Hyperscaler CapEx going back to 2020 — MAG, MANGOS, neoclouds, and the four public hyperscalers.
Hyperscaler spend is capital expenditure by the large cloud and AI buyers that funds server, accelerator, and memory demand. MAG is Microsoft, Amazon, and Google. MANGOS adds other large spenders (Google is listed in the name but counted once). Neoclouds are the GPU-centric hosts. Company rows show AWS, Google, Microsoft, and Meta, plus research estimates for newer AI labs.
Rising CapEx pulls demand through the stack: more dollars for AI infrastructure mean more CPU, GPU/ASIC, and HBM. If supply coverage is already below 100%, a CapEx surge tightens the market further — longer lead times and stronger vendor pricing. Flattening or falling CapEx eases component demand and is usually the first sign that shortages will loosen.
HyperScalerCapex: MAG, MANGOS (Google excluded from the MANGOS sum), and neoclouds. Company rows below.
Download PNG · Through 2025
AWS, Google, Microsoft, and Meta link CapEx_Annual. Anthropic / OpenAI / SpaceX are yellow estimates.
Download PNG · Through 2025