AI supply-chain thesis — mapping bottlenecks, focus companies, and supply-chain exposure for investors.
**Bottleneck theme:** Power & Grid **Focus:** $VICR — VICOR CORP Vicor sells last-inch power-delivery modules — the high-density DC-DC conversion stages that sit between the rack-level 48V (or, increasingly, 800V HVDC) bus and the GPU package. As GPU power has climbed from ~400W (H100) to ~1,000W (Blackwell) and is heading toward 1,800W+ on Rubin Ultra, the conversion problem is no longer trivial. Every additional ampere of current at the package is wasted heat and lost compute density; vertical power delivery (VPD) — Vicor's specialty — has become the only credible path to feeding 1.5+ kW into a GPU socket without melting the substrate. Vicor's Factorized Power Architecture and Pre-Regulator/Current-Multiplier topology have been NVIDIA reference designs since the H100 generation. The thesis is straightforward: power per GPU continues to climb, last-inch modules are a per-socket tax that Vicor charges, and competitive intensity is constrained by patent portfolio and customer qualification cycles. Execution risk is real — Vicor has historically been opex-heavy and capacity-constrained — but the demand backdrop is unambiguous. Pair with $MPWR (broader power management), $NVTS (GaN front-end), and $ETN (rack-level integration).
Browse all AI supply-chain theses · Macro trends · Industries · Product categories