Kopin, Fabric.AI partner on MicroLED optical interconnect for AI data centers

WESTBOROUGH, Mass. - Kopin Corp. in Westborough, Mass., has announced a strategic collaboration with Fabric.AI in New York to develop MicroLED-based optical interconnect technology intended to replace traditional copper wiring between graphics processing units (GPU) and high-performance processors in artificial intelligence (AI) data centers.

Under the collaboration, Fabric.AI has placed a $15 million purchase order with Kopin to fund a demonstration chipset for the jointly developed Neural I/o optical interconnect technology. The companies said the effort is aimed at addressing bandwidth and power constraints in AI computing architectures by shifting data movement from electrical signaling over copper interconnects to photon-based transmission using MicroLED devices.

The Neural I/o architecture is based on Kopin’s MicroLED technology and its bi-directional NeuralDisplay architecture, which repurposes programmable MicroLED pixels as optical transceivers. In this concept, individual pixels function as high-speed optical transmitters, encoding and sending digital data in parallel channels to enable high-bandwidth chip-to-chip and system-to-system communication while reducing energy consumption per bit.

Related: Kopin to provide microdisplay for Lockheed Martin version of future anti-air missile guidance and control

AI improvements

The companies said current AI accelerator systems rely heavily on dense copper interconnects to enable communication between GPUs and processors, creating significant power consumption and thermal management challenges as system scale increases. In modern AI infrastructure, these constraints are increasingly being addressed through silicon photonics and co-packaged optics, which are now widely deployed in hyperscale data centers and AI training clusters to replace long copper traces with optical links. These systems already support rack-scale and multi-rack GPU communication at data rates in the hundreds of gigabits per second to terabit-class throughput, reducing electrical losses and improving energy efficiency in large-scale compute environments.

Within defense and aerospace applications, silicon photonics and optical interconnect technologies are increasingly foundational to high-performance computing architectures that support intelligence, surveillance, reconnaissance data processing, and AI-enabled command and control systems.

Co-packaged optics is also emerging as a key design approach, integrating optical engines closer to GPU and switch silicon to reduce latency and power consumption in large-scale AI clusters that underpin defense workloads. These existing technologies represent the current production baseline for high-bandwidth military-relevant compute systems, while next-generation photonic approaches remain in development.

Related: 2026 Military + Aerospace Electronics Innovators Awards now open for entries

Kopin said the proposed Neural I/o architecture seeks to extend beyond silicon photonics by eliminating discrete laser-based optical subsystems and instead using MicroLED emitters as direct optical transceivers. The companies said this approach could reduce interconnect complexity while further improving power efficiency in dense AI compute environments, although the technology remains in the demonstration phase.

Kopin Chief Executive Officer Michael Murray said the technology is designed to address interconnect bottlenecks limiting large-scale AI deployments. Fabric.AI Chief Executive Officer Josh Silverman said MicroLED-based interconnects could form a foundational layer for future AI data center architectures.

The companies said the Neural I/o product line remains in development and is intended for demonstration-level validation as part of the collaboration.