AI Infrastructure Demand Widens Beyond Chips: Networking, Optics, Servers Rally

The narrative around AI capex is maturing beyond "NVIDIA is the only winner." Cisco's recent earnings revealed that AI infrastructure demand is now pulling in switches, optics, and entire networking stacks as hyperscalers build out not just training infrastructure but long-term inference capacity. This signals a multi-year capex boom that extends well beyond the GPU oligopoly.

Broadcom and AMD are capturing disproportionate share gains in the networking and switching layer. Analysts note that AMD is now supplying switches in 20-25% of deployments, meaningfully higher than historical levels. Broadcom chips face constraints in certain applications, creating a bifurcated supply chain. Memory chip makers, particularly in DRAM and NAND, are seeing insatiable demand and yet their valuations are compressing even as share prices rally, a sign that the market is pricing in a much longer capex cycle and lower future returns.

Microsoft's AI investment cycle is being misunderstood by many investors, according to recent analysis. The key question is not whether Microsoft can monetize AI but how quickly that monetization shows up in margins and earnings. Most of the company's AI spending is still pre-revenue: massive capex on data centers and infrastructure, but limited near-term earnings accretion. This has created a lag between the capex deployed and revenue recognition, keeping some investors cautious despite bullish near-term technicals.

The risk is peak capex fears resurface if hyperscalers signal pullback. But current visibility suggests the AI infrastructure build is only in its second inning. JPMorgan's upgraded Taiwan targets to 50,000 reflect confidence in the TSMC supply chain holding up for years. The debate is whether valuation multiples remain compressed (implying long cautious outlook) or re-rate higher once inference monetization accelerates.