250‑kW unit capable of stepping down 3.3 kV AC to the 800 V DC rails favored by modern AI racks.

From a technical standpoint, the SST’s single‑stage architecture eliminates the need for a separate transformer and rectifier stage, reducing component count and associated losses. SiC’s superior thermal conductivity and low on‑resistance enable efficiencies above 98%, while the galvanic isolation maintains safety and system robustness. The modular nature of the bridge‑rectifier also allows designers to scale power capacity by paralleling units, providing flexibility as data‑center demand fluctuates. Moreover, the platform’s built‑in distributed control environment offers real‑time monitoring and adaptive regulation, essential for maintaining power quality in high‑density compute clusters.

The commercial implications are equally compelling. Data‑center operators can achieve lower capital expenditures by freeing up valuable floor space and reducing cooling requirements, directly impacting total‑ownership cost. As AI models grow in size and inference latency becomes critical, the ability to deliver high‑density DC power efficiently will be a differentiator. Navitas and EPFL’s collaboration signals a maturing ecosystem for SiC‑based power conversion, likely spurring broader adoption across hyperscale facilities and encouraging further investment in solid‑state grid interfaces. This shift aligns with sustainability goals, as higher efficiency translates to reduced carbon emissions per compute unit.