The packaging ladder is the optical story of this decade, and it runs from the faceplate to the silicon. Today, pluggable optical modules sit at the switch faceplate, and the copper run from the switch ASIC to the module costs power at every speed step. The industry’s answer has been to walk the optics toward the silicon — first linear pluggable optics and half-retimed modules, then near-packaged optics (NPO) beside the ASIC, then co-packaged optics (CPO) on the ASIC package itself. We published this progression in our Widening Chasm analysis: the industry is transitioning from copper-based DACs to LPO, half-retimed LRO, and then to near-packaged (NPO) and co-packaged optics (CPO), trading reach for power per bit at every rung.
CPO crossed into commercial reality in 2026, with volumes ramping in 2028. The milestones are in our published record: at GTC 2025, NVIDIA’s Quantum InfiniBand CPO shipments were expected in 2025 and Ethernet CPO shipments in 2026, and by mid-2026 we reported that co-packaged optics had entered production, cutting optical network power consumption by nearly 5x — with optical connectivity already nears 10 percent of total AI factory power. This past week, our blogs said it four more ways: our Marvell investor day analysis framed CPO as the key enabling technology to get scale-up to multi-rack in volume; our AI buildouts analysis centered the laser and indium phosphide supply chain on co-packaged and near-packaged architectures; Broadcom moved its third-generation Tomahawk 6-Davisson 102.4T CPO systems into demonstration deployments; and our OCP Global Summit 2026 read put Tomahawk 6-Davisson 102.4T CPO on the show floor, where NVIDIA’s Rubin Ultra NVL576 carries its scale-up fabric over co-packaged optics. The financial framing is ours and public: we forecast scale-up optics including CPO to add $10B+ by 2030, and with the incorporation of CPO we see the market for AI networking exceeding $1T in the early part of the 2030s.
NPO is the interim rung, and it matters precisely because CPO is a commitment. Near-packaged optics keeps the optical engine close to the ASIC — short copper runs, lower power per bit than a faceplate module — while preserving a level of serviceability that a co-packaged design gives up. Our Marvell investor day analysis made the point directly: CPO is the key enabling technology to reach multi-rack scale-up in volume, with some customers using NPO as an interim step. The logic is the same on the copper side of the house, where near-packaged and co-packaged copper extend DAC reach in the 400G-era systems. An operator chooses NPO when the power savings are worth taking now but the package-level integration is not.
Lasers play a different role in NPO and CPO than they do in a faceplate transceiver, and the categories split by reach and architecture: high-power lasers serve as external light sources that keep heat and the failure point off the ASIC package in CPO designs; VCSELs and micro-VCSELs serve short-reach rack-scale links, as in the 1060 nm VCSEL-based PCIe-over-optics demonstrations at OCP 2026 from Lumentum, Astera Labs, and Corning that we covered from the show floor; and micro-LEDs are emerging at the very-short-reach end. We forecast laser volumes exceeding 1B units by the end of the decade across these NPO and CPO roles — a unit-volume market, not a niche. Our AI buildouts analysis put the supply constraint plainly: the industry does not need a modest number of lasers, and indium phosphide is one of the binding constraints on scale-up, CPO, NPO and the larger AI networking market.
The upcoming 650 Group report on CPO, NPO and Lasers covers this stack end to end: the packaging ladder and its power-per-bit economics, CPO commercialization and the 2028 volume ramp, NPO as the interim architecture, and the laser categories — high-power, VCSEL, micro-VCSEL, and micro-LED — with unit volumes by architecture through the end of the decade. We expect scale-up fabrics to stay copper as long as possible and become a larger CPO opportunity during the crossover, as we laid out in our Tale of Three Networks analysis, and we track the market quarterly by architecture and vendor.
650 Group researches several of the technologies we covered in this blog, including Data Center, Co-Location & Cloud and Enterprise, and more.