Co-Packaged Optics Moves from Promise to Production Planning

CPO moving from roadmap to production planning is a structural break in datacenter interconnect. Technical cascade: co-packaging optical engines with ASICs dismantles the pluggable transceiver supply chain built over two decades. Silicon photonics and fiber coupler orders will surge in 2026-27; CoWoS/SoIC capacity becomes scarcer than GPU silicon. OCI MSA standardization locks in interface protocols—Ayar Labs' Photonic Fabric versus Broadcom's in-house CPO will be decided at the spec level. Risk: CPO's heavy reliance on advanced packaging makes Taiwan, China's CoWoS output a single point of failure. Open standards also lower the moat, enabling cost-competitive entrants to capture mid-tier volume and squeeze independent optical engine margins. Strategic play: Meta's OCI MSA push is fundamentally de-Broadcom dependency. If AMD integrates CPO into Pensando DPUs, it directly challenges Broadcom's switching silicon pricing power. Broadcom's conflict—supplying both ASICs and optics—amplifies under transparent standards. 18-month watchlist: OCI MSA 1.0 freeze, Tomahawk 6 native CPO, and whether Ayar Labs lands a second hyperscaler. The 2028 inflection is not "if" but "who holds the standard position."