Intel says it is the only high-volume fab for 12-inch InP silicon photonics as co-packaged optics nears deployment

Intel's OCP Summit declaration is less a capacity announcement than a strategic lock on the highest-barrier node in the CPO value chain: the InP light source. Silicon cannot emit efficiently at 1310 or 1550 nm, making InP lasers the non-negotiable heart of co-packaged optics. Scaling InP to 12-inch wafers at volume is, as far as the industry can verify, a singular capability — a moat spanning MOCVD epitaxy through wafer-level bonding that no second player has replicated. The ripple effects cut both directions. Upstream, 12-inch InP substrate supply from Sumitomo and JX Metals becomes the new chokepoint. Downstream, Broadcom, Marvell, and NVIDIA's custom ASIC CPO modules will embed Intel's light source as a rigid BOM dependency, shifting pricing power decisively toward Intel. Coherent and Lumentum dominate discrete InP devices but lack wafer-scale foundry infrastructure. TSMC's silicon photonics platform focuses on waveguides and modulators, still sourcing lasers externally. Intel has chosen the picks-and-shovels posture: it does not need to win the AI silicon war — it collects a toll on every laser that ships. The critical risk: if Broadcom or NVIDIA vertical-integrates light-source fabrication by 2027, Intel's foundry window compresses to roughly 18 months. But through the 2026–2028 CPO deployment cycle, Intel has already secured its seat at the table, and the first-mover premium is closing fast.