Hollow-core fiber startup lands $22M to cut data center latency
Relativity Networks has raised $22 million in SAFE notes from investors including Rhapsody Venture Partners and Bell Ventures. The company's hollow-core fiber transmits light through a vacuum, cutting signal travel time by about 30% compared to conventional fiber. It also secured a $40 million follow-on order from an unnamed hyperscaler, aiming to help multi-campus data centers operate as synchronized systems.
The funding arrives as data center construction faces mounting pressure from power availability and zoning limits, pushing operators to reuse existing buildings across wider areas. Relativity’s hollow-core fiber, which routes light through a vacuum rather than glass, cuts per-kilometer signal delay from about five microseconds to 3.5. That 30% reduction lets synchronized compute clusters stretch their effective radius, so a campus spanning dozens of buildings—or multiple campuses—can behave as one machine. The unnamed hyperscaler’s $40 million follow-on order suggests early commercial traction beyond the seed stage.
The company’s approach targets the physical bottleneck of AI scale: as GPU clusters grow, they must be split across more buildings, and the time light takes to travel between them becomes a real constraint. By shrinking that latency, Relativity effectively relaxes the distance limit for tightly coupled workloads. This could let operators retrofit older, power-rich sites instead of building new ones, shifting the economics of where and how quickly AI infrastructure can be deployed.
If hollow-core fiber proves reliable at scale, it could reshape where data centers are built, easing pressure on power-strained regions and enabling reuse of existing facilities. Operators may gain more flexibility in site selection, potentially lowering costs and speeding deployment. Society could see faster AI services and reduced environmental impact from new construction, though the technology’s real-world durability and integration costs remain unproven.