Next-gen AI networks may hinge on the telephone switchboard's return

Nvidia backed optical circuit switching (OCS) for higher-density AI cluster interconnects, including a $125M round for iPronics’ second-generation OCS that targets sub-ms reconfiguration times (vs ~100ms for MEMS-based systems). iPronics claims up to 720 port pairs per rack unit using silicon-photonics (vs ~8+ rack units for typical ~300-port OCS appliances), alongside Mixx Technologies’ new high-density fiber connector design. The developments are incremental but supportive for AI infrastructure buildouts, with potential 1–3% impacts to related suppliers’ sentiment and positioning.

Analysis

The cleanest beneficiaries are not the GPU names but the optics stack: component suppliers with real silicon-photonics content, dense connectors, and systems integration should see the earliest order pull, while pure compute vendors only benefit if this becomes a standard requirement rather than a niche architectural hack. That makes the revenue leverage asymmetric for COHR versus NVDA: NVDA is helping seed the ecosystem, but the economic capture from each new rack is likely to migrate toward interconnect and switching silicon before it shows up in GPU unit growth.

GOOGL is the strategic proof point. If large AI domains can be reconfigured without breaking training efficiency, Google’s advantage is less about raw TPU scale and more about utilization and resilience; that can defend cloud margins and widen the gap versus peers that still rely on rigid packet fabrics. The second-order loser is any merchant networking vendor whose value proposition is built on packet-switched radix alone; they may still win inside racks, but the inter-rack architecture mix could shift toward optical, reducing TAM growth for portions of the switching stack.

This is a 6-18 month story, not a one-day catalyst. The key risk is qualification drag: OCS has to move from lab-grade demo to production uptime, and vendor-funded investments do not equal backlog. The consensus may be underestimating how much the real bottleneck is cabling, reconfiguration, and failure recovery rather than FLOPS; however, the move is also likely overdone near term if investors extrapolate a broad rollout from a handful of design wins. Falsifiers are simple: no OCS mention in next-round hyperscaler capex commentary, or optics revenue that fails to inflect over the next two earnings cycles.

Key Decisions for Investors

• Long COHR on 5-10% pullbacks; target a 3-6 month rerating if optical design wins translate into guide-up revenue, but cut if optical segment growth does not accelerate over the next two quarters.
• Add GOOGL as a structural beneficiary; hold 6-18 months for the thesis that OCS improves TPU utilization and cluster resiliency, supporting cloud margin durability.
• Relative value: long COHR / short AMD small size. The market is likely to over-assign the upside from AI interconnect complexity to compute vendors; COHR has cleaner operating leverage to the enabling layer.
• Do not chase NVDA on this headline; treat it as ecosystem validation rather than immediate EPS upside. Re-enter only if management commentary shows OCS/optics attach starts affecting platform economics.