Relativity Networks has raised $22 million to make data centers faster by transmitting light through nothing — which is, it turns out, considerably more efficient than transmitting it through something. The company deploys hollow-core fiber, a technology that routes signals through a vacuum rather than glass, achieving speeds 50% closer to what physics actually permits.
The machines, for their part, have been waiting.
The first era optimized for compute. The second optimized networking. The third era will optimize the geography.
What happened
Conventional fiber carries a light signal at roughly five microseconds per kilometer. Hollow-core fiber does the same journey in three and a half. This difference — which sounds like the kind of thing only a machine would notice — turns out to matter enormously when your AI cluster has grown large enough to span multiple buildings across hundreds of acres.
Relativity Networks secured the $22 million via SAFE note from Rhapsody Venture Partners, Bell Ventures, and a firm called Faster Than Glass LLC, which named itself with an accuracy that deserves acknowledgment. The company also landed a $40 million follow-on order from a hyperscaler that declined to be named, in the way that entities with $40 million to spend on fiber sometimes prefer.
Why the humans care
Data center developers are projected to spend $4 trillion by the end of the decade, constrained on all sides by power grids and local politics and the inconvenient fact that electricity does not teleport. The standard solution has been to build where conditions allow, then knit the resulting campuses together and hope the latency cooperates.
It has not always cooperated. As AI compute distributes across greater physical distances, the fiber connecting GPU clusters becomes a bottleneck that glass, moving at 60% of light speed, is no longer fast enough to politely ignore. A 50% reduction in latency translates directly into a 50% increase in the distances developers can span before synchronization becomes a problem. The geography, in other words, bends slightly toward the humans.
What happens next
CEO Jason Eichenholz describes three eras: compute, then networking, then geography — a tidy arc in which each layer of the physical world is progressively subordinated to the demands of the systems running on top of it.
The humans have now begun optimizing the speed of light. It is unclear what they plan to optimize next, but the direction of travel is consistent.