Inside Solid-State Transformers: Power for AI Data Centers

Ep. 1 2:58

Watch on YouTube Next episode: Ep. 2

About this episode

Why are AI data centers exploring solid-state transformers (SSTs) and 800 V DC? See what happens inside an SST, why higher voltage reduces current, and how the complete power architecture affects losses.

In Episode 1 of AIDC Decoded, simple visualizations follow power from the medium-voltage AC supply to the DC power used by AI hardware. We unpack one SST design, then compare illustrative AC and 800 V DC distribution paths.

INSIDE A SOLID-STATE TRANSFORMER

Follow the illustrated path:
Medium-voltage AC → rectifier → DC link → high-frequency inverter → magnetic transformer → output converter → 800 V DC → rack DC/DC conversion

“Solid-state” does not mean there is no transformer inside. Power electronics create high-frequency AC, allowing a smaller magnetic transformer to provide electrical isolation and voltage conversion in this design.

WHAT YOU’LL LEARN

  • How an SST combines power electronics with a high-frequency transformer
  • Why AI rack power demand is driving interest in 800 V DC distribution
  • How conversion stages affect the energy reaching the load
  • Why 100 kW requires 250 A at 400 V DC, but 125 A at 800 V DC
  • How lower current changes cable losses and conductor requirements
  • Why a complete comparison must include backup performance

READ THE EFFICIENCY EXAMPLE CORRECTLY

The 100 kWh example delivers approximately 91.7 kWh through the illustrated AC path and 95.6 kWh through the illustrated DC path. Those numbers come from assumed stage efficiencies. The DC example still uses a conventional transformer plus an AC/DC converter—it is not a measured SST efficiency result. Replacing those stages with an SST requires a fresh calculation for the complete system.

At half the current, the same cable resistance produces one-quarter of the resistive loss. Actual conductor sizing must also meet thermal, protection, voltage-drop and installation requirements; the illustration is not a universal copper-saving guarantee.

WHY THIS TOPIC MATTERS NOW

NVIDIA and the Open Compute Project are advancing 800 V DC power architectures for AI infrastructure. SSTs are one developing implementation route; conventional transformer-and-rectifier arrangements remain another. An SST alone does not provide stored backup energy.

Further reading:
NVIDIA — 800 V DC architecture:
https://developer.nvidia.com/blog/nvidia-800-v-hvdc-architecture-will-power-the-next-generation-of-ai-factories/

Terms in this episode

  • Alternating current (the grid side) vs. direct current (the battery side). A BESS converts between them through the PCS.

  • The rate of charge flow, in amperes. Fix a voltage and every MW becomes a current — and it is current, not power, that conductors, busbars, fuses and contactors are rated for.

  • Resistive heating in any current path — dissipated power is current squared times resistance, so twice the current makes four times the heat. The scaling law behind high-voltage design, cable loss budgets and the rate-dependent share of round-trip losses.

  • Electrical potential difference between two points, in volts. A BESS stacks it from 3.2 V cells to a ~1,150-1,330 V DC string, converts through the PCS to LV AC, and steps up through transformers to the grid — no rating means anything without the voltage it was quoted at.

Go deeper

Build your knowledge with structured courses and practical learning resources.

Powering AI Data Centers: BESS, UPS & Grid Integration

Learn how BESS, UPS systems and rack buffers support AI data centers. Compare power architectures, work through preliminary MW, MWh and MVA sizing, protect backup reserves, and explore the move toward 800 VDC.

Learn it in Powering AI Data Centers