Optical fiber startup Relativity Networks has secured $22 million in SAFE note funding and booked a $40 million follow-on order from an unnamed hyperscaler to deploy hollow-core fiber across distributed AI data center facilities.
The funding round included participation from Rhapsody Venture Partners, Bell Ventures Inc., and Faster Than Glass LLC. The capital will support scaling production and deployment of hollow-core fiber cables engineered specifically for low-latency interconnects between AI computing clusters.
The Physical Constraints of Distributed AI Clusters
Scaling modern frontier AI clusters has increasingly collided with regional electrical grid constraints. Single facilities often cannot secure the hundreds of megawatts or gigawatts required for next-generation training runs, forcing operators to distribute GPU clusters across multiple buildings or distinct campuses.
Distributed collective communication operations, including AllReduce, All-to-All, and pipeline synchronization, require tightly bounded latency. In standard single-mode silica optical fiber, light propagates through solid glass with a refractive index of approximately 1.45, resulting in a signal propagation delay of roughly 5.0 microseconds per kilometer.
Hollow-core fiber replaces the solid glass core with an air-filled or evacuated capillary structure. Because the refractive index of air approaches 1.0, light travels at near-vacuum speed, reducing propagation latency to approximately 3.5 microseconds per kilometer. This represents a 30 percent reduction in transit delay compared to conventional glass fibers.
Expanding the Synchronization Radius for Campus Networks
The 30 percent reduction in optical latency directly translates to geographic flexibility. For a given synchronization time budget, cluster architects can place data center buildings roughly 30 percent farther apart without degrading collective communication throughput or stalling gradient synchronizations.
According to Relativity Networks CEO Jason Eichenholz, early AI infrastructure focused primarily on GPU compute density and intra-rack switching topologies. As cluster sizes expanded to tens of thousands of accelerators, optical transport delays across campus networks emerged as a primary scaling bottleneck.
The company previously entered into a manufacturing partnership with Prysmian Group to produce hollow-core optical cables at high volumes from a dedicated facility in Eindhoven, Netherlands. The new $40 million commercial order reflects increasing adoption among tier-one cloud operators seeking to link geographically adjacent data centers into unified logical computing nodes.
