The data center industry is on track to absorb as much as $4 trillion in investment before the end of this decade, yet developers face mounting constraints tied to land availability, political opposition, and power grid capacity. Most operators treat fiber connectivity as a solved problem - a reliable constant in an otherwise complicated equation. One startup, however, is making the case that fiber itself still has room to improve, and that faster transmission speeds could fundamentally reshape where large-scale computing infrastructure gets built.
Relativity Networks announced on Tuesday that it has raised $22 million through a SAFE note funding round, drawing participation from Rhapsody Venture Partners, Bell Ventures Inc., Faster Than Glass LLC, and additional investors. A SAFE note - short for Simple Agreement for Future Equity - is a financing instrument under which capital converts into a defined number of shares once the company completes its first priced funding round. The structure is widely used at the pre-seed and seed stages of startup development. Alongside the funding announcement, the company also disclosed a $40 million follow-on order from a major hyperscaler, which requested that its identity not be published.
Hollow-Core Fiber and the Physics Behind It
Relativity Networks is built around hollow-core fiber, a technology that remains rare in commercial deployments but offers a measurable performance advantage over conventional fiber-optic cable. Standard fiber-optic lines carry light pulses through glass, which slows transmission relative to what is physically possible in a vacuum. Hollow-core fiber replaces that glass core with an air or vacuum channel running through the center of the cable, allowing light to travel much closer to its theoretical maximum speed. The practical result is a transmission rate roughly 50% faster than what conventional fiber can achieve.
The gap between the two technologies comes down to microseconds - a unit of time that may seem trivial in most contexts but carries significant consequences in high-performance computing environments. According to Relativity Networks CEO Jason Eichenholz, a signal traveling through standard fiber covers one kilometer in approximately five microseconds. The same signal moving through hollow-core fiber covers that same distance in around three and a half microseconds. Over the distances involved in modern data center campuses, those fractions of a second accumulate into a meaningful latency difference.
Why Latency Matters More as AI Infrastructure Scales
For much of the early period of AI development, compute workloads were concentrated within a single rack of graphics processing units. At that scale, the latency introduced by fiber connections between components was negligible and largely ignored by infrastructure planners. That picture has changed substantially as AI models have grown larger and more demanding. Today, a single data center campus can span hundreds of acres and encompass dozens of separate buildings. The physical distance between GPU clusters has grown accordingly, and so has the sensitivity to transmission delays.
Eichenholz identifies multi-campus deployments as a particularly compelling use case for hollow-core fiber. In these configurations, separate data center facilities - often built at different times and in different locations - are connected and operated as a unified computing system. The motivation is practical: new power capacity is difficult to secure in concentrated areas, so operators are increasingly stitching together existing facilities in locations where power is already available.
"The largest systems are distributing the compute across multiple campuses to reach the power that exists. They're moving to where the warm shell is, but they still need to operate as one synchronized machine." - Jason Eichenholz, CEO, Relativity Networks
Rethinking the Geography of Compute
The latency reduction offered by hollow-core fiber has a direct effect on the geographic range within which interconnected facilities can operate effectively. Because the technology cuts transmission delay by approximately 50%, data center developers gain the ability to place interconnected buildings or campuses roughly 50% farther apart before latency becomes a limiting factor. In an industry where site selection is already heavily constrained by power availability and regulatory conditions, that additional spatial flexibility could open up locations that would otherwise be impractical.
Eichenholz frames this development as part of a broader evolution in how the AI infrastructure industry prioritizes its investments. In his view, the sector has moved through distinct phases, each defined by a different bottleneck. The first phase centered on acquiring and deploying as much raw compute power as possible, with GPU procurement as the primary objective. The second phase shifted attention to the internal networking architecture of data centers, optimizing how data moves between processors within a facility. The third phase, which he believes is now beginning to take shape, will focus on geography itself - on how the physical arrangement of infrastructure across locations can be optimized to support the next generation of AI workloads.
"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."
With $22 million in fresh capital and a substantial follow-on commitment from an unnamed hyperscale customer, Relativity Networks is positioning itself at the center of that transition. Hollow-core fiber remains a niche technology today, but the company is betting that the economics and operational demands of large-scale AI infrastructure will accelerate its adoption in the years ahead. If Eichenholz's read on the industry is correct, the cable running between buildings may soon matter as much as the hardware inside them.
