Copper is running out of reach inside the AI rack. The industry's response — an open optical interconnect standard backed by the largest compute and hyperscale names in the business — needs a name as direct as the architecture it describes. OpticalRacks.com is that name.
As AI accelerators push toward multi-terabit signaling, electrical links lose integrity over distance and burn disproportionate power compensating for it. That constraint doesn't sit at the edge of the network anymore — it sits inside the rack.
Rack-level AI power density is climbing toward megawatt-class deployments within the next several product generations, compounding the pressure on interconnect efficiency alongside power delivery and cooling.
Every additional watt burned pushing a signal through copper is a watt not spent on compute — and every additional meter of reach lost to copper is a rack topology constraint. The industry's answer is not a faster copper spec. It's a different physical layer.
In March 2026, the Optical Compute Interconnect (OCI) Multi-Source Agreement group formed to define an open, multi-vendor optical physical layer for scale-up links inside AI racks — replacing copper where it has stopped being viable.
Founding members of the OCI MSA — chip designers and the hyperscalers building on top of them, aligned on one open specification rather than competing proprietary links.
| Signal | What it means |
|---|---|
| Protocol-agnostic PHY | Designed to sit under both NVLink and UALink — vendor-neutral by construction |
| Gen 1 → Gen 2 roadmap | Wavelength-multiplexed links scaling from ~200 Gb/s to ~800 Gb/s per fiber, direction-dependent on generation |
| Silicon-centric optics | Shift from pluggable modules toward co-packaged, chip-integrated optical I/O |
| Scope | Intra-rack ("scale-up") and inter-rack ("scale-out") — the whole rack-to-rack fabric |
Independent of the MSA, optical component suppliers including Broadcom and Ayar Labs have been publicly building toward the same architecture — optics as a core rack-level primitive, not an accessory.
Once compute, power, cooling, and interconnect are engineered as one rack-level system, that architecture doesn't stay inside the data center. It's the same design unit showing up everywhere physical AI needs dense, low-latency compute close to the point of action.
The original OCI MSA use case — hyperscale training and inference racks hitting copper's ceiling first.
On-board compute racks for humanoid platforms face the same density-per-watt-per-reach math at a smaller scale.
World-model and simulation workloads pushing rack-adjacent inference out of the cloud and onto the factory floor.
Self-driving compute stacks are, architecturally, a rack shrunk into a chassis — same interconnect ceiling, tighter envelope.
No metaphor to explain. "Optical" + "racks" is the plain-language description engineers and buyers already use for this shift.
Not tied to one company's brand name or one product line — usable by a standards body, a component maker, an integrator, or a media brand covering the space.
Acquired ahead of, or in step with, the OCI MSA's public rollout — before the terminology fully commoditizes into generic industry vocabulary.
Comps-anchored estimate based on recent activity in adjacent AI-infrastructure and optical-interconnect namespaces. Preliminary — a full valuation memo with named comparables is available on request. Serious offers considered outside this range.
OpticalRacks.com is available for acquisition. Reach out directly — no broker markup, no bidding theater.