Data center developers are projected to pour as much as $4 trillion into construction by the end of the decade—yet their choices about where to build remain tightly constrained by political hurdles and power-grid limitations. While most in the industry treat fiber-optic speed as a fixed reality, one startup is wagering that faster fiber could fundamentally reshape the geographic calculus behind data center expansion. On Tuesday, Relativity Networks revealed it had raised $22 million in SAFE note funding, with participation from Rhapsody Venture Partners, Bell Ventures Inc., and Faster Than Glass LLC, among others. A SAFE note—an instrument that converts into a specific number of shares once the company completes its first priced funding round—is a standard vehicle for pre-seed and seed-stage investments. The company also locked in a $40 million follow-on order from a major unnamed hyperscaler.

Relativity Networks specializes in hollow-core fiber, a rarely used technology that moves data 30% faster than conventional fiber. Traditional fiber sends light through fiber-optic glass, but hollow-core fiber transmits that same light through a vacuum-filled channel at the line’s center, bringing it much closer to the physical limit of light speed. The gains are measured in microseconds. CEO Jason Eisenholz notes that a signal traveling one kilometer through standard fiber takes roughly five microseconds; with hollow-core, that drops to just three and a half microseconds. When AI workloads ran on a single rack of GPUs, such fiber latency was easy to overlook—but as computing scales have exploded, so has the physical distance between those GPUs. Today, it’s not unusual for a data center campus to stretch across hundreds of acres and dozens of separate buildings.

Eisenholz sees a particularly strong opportunity in multi-campus setups, where existing data centers are linked together to function as one cohesive system. “They’re moving to where the warm shell is, but they still need to operate as one synchronized machine,” he said. The technology offers a way to soften the harsh spatial logic that has hampered many ongoing data center projects. In latency terms, slicing transmission time by 30% means developers can extend their operational footprint by 30% before latency becomes a bottleneck. As compute projects continue to grow in scale, Eisenholz believes this could represent a major industry shift. “The first era of AI optimized for compute,” he explained. “It was GPU, GPU, GPU. The second era optimized the networking inside the data center to take advantage of that compute. The third era that we see coming is optimizing the geography.”