Learn
What the parts do and how they fit together, written for someone meeting this industry for the first time. Every part name links into the catalog. For the condensed version, seehow it fits together; for one real rack bolt by bolt, theNVL72 teardown.
- How power reaches the chip
The path electricity takes from the grid to a GPU die, and why 800 V DC is starting to change it.
- How heat leaves the building
Why AI racks moved from air to liquid cooling, and the path heat takes from a GPU die out to the atmosphere.
- How data moves inside an AI cluster
Scale-up vs scale-out, copper vs optics, and why the copper spine inside a rack is running out of road.
- From sand to substrate: how an AI chip gets built and packaged
Wafer, foundry, packaging, HBM stacking, substrate, test — why the constraint on how many AI accelerators ship has moved downstream of the die.
- The slow parts: transformers, turbines, and the queue
Servers ship in months. The substation transformer, the onsite turbine, and the grid queue behind them move on a completely different clock.
- Reading this catalog like an investor
What parts-first means, how confidence tags and sole-supplier flags work, why breadth isn't revenue exposure, and how to use the cross-cutting views.