🧬 Interactive map · 01 · Build the Machine
Stack Innovation Map: Two Decades of Hardware and Software Leaps
Every layer of the AI compute stack on one timeline: power and cooling, GPUs, TPUs and custom chips, HBM memory, packaging, interconnect, systems and storage at the bottom; CUDA, compilers, kernels, frameworks, parallelism, post-training, evaluation, inference, agents and data on top. Lines join generations of the same technology; the chains view shows how a leap in one layer unlocked the next.
⚠️ About this map. A curated selection of milestones, not a complete history, compiled from vendor announcements, papers and project pages (as of 27 October 2026). Dates are announcement or release months; many hardware items cite encyclopedia pages as secondary sources (hollow dots), and items marked ⚠️ in their summary are not fully confirmed. Unconfirmed 2025–26 products are left out, and 2026 shipment dates are vendor plans unless marked otherwise. Disclosure: Claude, made by Anthropic, helps write this site; Amazon and Google are Anthropic investors and compute suppliers. Cross-layer "enables" links are our editorial reading.