The market’s newest AI tell is that investors are starting to pay up for the plumbing, not just the processor. The June 2 catalyst was obvious: Reuters reported that Marvell shares surged more than 25% after Nvidia CEO Jensen Huang called it the next trillion-dollar company at Computex. That headline looked flashy, but the deeper point was more important than the sound bite. Marvell’s role in custom AI silicon and interconnects is now being priced as core infrastructure, not a sidecar to the GPU trade.
That shift spilled straight into optics. Investing.com reported that optical names such as Coherent, Lumentum, Corning, and Ciena jumped after Huang argued that copper can only take AI clusters so far and that optics become essential as scale increases across racks and buildings. Traders have heard the “picks and shovels” story for two years, but this was a more specific message: the bottleneck is moving outward from compute into connectivity.
Europe is reinforcing that read with actual product and lab investment. Nokia said on May 21 that it launched an AI Networking Innovation Lab to test switching silicon, congestion control, telemetry, and validated multi-vendor designs under real AI training and inference workloads. That matters because it turns AI networking from a vague narrative into an engineering race. If operators are demanding proof that fabrics can survive real workloads before they buy, the vendors with test environments and deployable blueprints gain leverage.
Japan’s angle is even more tangible. Furukawa Electric said on March 12 that it had started mass production of 13,824-count optical fiber cable for hyperscale data centers, opened a new plant dedicated to ultra-high-count cable, and more than doubled production capacity versus fiscal 2023. The company explicitly tied the move to the spread of generative AI, tighter in-building fiber space, and the need to connect GPU clusters across multiple facilities. That is not a speculative slide deck. It is physical capacity being built because hyperscalers need denser optical backbones now.
South Korea adds the memory and process side of the same story. Samsung said in its April 30 first-quarter results that the memory business posted record quarterly sales by serving high-value AI demand despite limited supply, and that demand should remain strong in the second quarter as AI infrastructure expands. It also highlighted silicon photonics as part of the foundry foundation it is building. Reuters then added the macro warning on June 3: Morgan Stanley sees AI-driven memory prices creating broader “chipflation,” with the squeeze spreading from data centers into devices and margins across the wider economy.
My cautious view is that this is a healthier way to read the AI trade than endlessly stretching GPU valuation narratives. When the market starts rewarding interconnects, optics, dense fiber, switching, and memory-adjacent enablers, it usually means hyperscaler capex is becoming system-wide rather than single-component. That is bullish for the infrastructure stack, but it also raises the risk that downstream hardware buyers get squeezed while expectations for every AI-adjacent supplier start to outrun execution. The opportunity is real; so is the danger of paying peak multiples for anything with an optics label.
Risk notice: This article is for market commentary only, not personalized investment advice. AI infrastructure names can be volatile and may react sharply to capex changes, supply-chain delays, export controls, valuation resets, and broader risk-off moves, so traders should verify their own risk tolerance and independent research before taking positions.
Sources:
Reuters via Investing.com: Marvell surges after Jensen Huang’s comments (June 2, 2026)
Investing.com: Huang’s copper-vs-optics call sends optical stocks surging (June 2, 2026)
Nokia: AI Networking Innovation Lab launch (May 21, 2026)
Furukawa Electric: Mass production of 13,824-count optical fiber cable for hyperscale data centers (March 12, 2026)
Samsung Electronics Q1 2026 results (April 30, 2026)
Reuters via Investing.com: AI chipflation spreading beyond data centers (June 3, 2026)
Reddit discussion: AI chipflation thread (June 3, 2026)
原创文章，作者：financial transaction，如若转载，请注明出处：https://www.fanbi.net/archives/228
