Trace the flow of AI capex backwards and the whole chain lines up. It starts with demand. OpenAI has committed to 10GW of NVIDIA systems, 10GW of Broadcom custom accelerators (~$350B) and $250B of Microsoft Azure, while Anthropic has locked roughly 10GW across Google TPUs and Amazon Trainium. These commitments are the first driver of all upstream revenue.
The money flows first into infrastructure. Foxconn and Super Micro assemble NVIDIA GB200 racks for hyperscalers, and Vertiv supplies the power and liquid cooling that handle 100kW+ per rack. Coherent's optical transceivers connect those racks.
Next come the chips. NVIDIA, Broadcom, Marvell, Alchip and GUC design them, and Synopsys, Cadence and Siemens EDA sell the design tools. TSMC manufactures, while SK Hynix, Micron and Samsung supply HBM memory.
Further upstream sit equipment and materials. ASML, Applied Materials, Lam Research and Tokyo Electron make the tools, MKS makes the critical subsystems inside those tools, and Photronics supplies the lithography masks. At the very end of the chain, the apex of EUV, sit just two companies: Carl Zeiss SMT, which makes the mirrors, and TRUMPF, which makes the light-source laser.
From demand (OpenAI, Anthropic) to the deepest supply (Zeiss, TRUMPF), the value chain is unbroken. That share prices moved along this order with a lag is shown in a separate piece. The use of this map is simple: when demand or a bottleneck anywhere is disturbed, you can trace in advance where the shock will propagate.