Faster fiber technology could reshape where data centers get built

The race to build AI data centers is hitting a bottleneck. Developers face mounting pressure from power constraints and geography, with industry spending projected to reach $4 trillion by decade's end. But one startup thinks a new approach to transmission cable could fundamentally change where these facilities can operate.

Relativity's hollow-core breakthrough

Relativity Networks just closed a $22 million SAFE round backed by Rhapsody Venture Partners, Bell Ventures, and Faster Than Glass, plus a $40 million order from an unnamed hyperscaler. The company's core innovation is hollow-core fiber—a rarely used alternative that sends data through a vacuum chamber inside the cable instead of through solid glass. The result: light travels 30% faster than in conventional fiber, cutting the time for signals to traverse a kilometer from roughly five microseconds down to three and a half.

Why latency matters at scale

When AI workloads fit on a single GPU rack, transmission delay was negligible. Today's largest systems sprawl across hundreds of acres and multiple buildings, sometimes entire campuses. CEO Jason Eisenholz points to the industry's next frontier: chaining together separate data center sites into a single coordinated computing cluster. "The largest systems are distributing the compute across multiple campuses to reach the power that exists," he explained. "They're moving to where the warm shell is, but they still need to operate as one synchronized machine."

A geography inflection point

By cutting latency by nearly a third, hollow-core fiber effectively allows developers to space out their infrastructure over 30% greater distances without hitting performance walls. Eisenholz sees this as part of a three-stage evolution: first came GPU optimization, then internal-network tuning—now comes geographic optimization. For an industry constrained by power grids and political boundaries, the ability to site facilities farther apart could unlock new clusters and ease the strain on overburdened regions.