Relativity Networks Raises $22 Million to Bring Hollow-Core Fiber to Data Centers

As data center developers race to meet surging demand from artificial intelligence workloads, they face mounting constraints from power availability and permitting hurdles that dictate where new facilities can rise. Spending on data center construction is projected to reach as much as $4 trillion by the end of the decade, yet the industry has largely treated the speed of fiber-optic connections as a fixed limitation. Relativity Networks is challenging that assumption, betting that faster fiber could fundamentally alter the geographic calculus of data center expansion.

On Tuesday, the company announced it had secured $22 million in SAFE note funding from investors including Rhapsody Venture Partners, Bell Ventures Inc., and Faster Than Glass LLC. A SAFE note, which converts into shares upon the company's first priced equity round, is commonly used in pre-seed and seed-stage financing. In addition to the investment, Relativity Networks said it has received a $40 million follow-on order from a leading hyperscaler that asked not to be named.

Relativity Networks specializes in hollow-core fiber, a technology that transmits data roughly 50 percent faster than conventional fiber. While traditional fiber sends light through solid glass strands, hollow-core fiber channels light through a vacuum-filled core, bringing transmission speeds closer to the theoretical limit of light speed. The performance gain is measured in microseconds, but at scale, it can make a significant difference.

CEO Jason Eichenholz estimates that a signal traveling one kilometer through conventional fiber takes about five microseconds. With hollow-core fiber, that drops to approximately three and a half microseconds. In early AI deployments, when compute was concentrated within a single rack of GPUs, such latency was negligible. But as AI systems have scaled, data center campuses now span hundreds of acres and dozens of buildings, increasing the physical distance between interconnected GPUs.

Eichenholz sees particular promise in multi-campus configurations, where existing facilities are linked to operate as a unified system. "The largest systems are distributing the compute across multiple campuses to reach the power that exists," he said. "They're moving to where the warm shell is, but they still need to operate as one synchronized machine."

By halving latency, hollow-core fiber effectively doubles the geographic radius over which data centers can maintain synchronized operations before performance degrades. That could ease some of the spatial pressures that have slowed many ongoing buildouts, allowing developers to tap into existing power infrastructure rather than waiting for new grid capacity.

Eichenholz frames the shift in broader terms. "The first era of AI optimized for compute," he said. "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."