Two separate foundries announced production silicon photonics moves in the same week. UMC delivered first production wafers for SILITH's 1.6T optical interconnect platform from its 300mm Singapore fab, and announced a customer development program starting in 2027. Tower Semiconductor announced a ~$4B expansion of silicon photonics, silicon-germanium, and advanced-packaging capacity in Japan. When two unrelated foundries make production commitments in the same week, the threshold question shifts from "will this be available" to "when do I design for it."
Silicon photonics manufacturing has been confined to a short list of specialty lines: GlobalFoundries' Albany facility, Intel's photonics group, ST Microelectronics' limited volume. Moving to UMC's mainstream 300mm process and Tower's specialty fab in Japan means silicon photonics chip design can route to non-captive foundries with real production capacity. SILITH's 1.6T platform is targeting optical interconnect for AI datacenters and HPC, exactly the application where PCIe and NVLink bandwidth ceilings are becoming a first-order constraint on system architecture.
Any team designing AI accelerators or HPC fabric chips that assumed silicon photonics would remain a specialty-fab bottleneck through 2028 should revise that assumption. UMC customer development starts in 2027; Tower's Japan capacity builds in parallel. Foundry-accessible silicon photonics for chip-to-chip optical interconnect is an 18-24 month runway from now. The design-point conversation for any chip targeting post-PCIe interconnect bandwidth changes when the fab is accessible, not when the physics works. The physics has worked for years. The fab is now accessible.