The premise of this wiki is that structure explains behavior. If supply relationships drive share prices, the AI capex cycle should spread along the value chain in stages. We test that against the annual returns of 100 nodes.
In 2023 the compute and design layer moved first. NVIDIA rose 239%, Meta 194%, Broadcom 105% and AMD 128%. In 2024 the foundry node TSMC led with 93%, while upstream tools, materials and wafers stalled or fell (ASML -8%, Lam Research -7%, SUMCO -48%, GlobalWafers -39%).
That upstream rose later, in 2025 and 2026. Memory and HBM peaked (SK Hynix 361% then 205%, Micron 240% then 298%, Kioxia 1082%), and Korean equipment and materials (Samsung Electro-Mechanics 714%, Wonik IPS 409%, Simmtech 393%) along with tools and test (Lam Research 140%, Advantest 196%, Applied Materials 141%) followed. Over the same period the demand-side megacaps cooled (Meta -12%, Microsoft -21%).
In short, the gains moved along the chain over one to three years, from demand and compute to design, to the foundry, and on to memory, materials and equipment. The hypothesis that value-chain structure largely explains the return rotation holds up.
Not everything is structural, of course. Intel's 263% gain owed more to a foundry turnaround than to chain propagation, and EDA lagged because its subscription model is insensitive to the cycle. The implication is clear. The parts of the chain that have not yet risen are the candidates for the next leg.