Hollow-core fiber could let data centers spread farther apart without slowing down

Key points

• Relativity Networks announced $22 million in SAFE note funding on Tuesday, drawn from investors including Rhapsody Venture Partners and Bell Ventures Inc.
• The company also secured a $40 million follow-on order from an unnamed major cloud operator.
• Hollow-core fiber transmits data 30% faster than conventional fiber-optic cable by routing light through a near-vacuum instead of glass.
• Data center campuses now commonly sprawl across hundreds of acres, making fiber speed a growing bottleneck for AI workloads.
• A 30% speed gain translates directly into a 30% wider physical footprint before latency becomes a problem.

The numbers behind AI data centers are staggering. Developers are expected to spend up to $4 trillion building them before 2030, first reported by TechCrunch AI. Yet land, power, and politics already limit where those buildings can go. Relativity Networks, a startup most people have never heard of, thinks the answer to that crunch is a different kind of cable.

On Tuesday the company announced $22 million in SAFE note funding. A SAFE note is a type of investment that converts into company shares once a formal funding round is priced, making it a common tool for early-stage startups. The round was led by Rhapsody Venture Partners, Bell Ventures Inc., and Faster Than Glass LLC, among others. Separately, an unnamed large cloud operator placed a $40 million follow-on order for the company's product.

What is hollow-core fiber, and why does it matter?

Hollow-core fiber is a cable that sends light through an air-filled channel rather than through solid glass. That single change makes a measurable difference in speed.

Conventional fiber-optic cables, the kind that already carries most of the world's internet traffic, push light through a strand of glass. Light travels slightly slower inside glass than it does through open air or a vacuum. Hollow-core fiber threads light through a near-vacuum running down the centre of the cable, bringing it much closer to the theoretical top speed of light.

Relativity Networks CEO Jason Eisenholz puts a concrete number on it. A signal takes roughly five microseconds, or five millionths of a second, to travel one kilometre in standard fiber. Hollow-core cuts that to about three and a half microseconds. That is a 30% reduction.

Why would a fraction of a second matter?

For everyday browsing, it probably would not. But AI training runs on GPUs, the specialised chips that do the heavy number-crunching AI needs, and those chips have to stay in constant, tight communication with each other.

A decade ago, a single rack of GPUs could handle an entire AI job, and all those chips sat close together. Today, training the largest AI models requires spreading thousands of GPUs across multiple buildings, or even across separate campuses connected by fiber links. Every extra microsecond of delay between chips means those systems fall out of sync.

Eisenholz frames it plainly: "They're moving to where the warm shell is, but they still need to operate as one synchronized machine." Hollow-core fiber, he argues, gives operators a larger margin. A 30% speed gain means engineers can place campuses 30% farther apart before latency becomes a hard barrier.

"The first era of AI optimized for compute," he said. "It was GPU, GPU, GPU. The second era optimized the networking inside the data center. The third era that we see coming is optimizing the geography."

For ordinary people, the practical effect is indirect: faster and cheaper AI infrastructure tends, over time, to produce faster and cheaper AI products. The more immediate audience for Relativity's pitch is the handful of cloud giants building those campuses right now.

What should readers watch for?

Hollow-core fiber has been around in research labs for years but has seen almost no commercial deployment. Relativity Networks is betting it can change that. Watch for whether the unnamed hyperscaler customer goes public with a deployment, and whether the claimed 30% speed improvement holds at real-world scale outside a controlled test environment.