Nvidia, SPIL target CPO volume shipments in 2027

Nvidia's 2027 CPO mass-production lock-in with SPIL signals an irreversible architectural migration from pluggable transceivers to chip-level optical interconnect. The value-chain impact is structural, not incremental: silicon photonics foundries (TSMC COUPE, GF) capture wafer demand; transceiver makers like Innolight and Eoptolink face systemic ASP compression in 2027-28 as CPO eliminates the discrete module BOM layer entirely; data-center TCO models are rewritten with 30-40% per-interconnect power reduction. Supply-chain risk is material. SPIL, as a Taiwan, China packaging asset, faces a 15x-to-30x capacity ramp spanning 2027-28—precisely one full export-control review cycle. The tail scenario: capacity built but customer access constrained by regulatory escalation. Competitively, Broadcom's CPO roadmap lags by roughly 12-18 months, yet its deep ASIC bindings (Trainium, TPU) create a credible 2028+ counter-attack vector. More critically, if TSMC opens COUPE to multiple customers, CPO ceases to be Nvidia's exclusive moat. The 18-month watch item: whether SPIL's CPO yield clears 85%, and whether Rubin Ultra mandates CPO as standard rather than optional.