Rosenblatt sees optical demand running years ahead of supply — visibility now stretches into 2028. CPO and NPO timelines aren't slipping; they're accelerating.
Scale-up optics could be 3x the TAM of scale-out. Why? Per-GPU bandwidth needs are 9x higher, but scale-out requires 3x more optical links due to multi-layer switching. Net result: scale-up wins on size.
For optical vendors, scale-up is 100% net new revenue. Scale-out just replaces legacy transceivers.
Laser pricing is surging. Spot prices for 100G/200G EML and CW lasers jumped high-single to low-double digits last quarter. As LTAs lock in, annual ASP growth should normalize to 15–20% — a sharp reversal from the 10–20% annual price erosion before AI.
Chinese laser makers are still ~2 nodes behind and a couple years away from 300mW output. They'll close the gap on lower-power 1.6T silicon photonics over the next 18 months, but U.S. suppliers hold the moat on high-power and ultra-low-loss CPO lasers.
Optical infrastructure is the new battleground. Supply can't keep up, pricing is firming, and the competitive lead is still intact.
Scale-up optics could be 3x the TAM of scale-out. Why? Per-GPU bandwidth needs are 9x higher, but scale-out requires 3x more optical links due to multi-layer switching. Net result: scale-up wins on size.
For optical vendors, scale-up is 100% net new revenue. Scale-out just replaces legacy transceivers.
Laser pricing is surging. Spot prices for 100G/200G EML and CW lasers jumped high-single to low-double digits last quarter. As LTAs lock in, annual ASP growth should normalize to 15–20% — a sharp reversal from the 10–20% annual price erosion before AI.
Chinese laser makers are still ~2 nodes behind and a couple years away from 300mW output. They'll close the gap on lower-power 1.6T silicon photonics over the next 18 months, but U.S. suppliers hold the moat on high-power and ultra-low-loss CPO lasers.
Optical infrastructure is the new battleground. Supply can't keep up, pricing is firming, and the competitive lead is still intact.