Data center developers are expected to spend up to $4 trillion by the end of the 2010s, and are already heavily constrained by both political and grid considerations on where they can build. But while most companies treat fiber speeds as a given, one company is betting that faster fiber could change the geographic calculus behind data center construction.
On Tuesday, Relativity Networks announced $22 million in funding in a SAFE note raised from Rhapsody Venture Partners, Bell Ventures Inc., Faster Than Glass LLC and others. SAFE notes transfer investment into a specific number of shares after a company has raised an initial price round and are a standard method used in pre-seed and seed rounds. The company also secured an additional $40 million order from a major hyperscaler, which declined to name the work.
Relativity Networks works with hollow-core fiber, a rarely deployed technology that can transmit data 30% faster than traditional fiber. While traditional fibers transmit light through fiber optic glass, hollow core fibers transmit the same light through a vacuum chamber at the center of the line, much closer to the theoretical limit of the speed of light.
The difference is a matter of microseconds. CEO Jason Eisenholtz estimates that it takes a signal about 5 microseconds to travel one kilometer over traditional fiber. Switching to a hollow core reduces this number to just 3.5 microseconds.
Fiber latency was easy to ignore when AI computations were being done across a single rack of GPUs, but as we scaled up, so did the physical distance between GPUs. Today, it is common for data center campuses to span hundreds of acres and dozens of buildings. Eisenholtz sees particular opportunity in multi-campus deployments, where existing data centers are consolidated to operate as a single unit.
“The largest systems spread computing across multiple campuses to maximize existing capacity,” he told TechCrunch. “They’re moving into a warm shell location, but they still need to operate as one synchronized machine.”
The result is a way to partially alleviate the harsh spatial logic that has constrained the construction of many data centers currently underway. From a latency perspective, reducing time by 30% gives developers the opportunity to travel 30% more distance before latency becomes an issue. Einholz believes this could be a major change for the industry as computing projects become increasingly large.
“The first era of computing-optimized AI,” he said. “It was GPU, GPU, GPU. The second era was optimizing the networks within the data center to take advantage of that computing. The third era that I think is coming is geo-optimization.”
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
