Beyond Single-Mode Fiber: Rethinking Optical Switching for HCF and MCF

A physical connector, an optical path, and a logical network port are not the same thing. As hollow-core fiber, multi-core fiber, and parallel optics reshape cloud and AI infrastructure, optical switching needs to accommodate both changing media and different ways of organizing connectivity.

The fiber itself is evolving

Hollow-core fiber (HCF) guides most of the signal through air rather than glass, reducing propagation latency and offering low nonlinearity. Microsoft has deployed HCF for metro data-center interconnection in Azure, bringing it into early production networks. Azure HCF deployment; HCF transmission research.

Multi-core fiber (MCF) places multiple optical cores within one cladding, creating multiple spatial channels. NTT has demonstrated field-installed four-core MCF and is developing technologies for terrestrial and submarine networks. Field demonstration; MCF development; Further advances.

Parallel optics use multiple optical paths through interfaces such as MPO. Those paths may carry one aggregated link or separate links routed to different destinations. The connector alone does not define the network topology. MPO connector overview.

A port is not a path

In MPO/parallel single-mode (PSM) systems, parallel channels need not share a destination. A GPU or switch interface with eight 200G optical lanes presents eight paths; where the architecture supports independent links, those paths can be shuffled to different endpoints. Switching must preserve the intended lane mapping and polarity - not assume every fiber in a connector belongs to one logical link.

The same eight 100G lanes can support 8×100G, 4×200G, 2×400G, or 1×800G Ethernet, depending on equipment and configuration. Aggregating all lanes into one port is one use case, not a universal model for AI scale-up clusters.

The switching layer must support the topology - not infer it from the connector.

Multi-core fiber: density, not breakout

MCF’s near-term value is front-panel and cabling density, not routinely breaking each core out to a separate connector. Until lower-cost breakout solutions emerge, keeping cores together makes connector-level optical circuit switching (OCS) density especially important.

Flexibility beyond today’s transceivers

The Telescent Network Topology Manager (NTM) offers optionality across transceiver and fiber choices: DR, FR, LR, OCI MSA, CL, SR over multimode fiber, parallel SR, SR WDM, and BiDi. Because it establishes a passive physical path without processing the signal, it does not require a particular Ethernet rate or modulation format - though fiber, connectors, wavelength range, and optical budget must still match the application.

HCF also suggests another use: automated access to specialized monitoring equipment. Its low backscatter can require different reflectometry approaches, a challenge addressed in Microsoft and Nokia research. Physical automation could connect monitoring gear to different HCF routes under software control. HCF monitoring research.

Build for the connection, not the generation

The question is no longer simply, “How do I switch today’s optical signal?” It is, “How do I build an automated fiber layer that can support tomorrow’s optical interfaces?”

Telescent’s approach is to automate the physical connection and let the network around it evolve. Whether the application calls for aggregated lanes, a parallel-path shuffle, or dense multi-core connectivity, the fiber automation layer should support the intended topology rather than prescribe it. The NTM robotically establishes a physical fiber-to-fiber connection rather than routing the signal through an internal switching medium; once connected, the path is passive and mechanically latched.

The fiber of tomorrow may look different. The infrastructure connecting it should be ready.