Co-packaged optics remain the long-term answer for AI scale-up and scale-out power budgets. They are not yet the only answer shipping in volume. Through 2026, linear pluggable optics (LPO) and related pluggable architectures have taken a large share of new high-speed transceiver spend—moving bits with less DSP power than classic pluggables while keeping the serviceability of a module you can swap in a rack.
That matters for operators who need 800G/1.6T links now without waiting for every switch and accelerator to redesign around on-package optical engines.
What LPO is meant to buy
- Power relief versus copper and heavy DSP pluggables — More watts left for GPUs.
- Field replaceability — Failed optics stay a FRU, not a package scrap event.
- A bridge — Deploy density today while CPO/NPO supply chains and qualifications catch up.
What still decides the mix
Reach, interoperability, and switch ASIC roadmaps. Some fabrics will jump straight to CPO where power and density leave no choice. Others will stay pluggable for years because operations teams refuse to couple optics yield to silicon yield. Laser supply (including InP) and foundry photonics capacity constrain both paths.
LPO is not “CPO lite.” It is a different integration point with different failure modes and sparing models.
What to watch next
- Share of new AI cluster ports shipping LPO/LRO versus classic DSP pluggables versus early CPO.
- Whether hyperscalers standardize one transitional optic SKU across regions.
- How fast CPO cost and service models erase the pluggable advantage.
CPO is the package-level endgame. LPO is the 2026 networking chapter: optical bandwidth you can still pull from a cage.