MTIA
A visual systems guide to the evolution, architecture, deployment and strategic economics of Meta’s Training and Inference Accelerator.
Google TPU
From the first inference TPU to pod-scale training systems and Trillium, explained as one continuous silicon-to-software architecture.
NVIDIA Blackwell
A visual guide to dual-die GPUs, Transformer Engine, NVLink, HBM and the NVL72 rack-scale system.
High Bandwidth Memory
A visual guide to stacked DRAM, TSVs, interposers, bandwidth evolution, capacity economics and HBM4.
NVLink & NVSwitch
How NVIDIA connects GPUs into high-bandwidth domains, from point-to-point links to NVL72 and beyond.
Direct-to-Chip Liquid Cooling
Cold plates, CDUs, facility water, thermal resistance, redundancy and the economics of dense AI racks.
800VDC Power Architecture
Why higher-voltage DC is being considered for dense AI racks, with current, conversion, safety and adoption tradeoffs.
The AI Factory
A visual journey from electricity and cooling to networks, accelerators, runtimes, useful tokens and economics.
From eight flagship chapters to a connected infrastructure graph.
This release establishes the article system and visual component library. The remaining research archive can now be converted topic-by-topic without returning to generic cards.
Phase 1 · Foundation
Eight flagship visual chapters and reusable diagrams.
Phase 2 · Expansion
Hyperscaler, silicon, networking, power and cooling families.
Phase 3 · Knowledge graph
Entity relationships, guided learning and semantic discovery.