Apr. 23 at 3:20 PM
$NVTS vs
$VICR — important distinction in AI data center power:
$NVTS is pushing a single-stage architecture \(800V → 6V\) using GaN, eliminating the 48V intermediate bus entirely. Fewer steps = potentially higher efficiency, lower losses, and smaller footprint. This is a disruptive redesign approach.
$VICR, on the other hand, uses its proven factorized power architecture:
\(800V → 48V → ~1V\), with modular converters and point-of-load regulation near the GPU. This enables better transient response and tighter voltage control, which is critical for AI workloads—and it’s already deployed with NVIDIA.
Tradeoff:
-
$NVTS = simpler, fewer stages, higher theoretical efficiency
-
$VICR = modular, highly optimized, already scaled in production
Bottom line:
$VICR leads today on deployment,
$NVTS is betting on next-gen architecture shifts.
