Skip to content
Interconnect hardware for AI infrastructure

The boards behind AI infrastructure

Betelgeuse Labs designs and builds the cards, modules, and systems that turn advanced connectivity silicon into deployable, rack-ready infrastructure.

Built for
PCIe 6.0CXL 3.xEthernetUALink
Flagship

Rigel retimer add-in cards, now sampling

PCIe Gen 6 / CXL 3.x reach extension in a standard slot — SI-validated and instrumented out of the box.

See the family
Approach

Best-in-class silicon, made deployable

We integrate leading connectivity silicon so you get the right part for the workload — without running a silicon program of your own.

Our approach
Portfolio

Boards for every link in the fabric

Three problem areas, one integration discipline — from signal extension to full rack systems.

Hardware + software

Every board ships instrumented

Meissa firmware brings BMC-integrated telemetry, link bring-up, and diagnostics to every design — so the fabric is observable and manageable the day it lands in the rack.

Meet Meissa
Signal integrity

The board is the performance

At 64 GT/s the datasheet is only a starting point. Our materials, stack-up, and layout discipline are what turn silicon capability into a link that actually closes at scale.

Inside the engineering
Our thesis

AI clusters are bottlenecked by interconnect, not compute.

Everyone sells silicon into that bottleneck. Almost no one builds open-market boards and systems around it. That seam — between the chip and the rack — is where Betelgeuse Labs lives.

Why we exist
Interop-first

Validated against the silicon vendors’ own ecosystems

We test every design for standards compliance — PCIe-SIG CEM, OCP, E3.S, OAM — so your fabric stays interoperable and future-proof.

PCIe CEMCXLEthernetE3.SOCPOAM
Insights

From the lab

Notes on signal integrity, interconnect, and building deployable AI hardware.

Engineering

Why the board decides whether a PCIe Gen 6 link closes

Signal integrity at 64 GT/s is a materials-and-layout problem long before it is a silicon problem.

Read more
Perspective

The interconnect is the bottleneck, not the compute

AI clusters now stall on how fast accelerators can talk — which is exactly where a board partner adds value.

Read more
Product

Designing CXL memory boards for pooling at rack scale

What it takes to attach terabytes of fabric-addressable memory without wrecking latency.

Read more

Build the hardware behind AI

We’re a small team of board, signal-integrity, and systems engineers. If turning the newest connectivity silicon into shipping product sounds like your kind of problem, let’s talk.

Silicon-agnostic by design

We integrate the right connectivity silicon for each workload, so you are never locked to a single vendor’s roadmap.

Signal integrity is the product

At 200G+ per lane and PCIe Gen 6, the board is the performance. Our SI/PI engineering turns a datasheet into a link that actually closes.

Rack-ready, not reference-only

Manufacturable, thermally-validated, standards-compliant hardware — from schematic to volume, not lab curiosities.