IN Brief:
- AI cluster scale-up is pushing optics toward blind-mate backplanes, detachable fiber interfaces, and flatter switching fabrics.
- Molex is combining 192-fiber VersaBeam backplane links, TeraVERSE detachable chip coupling, VFI backplanes, and ELSFP laser connectivity.
- The new OCS platform takes the story beyond connectors, with a 544×544 optical switch and a roadmap beyond 1,000 ports.
Molex is widening its AI infrastructure push from cabling and connectors into something closer to a full optical systems play. The latest additions combine blind-mate backplane optics, detachable fiber-to-chip coupling, external laser connectivity, and a new optical circuit switch platform, all aimed at reducing the installation, servicing, and scaling friction that has become harder to ignore in large GPU clusters.
At the front end of that stack is the VersaBeam EBO Backplane Connector, which consolidates up to 192 fibers into a compact blind-mate interface. Molex is pairing it with Teramount’s TeraVERSE detachable fiber connection, a serviceable silicon-photonics interface designed to preserve high fiber density while making post-reflow assembly and in-field maintenance more manageable than fixed fiber attachments.
The practical point is serviceability at scale. Molex’s expanded-beam optical approach already targets lower sensitivity to dust and reduced cleaning and inspection overhead, and the company has previously pointed to deployment time savings of roughly 85% versus conventional MPO-based installation in hyperscale environments. In the newer roadmap, that service-first logic continues with VFI optical backplanes and ELSFP interconnects that move laser sources off the chip and align with the OIF’s ELSFP 2.0 work on blind-mate, field-replaceable external laser modules.
The more consequential move may be the High-Radix Optical Circuit Switch platform. Molex says the MEMS-based OCS supports up to 544×544 ports, with a roadmap beyond 1,000 ports, creating scope for flatter east-west fabrics with fewer switching tiers, fewer optical-electrical-optical conversions, and less pressure on power and thermal budgets as AI clusters expand.
Molex will demonstrate the wider optical stack and the OCS platform at OFC 2026, which makes this as much an integration statement as a product announcement. The company is positioning itself higher up the architecture stack, where AI networking is increasingly shaped by field service, optical topology, and how quickly operators can rework dense systems once they are live.
