ChronoCore is designed to reduce the time required for optical signals to travel between geographically separated computing infrastructure. Conventional fiber carries light through solid glass, while hollow-core fiber instead guides it through an air-filled central region. Relativity Networks says its technology can carry light roughly 47% faster than solid-core glass fiber.
The difference becomes more consequential as AI computing infrastructure spreads across larger physical areas. Relativity Networks CEO Jason Eichenholz estimates that signals traveling through conventional fiber require about five microseconds to cover one kilometer, compared with roughly three and a half microseconds using hollow-core fiber.
That latency reduction is central to the company’s pitch to hyperscalers. AI systems increasingly depend on large groups of GPUs working together, while access to sufficient electrical power can require operators to distribute that computing infrastructure across separate sites. Greater physical separation adds propagation delay that networking equipment and software cannot eliminate.
“The largest systems are distributing the compute across multiple campuses to reach the power that exists,” Eichenholz said. “They’re moving to where the warm shell is, but they still need to operate as one synchronized machine.”
Relativity Networks describes the underlying constraint as the “Propagation Tax,” referring to latency created simply by the distance a signal must travel. The company argues that faster optical transmission can give data center operators more flexibility over how far apart computing resources can be placed while remaining within a given latency threshold.
“The first era of AI optimized for compute,” Eichenholz 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.”
The manufacturing work with Prysmian addresses another part of that strategy: fitting more hollow-core connections into infrastructure that can be installed using existing deployment methods. The companies have now packaged 24 fibers inside a single 10-millimeter cable. Testing conducted with Dura-Line showed that the cable could be installed in standard microducts.
Relativity Networks says the resulting design provides 24 low-latency optical paths between AI data centers. ChronoCore fiber is manufactured at Prysmian’s Eindhoven facility in the Netherlands before being made into cable at its Claremont, North Carolina plant. Relativity Networks handles couplers, fiber characterization and installation training in Orlando, Florida.
The combination of lower propagation latency and higher-density cabling is intended to make hollow-core fiber more practical for connecting computing resources spread across multiple sites. Rather than addressing processing performance inside an individual server or data center, Relativity Networks is targeting the physical links between facilities that still need to function as part of a coordinated AI system.
“AI is no longer scaling inside a data center. It is scaling across geography,” Eichenholz said. “The next great AI infrastructure challenge is making thousands of distributed GPUs behave like one machine, even when the power they depend on is miles apart. ChronoCore™ is purpose built for that world, giving hyperscalers the low-latency connectivity needed to scale AI wherever power is available.”
The company’s latest milestones also accompany a new white paper, The AI Geography Era, which lays out Relativity Networks’ view that physical distance is becoming a more important consideration in AI infrastructure design as computing expands across distributed campuses. The paper further develops the company’s Propagation Tax concept and its argument for treating geography as part of network architecture.
This analysis is based on reporting from PR Newswire.
Image courtesy of Relativity Networks.
This article was generated with AI assistance and reviewed for accuracy and quality.
