The key implication is that AI infrastructure is shifting from a compute scarcity trade to a transport bottleneck trade. That usually extends the cycle because networking/optics content rises with every rack, every cluster, and every topology change, so the spend burden migrates from one-time chip purchases toward a broader bill of materials that also benefits power, cooling, and optical component vendors. The underappreciated winner is the layer that sits closest to utilization: anything that keeps expensive accelerators from idling should capture budget before incremental GPUs do.

The main risk is timing, not thesis. These businesses can look ahead of fundamentals for months, then digest sharply if hyperscaler capex pauses or topology changes reduce content per compute node. The immediate upside is already partly in the group, so the better expression is relative: names with visible bookings and diversified networking exposure should outperform if AI capex stays hot, while purely sentiment-driven AI hardware multiples are more vulnerable to any order hiccup. Watch the next 1-2 quarters of hyperscaler commentary; if large customers guide AI spend lower or networking growth decouples from accelerator shipments, the trade weakens quickly.

Contrarian take: the market may be underestimating how long the plumbing phase lasts, but overestimating how cleanly every dollar of AI capex converts into recurring network revenue. If model efficiency improves faster than cluster scale, the content-per-gigawatt thesis can plateau. That makes this a good momentum-to-quality rotation rather than a blind chase: the best risk/reward is in the supplier with the strongest backlog/visibility rather than the most extended stock chart.