Relativity Networks Raises $22 Million to Improve Fiber Connectivity for Data Centers

Data center developers are expected to invest as much as $4 trillion by the decade’s end, encountering notable constraints due to political and power-grid factors that influence their construction sites. While it’s commonly thought that fiber speed remains unchanged, one company asserts that advancements in fiber technology could reshape the geographical landscape of data center construction.

On Tuesday, Relativity Networks announced it has secured $22 million through SAFE note funding, with contributions from Rhapsody Venture Partners, Bell Ventures Inc., and Faster Than Glass LLC, among others. A SAFE note allows an investment to convert into a predetermined number of shares once the firm finalizes its first priced round, serving as a standard financing strategy for pre-seed and seed rounds. Furthermore, the company has received a $40 million follow-on order from a significant hyperscaler, whose identity is not disclosed in this article.

Relativity Networks focuses on hollow-core fiber, a seldom-used technology that enables data transmission at rates 30% faster than conventional fiber optics. In contrast to traditional fiber, which transmits light through fiber-optic glass, hollow-core fiber allows light to travel through a vacuum core within the cable, nearing the theoretical limit of light speed.

The speed difference is measured in microseconds. CEO Jason Eisenholz estimates that a signal requires about five microseconds to travel one kilometer in standard fiber, a duration that can be reduced to just three and a half microseconds with hollow-core technology.

In the early days when AI processing was confined to a single rack of GPUs, fiber latency was a minor concern. However, as scaling has progressed, the physical gap between GPUs has considerably enlarged. Nowadays, it’s common for a data center campus to span hundreds of acres and multiple buildings. Eisenholz envisions substantial potential in multi-campus setups, where existing data centers are interlinked to operate harmoniously as a single unit.

“The largest systems are distributing compute across multiple campuses to leverage the available power,” he told TechCrunch. “They are moving to regions equipped with the necessary infrastructure, yet still require functionality as a cohesive, synchronized system.”

This innovation could alleviate the strict spatial requirements that have limited numerous data center initiatives. Regarding latency, a 30% decrease in time empowers developers to broaden operations over 30% longer distances before latency complications occur. As computational demands keep rising, Eisenholz believes this could signify a substantial shift for the industry.

“The first era of AI concentrated on compute,” he remarked. “It centered on GPU, GPU, GPU. The second era advanced networking within data centers to maximize that compute. The third era on the horizon focuses on optimizing geographical placement.”