AI infrastructure is usually discussed in terms of processors, power and cooling. Relativity Networks argues that the glass connecting machines deserves equal attention. The startup has raised $22 million through a SAFE note to commercialise hollow-core fiber, a technology it says can carry data about 50% faster than conventional optical fiber.
Rhapsody Venture Partners led the financing, with Bell Ventures and Faster Than Glass among the other participants. Relativity also says it has received a $40 million follow-on order from an unnamed leading hyperscaler. The customer’s identity and the commercial terms were not disclosed, so the order signals demand without revealing how broadly the technology has been validated in live deployments.
Light travels through an empty channel
Standard fiber guides light through solid glass. Hollow-core fiber instead sends it through a vacuum-like central channel, reducing the delay introduced by the material. Relativity chief executive Jason Eichenholz told TechCrunch that conventional fiber adds roughly five microseconds of latency per kilometre, compared with about 3.5 microseconds for his company’s product.
That difference appears small until it is multiplied across enormous computing estates. Modern AI campuses can cover hundreds of acres and contain dozens of buildings, with accelerators constantly exchanging data. Every network delay can limit how effectively those machines act as a coordinated system.
Relativity’s pitch is therefore not merely faster cabling inside one building. Eichenholz says a 50% reduction in latency could allow an operator to expand the physical span of a connected campus by roughly the same proportion while preserving response times. In principle, several nearby campuses could function more like a single computing environment.
The claim is technically intuitive, but the source does not provide independent benchmark results, installation data or comparisons under production workloads. Fiber performance is also only one part of an end-to-end network, where switches, routing and software add their own delays.
Faster links may loosen geographic constraints
Data center development is increasingly constrained by power availability, grid connections and political resistance. TechCrunch reports that developers could spend as much as $4 trillion on facilities by the end of the decade. Finding one site with enough land and electricity for an entire project can be difficult.
Lower-latency links could make distributed construction more practical. Instead of forcing all capacity into one vast location, an operator might connect facilities separated by greater distances and still treat them as a coherent cluster. That could widen the pool of viable sites and let developers assemble capacity around fragmented supplies of land and power.
Eichenholz describes this as a “third era” of data center design, in which operators optimise geography as well as buildings and equipment. The framing is the company’s vision, not an established industry transition. Whether hollow-core fiber changes siting decisions will depend on price, manufacturing scale, reliability, installation practices and compatibility with existing network equipment.
Those operational questions matter because conventional fiber has an immense installed base and a mature supply chain. A performance advantage must translate into sufficient economic value to justify changing a foundational component of critical infrastructure.
A large order raises the stakes for execution
The combination of new financing and the reported hyperscaler order gives Relativity more than a laboratory story. It now has to prove that hollow-core fiber can be manufactured, installed and supported at the volume expected by the world’s largest computing customers.
The SAFE structure means the $22 million investment will convert into equity under future terms rather than establishing a conventional priced valuation now. That provides capital for expansion while leaving the company’s eventual valuation open.
If its latency claims hold at scale, Relativity could address a genuine pressure point in AI infrastructure. Faster connections would not create power or land, but they could give developers more freedom in how those scarce resources are combined. The opportunity is substantial precisely because fiber is easy to overlook: a modest delay per kilometre can become a strategic limit when a data center grows into a regional computing network.
