Nvidia is up about 12 percent year to date in 2026. By the standards of any other company that would be a strong year. By the standards of the AI trade it is a curiosity, because the PHLX Semiconductor Sector index is up 74 percent over the same window, and several smaller names tied to Nvidia's own supply chain have been running at multiples of that. The interesting question is no longer whether AI infrastructure spend is real. It is which seat in the supply chain has the most pricing power once the GPU itself becomes a commodity input.
The clearest example is Lumentum Holdings, up 121 percent on the year. Lumentum makes the lasers, transceivers and optical amplifiers that move data between GPUs inside a data center. Its first nine months of fiscal 2026 brought just over 2 billion dollars in revenue, up 72 percent year over year, and the company has guided for 985 million in the current quarter alone, more than double the same period last year. The reason the numbers are accelerating rather than tapering is structural: as model training runs scale across thousands of accelerators, the bottleneck stops being compute and starts being the speed at which data moves between chips. The optical interconnect market is forecast to compound at 21 percent annually through 2029 to roughly 30 billion dollars, and Lumentum notes that its data-center products carry higher margins than its legacy lines.
Applied Materials sits one rung further upstream and is up 67 percent. The company sells the equipment used to fabricate chips, plus the software to optimise it. Fiscal Q2 revenue rose 11 percent to 7.91 billion, with earnings per share up 20 percent, and management is guiding to a 23 percent revenue jump and 36 percent earnings jump in the current quarter. The reasoning is mechanical: leading-edge foundry equipment, DRAM and advanced packaging are expected to drive 80 percent of wafer-fab equipment growth this year, and Applied has named partnerships with TSMC, Micron, SK Hynix and Samsung pointed exactly at those segments. None of that is secret. What it shows is how the market is now willing to pay for a vendor whose customers are themselves expected to overbuild for AI.
Nvidia's own portfolio tells the same story in miniature. In the first quarter the company put 1.9 billion dollars more into CoreWeave, the GPU cloud operator, lifting its position to about 9 percent of the company. And in March it announced a non-exclusive agreement with Coherent that combined a multi-billion-dollar purchase commitment for silicon photonics, used in Nvidia's Spectrum-X switches, with a fresh 2 billion dollar investment in R&D. Coherent stock is up nearly 370 percent over the past year. Nvidia, in other words, is using its cash to lock in capacity from the parts of its own supply chain where pricing power is rising fastest, which is also an implicit admission of where the leverage actually sits.
There are obvious caveats. CoreWeave's debt-to-equity sits around 5.2, its total-liabilities-to-equity at 10.6, and the share count has more than doubled in a year on convertible offerings. If the AI capex cycle wobbles, the most leveraged operators get hit first and hardest. Lumentum at 56 times forward earnings, and Coherent at 70, are priced as if the optical-interconnect story compounds for years without a flat patch. These are not safe stocks. They are leveraged bets on the assumption that hyperscalers keep ordering at current rates and that the bottleneck stays in the same place.
The interesting takeaway is editorial rather than investment. For two years the public conversation has treated AI infrastructure as a synonym for Nvidia. In 2026 the market is starting to price it as a stack, and the parts of the stack that move data and make the chips are doing better than the part that designs them. Nvidia is not losing. It is being treated, finally, as one component in a layered system, which is what an industry looks like once the early monopoly phase ends.