Faster Fiber Could Change AI Data Center Design

AI infrastructure is starting to run into a problem that sits outside the chip itself: distance. As large AI systems spread across more buildings, campuses, and power-constrained locations, the speed of the connections between those systems matters more.

Relativity Networks is pushing hollow-core fiber as one possible answer. The company says the technology can move data about 30% faster than conventional fiber by sending light through an air-filled core instead of solid glass. It has announced $22 million in early funding and a $40 million follow-on order from an unnamed hyperscale customer.

Why This Matters For AI Infrastructure

• Large AI clusters increasingly need to connect compute across bigger physical footprints.
• Lower latency can help separate data center buildings or campuses act more like one synchronized system.
• Power availability is pushing AI infrastructure toward locations where capacity exists, not only where networking is easiest.
• Networking improvements may become a practical way to expand AI compute without waiting for every workload to fit inside one dense campus.

The funding number is small compared with the biggest AI chip and data center deals, but the technical direction is notable. AI compute competition is now moving through the full stack: processors, cooling, power, cloud access, and the physical links that hold distributed systems together.

If hollow-core fiber becomes more widely deployed, it could give data center operators more room to spread infrastructure across available power and real estate while still keeping latency under control.

The larger takeaway is straightforward: AI infrastructure is becoming a geography problem as much as a hardware problem. Faster networking will matter most where compute demand is growing faster than single-site data centers can handle.