Who this is for: chip procurement leads, hardware engineers, and infrastructure buyers evaluating GPUs, custom AI silicon, and inference hardware. You are comparing Blackwell, MI400, and Trainium2 against real inference TCO, sizing HBM and NVLink for your workloads, weighing neocloud and bare-metal capacity against owned silicon, and owning the silicon-vs-GPU decision at the board level — and you need analysis that maps to those procurement decisions, not vendor spec sheets.
How this layer is organized
Gen α AI sorts its coverage into five layers of the AI stack — Energy, Chips, Infrastructure, Models, and Applications — using a computed taxonomy applied to every article at render time. This hub collects every piece the taxonomy classifies into the Chips layer: GPUs and custom AI silicon, inference hardware and accelerators, HBM and NVLink, TPUs, neoclouds and bare-metal compute, the silicon-vs-GPU cost math, and inference TCO by accelerator. Chips is the second-highest-commercial-priority layer in that taxonomy — after Infrastructure — which is why it gets a dedicated hub.
The article list and the count above are computed at render time from the same taxonomy rules in taxonomy.js that tag each article — there is no hand-curated selection and no traffic or popularity ranking behind the order. Pillars surface first, then pieces sort by editorial quality and recency. If a piece is missing, the taxonomy rules did not classify it here; the rules are iteratively refined.
The Chips library
8 articles in this layer. The grid below renders every one of them.
