Nvidia surged approximately 20% following its latest earnings, with the Vera Rubin architecture launch confirming AI GPU demand remains structurally elevated. Hyperscalers are accelerating capital commitments into next-generation compute infrastructure.
A second dynamic is reshaping supply chains. The U.S. government's 2027 ban on Chinese-origin rare earth materials in defense systems forces domestic sourcing for magnets, substrates, and power devices. This accelerates investment timelines for alternative semiconductor suppliers.
Three investable categories sit below Nvidia in the stack.
Optical Interconnects
POET Technologies' Optical Interposer products offer lower cost, lower power consumption, and smaller form factor than comparable alternatives.1 A Lumilens purchase order could scale to $500M+ in cumulative purchases over five years.1
Lumentum operates at the same optical networking layer, supplying components between AI clusters.2 As GPU density scales, interconnect bandwidth becomes the binding constraint.
Edge Inference Silicon
Arteris technology is embedded in Li Auto's autonomous driving SoC, delivering 2,560 TOPS of compute for the L9 Livis SUV.3 The automotive deployment confirms inference workloads are migrating into purpose-built silicon beyond the data center.
Blaize is building dedicated edge inference hardware targeting defense, automotive, and industrial applications.4 Edge AI requires efficiency profiles that GPU-based data center architecture cannot cost-effectively meet.
Power Semiconductors
Ideal Power is on track to complete Stellantis deliverables under its existing purchase order by mid-2026.5 Automotive electrification and defense modernization are converging demand drivers for next-generation power devices.
The 2027 rare earth ban directly constrains sourcing for permanent magnets and substrates. Domestic power semiconductor suppliers gain structural demand independent of AI cycle timing.
The Investment Case
Nvidia's training compute position is not the question. The opportunity is in enabling infrastructure — interconnects, power devices, edge processors — that scales alongside hyperscaler capital expenditure.
Each category addresses a supply chain node where geopolitical risk or architectural transition creates demand incumbents cannot satisfy alone. The rare earth ban, bandwidth bottlenecks, and edge AI deployment are converging timelines, not distant catalysts.
