Hyperscalers: AI bottlenecks

Hyperscalers is editorial research for the AI infrastructure trade. In 2026, roughly $340B will flow through the AI infrastructure stack. It doesn't end at Nvidia. Capex cascades down a deep waterfall: from accelerators to networking, memory, packaging, foundries, materials, and the power that runs it all. Hyperscalers maps that waterfall so you can see, layer by layer, where the dollars actually go. What's inside • Capex waterfall — an interactive breakdown of every layer Nvidia, Broadcom, and AMD silicon depends on: HBM, CoWoS, optical interconnect, substrates, wafer supply, EUV, power, cooling. • Bottleneck explainers — short editorial briefings on why each layer matters and who supplies it (TSMC, Samsung, SK hynix, Micron, ASML, Marvell, Coherent, Vertiv, Eaton, and more). • Spend dollar — pick any amount and see how it propagates through the stack. • Ticker detail — drill from a public ticker into the bottlenecks downstream of its revenue. How it differs This is not a charting app and not financial advice. Hyperscalers is a research tool for thinking about how AI infrastructure spending decomposes — useful for analysts, traders, and engineers building a mental model of the supply chain. Pro subscription A subscription unlocks every layer past Accelerators ($25B). Pricing: $19.99/week, auto-renewing weekly. Cancel anytime in Settings. Privacy policy: hyperscalers.lilinsystems.xyz/privacy. Terms: hyperscalers.lilinsystems.xyz/terms. Editorial research tool. Not financial advice. Always do your own diligence before acting on any data shown.