Every physical part of an AI data center, and the companies that make each one.
A reference catalog from the GPU die to the grid interconnection: 130 part categories across 11 layers, 1,199 companies (685 with listed equity), and 1,864 part–company placements. Start from a part to see who makes it, or from a company to see what it makes.
Start here: the GB300 NVL72 teardown
One rack, walked from the cabinet inward, every line mapped to parts and suppliers.
How it fits together
The four chains — power, heat, data, shell — as linked paths through the catalog.
Parts by facility scale
What a 1 MW room shares with a GW campus, and what it doesn't.
The eleven layers
Layer 1 — Compute silicon & packaging
The silicon at the heart of the build: GPUs and custom accelerators, the CPUs beside them, the foundries and advanced packaging that put them together, and the substrates, wafers, equipment and materials one step upstream. Almost every dollar of an AI data center is ultimately spent to keep this layer busy.
20 parts
Layer 2 — Memory & storage
Memory and storage attached to the accelerators and servers: HBM stacked on the package, DDR5 on the board, the interface silicon between them, and the NAND, SSDs and nearline drives that hold checkpoints and training data.
10 parts
Layer 3 — Networking & interconnect
How chips, servers and sites talk to each other: the scale-up fabric inside a rack, the switch ASICs and systems of the scale-out network, the optical transceivers and copper cables between them, and the fiber and coherent optics linking buildings and campuses.
13 parts
Layer 4 — Servers, racks & integration
The server and the rack it lives in: rack-scale systems and their ODMs, GPU baseboards and trays, the high-layer-count PCBs, connectors, VRMs and passives on every board, plus power shelves, busbars, PDUs and the sheet metal that holds it together.
17 parts
Layer 5 — Liquid cooling (IT-side: chip to CDU)
Liquid cooling on the IT side, from the chip to the coolant distribution unit: cold plates, manifolds, quick disconnects, hoses, CDUs, the pumps and plate heat exchangers inside them, the fluids in the loop, and the sensors watching for leaks.
11 parts
Layer 6 — Facility cooling & HVAC
Facility cooling and HVAC: chillers, dry coolers and cooling towers, CRAH and fan-wall units, the facility water loop with its pumps and valves, water treatment, heat reuse, building controls, and fire detection and suppression.
9 parts
Layer 7 — Electrical: power distribution (grid to rack)
Electrical distribution from the grid to the rack: large power transformers and HV switchgear, MV/LV distribution, busway, UPS and batteries, backup generators and their controls, and the emerging 800 V DC architecture.
16 parts
Layer 8 — Power generation & energy supply
Where the power comes from: gas turbines and reciprocating engines behind the meter, fuel cells, nuclear PPAs and SMRs, the utilities and IPPs signing data-center deals, renewables, storage, transmission, and the gas supply chain.
10 parts
Layer 9 — Building shell, site & construction
The building and the site: developers, operators and REITs, the hyperscalers and labs that fill them, the architects, contractors and MEP trades that build them, structure and envelope materials, site utilities, security, and commissioning.
10 parts
Layer 10 — Materials & upstream inputs
Raw materials several steps upstream that AI demand still moves: copper, aluminum, electrical steel for transformer cores, rare-earth magnets, specialty glass, refrigerants, industrial gases, and export-control-sensitive critical minerals.
8 parts
Layer 11 — Software, tools & services touching the hardware (optional)
Software, tools and services that touch the hardware: DCIM and digital twins, cluster management, site selection and market intelligence, the capital behind the buildout, test instruments, and operations robotics. Optional context, not parts.
6 parts