Kopin Corporation and Fabric.AI unveiled a MicroLED-based optical interconnect architecture aimed at replacing copper links in AI data centers. The companies secured a $15 million initial development order to fund a demonstration chipset for their jointly developed Neural I/O™ platform, which targets GPU-to-GPU, board-level, and rack-scale communications.
The Neural I/O architecture repurposes programmable MicroLED pixels as high-speed optical transceivers, using photons rather than electrons to move data across systems. The approach seeks to address two persistent bottlenecks in AI infrastructure—bandwidth scaling and power consumption—by reducing the energy required per bit and eliminating reliance on both copper interconnects and conventional laser-based optical modules. The companies position the technology as a path toward real-time, large-scale data exchange across tightly coupled accelerator clusters.
The collaboration combines Kopin’s MicroLED materials and manufacturing expertise with Fabric.AI’s system-level design for AI “factory” infrastructure. Kopin will serve as the exclusive manufacturer of the Neural I/O™ chipsets and holds a 19.9% ownership stake in Fabric.AI. The companies frame the effort as an expansion of MicroLED applications beyond display markets into high-performance interconnects for next-generation AI systems.
- $15 million initial order funds development of Neural I/O™ demonstration chipset
- MicroLED pixels used as bi-directional optical transmitters for chip-to-chip and rack-scale links
- Targets GPU-to-GPU, board-to-board, and rack-to-rack interconnect use cases
- Aims to reduce power per bit and eliminate copper and traditional laser-based optics
- Kopin to act as exclusive manufacturing partner; holds 19.9% stake in Fabric.AI
- Positions MicroLED as a potential alternative to pluggable optics and co-packaged optics approaches
“The marriage of our MicroLED technology with our bi-directional NeuralDisplay™ architecture is exactly what the industry needs to break through current interconnect bottlenecks… we are creating a faster, more efficient optical interface capable of supporting GPU-to-GPU communication at massive scale,” said Michael Murray, CEO of Kopin.
🌐 Analysis: The announcement aligns with broader industry efforts to move beyond copper and conventional pluggable optics as AI clusters scale toward higher bandwidth densities and tighter power envelopes. MicroLED-based interconnects represent a less conventional path, but if manufacturability and integration challenges are addressed, the approach could offer a differentiated power and density profile relative to laser-based silicon photonics.