Micro LED optical interconnect targets AI racks amid unsettled standards

Micro LED entering AI rack interconnect is a light-source displacement war, not an incremental upgrade. VCSELs currently own the 1-3m intra-rack segment, constrained by modulation bandwidth ceilings and thermal drift. If Micro LED achieves >100GHz bandwidth with ±1nm wavelength stability, it directly threatens Coherent and Lumentum's VCSEL fabs, forcing complete re-qualification of transceiver driver circuits, thermal packaging, and switch ASIC SerDes parameters. The Largan-AUO split mirrors a deeper route conflict: external optical coupling versus integrated waveguide—precision optics versus semiconductor-level integration. The core risk is the standards vacuum. OIF and IEEE 802.3 have no Micro LED-specific wavelength or eye-diagram specs. Early mass production invites the 2019 400G "ship-before-spec" inventory trap. Critically, telecom-grade Micro LED demands MOCVD + lithography + bonding hybrid processes, entirely different from display-oriented mass-transfer lines, making capacity conversion in Taiwan, China prohibitively expensive at scale. Broadcom and Marvell will likely accelerate silicon photonics integration to lock in customers. If NVIDIA's NVLink 5.0 stays copper-based, Micro LED's window compresses past 2027. Coherent will defend with VCSEL-plus-silicon-photonics hybrids. Within 12 months, 2-3 hyperscalers will run POCs; a standards whitepaper lands within 24 months; volume production targets 2027-2028. The inflection point is never the technology itself—it is who secures design-in first.