AI semiconductor companies that expand from pure-silicon (signal conditioning chips, retimers) into integrated systems (cable modules, switch fabrics, optical) experience a structural gross margin step-down of 300–500bps as the hardware-intensive products scale. This is different from the SaaS-to-hardware pattern: the base is already high-margin silicon (~77–78%), not software, and the expansion is into systems engineering rather than consumer devices.
For AI semiconductor companies, track product mix at the silicon vs. systems level — not just semiconductor vs. non-semiconductor. A company expanding into cable modules, optical transceivers, or switch fabrics is making a deliberate TAM expansion tradeoff: more revenue, lower margins. When evaluating these companies, build a two-segment model (pure-silicon GM vs. systems GM) and project blend forward as the product roadmap matures. A 300–500bp margin step-down is the expected structural cost of becoming a platform rather than a point-solution chip supplier.