A critical assessment of the $22M SAFE promising light that moves ~47% faster through air instead of glass — led by Rhapsody Venture Partners, with a $40M hyperscaler order attached. The physics is real. The moat is the problem.
Effectively, yes. Microsoft acquired hollow-core leader Lumenisity in 2022, runs a dedicated HCF plant in the UK, and has stated plans to deploy ~15,000 km of HCF across Azure. The single biggest buyer of inter-data-center fiber has vertically integrated the exact technology Relativity is selling.
No. Light travels ~1.47× faster in air than in silica — a physics constant every HCF maker inherits. Relativity’s own materials quote it as 30%, 47%, and “nearly 50%” in different places. Same ceiling, no disclosed edge in loss, splicing, or cost.
Prysmian — the world’s largest cable maker — draws and cables every fiber, and is free to serve rivals. Relativity is a design-and-coupler layer on someone else’s factory, funded by a $22M SAFE that does not pay for a fab.
Key Finding: Relativity has real, credentialed hollow-core science and a genuine AI-capex tailwind. But the “47% faster” advantage is a shared physics ceiling, not a proprietary breakthrough — and the incumbent who got there first, Microsoft, is simultaneously the archetypal customer. A $22M SAFE with no priced valuation, one unnamed anchor customer, and 100% manufacturing dependency on Prysmian is a thin foundation for out-executing a $3T competitor.
Relativity frames latency as a “Propagation Tax” on distributed AI — “AI is no longer scaling inside a data center. It is scaling across geography.” The chain below is genuine light physics. It is also available to anyone who makes hollow-core fiber.
Conventional fiber sends light through silica (index n≈1.468), slowing it to ~68% of vacuum speed. ~5 µs per km.
Hollow-core fiber guides light through an air/vacuum microstructure — near vacuum speed.
The 1.47× index ratio yields ~47% higher speed / ~30–32% lower latency. A fixed physics ceiling.
Latency drops from ~5 to ~3.5 µs per km — the same figure Lumenisity published years ago.
Extends inter-site reach from ~60km to ~90km within one latency budget.
The pitch: let operators spread AI training across campuses. True only for latency-tolerant workloads.
Every step above is correct — and none of it is proprietary. Relativity’s differentiation is incremental: cable density (a claimed 24 fibers in a 10mm cable with Prysmian), packaging, couplers, and go-to-market. The foundational advantage belongs to physics, and the manufacturing belongs to Prysmian.
The identical “47% faster” headline was Lumenisity’s — the hollow-core company Microsoft acquired in December 2022. Microsoft now runs the world’s first dedicated HCF plant (Romsey, UK), has announced the lowest signal loss ever recorded, signed Corning + Heraeus supply deals, and stated plans to deploy ~15,000 km of HCF across Azure. Relativity’s entire market — hollow-core fiber for hyperscaler campuses — is one a $3T incumbent is already building in-house. The biggest buyer is also the competitor.
Strip away the physics everyone shares, and three dependencies define what Relativity actually controls — and what it doesn’t.
Microsoft (via Lumenisity) has the same physics, in-house, at hyperscale — and is the archetypal customer for inter-DC fiber. Corning and Heraeus supply the preforms. The one unnamed hyperscaler behind the $40M order could become an in-house program at any time.
Prysmian draws and cables every fiber. It is the world’s largest cable maker and is free to serve any HCF designer. Relativity carries single-supplier risk with a thin capex moat — and $22M does not fund a fab.
Splicing hollow-core to standard fiber still adds 0.3–2 dB loss (vs <0.1 dB for glass). Connectors, test gear, and repair ecosystems are all built for silica. Scaling precise microstructure fabrication economically is an industry-wide, unsolved challenge.
Financing note: A $22M SAFE means no institutional lead set a price via a Series round — there is no disclosed valuation or cap. One named investor, “Faster Than Glass LLC,” is titled after the product and reads like an insider/SPV vehicle we could not independently identify as a conventional fund. Flag, not proof — but worth noting against a $500M+ peer set (Ayar Labs) competing for the same optical-AI-infrastructure capital.
Seven structural risks the $22M SAFE does not resolve.
The dominant HCF player owns the technology in-house at hyperscale, deploying ~15,000 km on Azure — and is also the archetypal customer Relativity is trying to sell.
The “47% faster” edge is a shared physics ceiling every HCF maker inherits. No disclosed differentiator in loss, splice quality, or cost per km.
No owned fab capacity; both fiber draw and cabling happen in Prysmian plants. Single-supplier risk and a thin capital moat in a capital-intensive business.
The $40M is one unnamed hyperscaler’s order off a two-DC pilot — an order, not recurring or booked revenue, with no disclosed contract term.
Independent analysts note 30–40% latency cuts are “not necessarily game-changers” for most workloads; even HCF is too slow for the most latency-critical apps.
$22M SAFE with no priced valuation; a product-named “Faster Than Glass LLC” investor; and a speed claim that shifts across the company’s own materials (30 / 47 / 50%).
CEO Eichenholz co-founded Luminar, which SPAC’d near a $54B peak and now sits near bankruptcy — though that fallout centers on Austin Russell, who exited after an ethics inquiry, not Eichenholz, who had already left. No wrongdoing found; relevant only as a pattern-of-hype caution for a founder now making bold AI-infra claims.
Relativity Networks is selling real physics into a real tailwind — against the one competitor who already owns both. The hollow-core science is genuine and the founders are credible, but the “47% faster” edge is a ceiling every HCF maker shares, the factory is Prysmian’s, and Microsoft has spent three years and a Lumenisity acquisition building the same thing in-house for Azure. The key diligence question is defensibility: what, exactly, does a $22M design-and-coupler layer own that a $3T incumbent and the world’s largest cable maker cannot simply out-scale?
Based entirely on publicly available information, including the TechCrunch announcement of August 19, 2026. Numbers labeled CONFIRMED / DERIVED / EST throughout; company-claimed figures are never presented as independently verified.
