Surging AI compute has pushed optical-interconnect valuations higher and
amplified stock volatility. A SemiAnalysis report alleging delays to CPO
(co‑packaged optics) mass production sparked a global sell-off in
optical-communications names; NVIDIA executives denied a delay at a Taipei event
and industry contacts say CPO mass production has not been postponed. Market
confusion appears to conflate small‑batch validation with full industry rollout.
Multiple industry sources identify indium‑phosphide laser chips as the binding
constraint: long photonic‑chip fab buildouts and extended customer validation
cycles make lasers the current capacity bottleneck. Strategically, CPO remains
the long‑term solution for hyperscale model training clusters, but near term
NPO, LPO and conventional pluggable modules are likely to operate in parallel to
meet ultra‑high bandwidth demand. Deployment will be upstream‑capacity led and
phased; compute‑bandwidth demand remains robust.
