★
Menu
Home How It's Built Watchlist The Stack
Hardware ChipsCompare Chips Servers & SystemsNetworking MemoryFoundries & Packaging Cooling
Facilities Data CentersPower ProjectsCloud & GPU Compute
Intelligence AI ModelsSoftware AI ProductsBenchmarks Research PapersDatasets Concepts
Ecosystem CompaniesStocks PeopleTimeline TrendsFuture of AI
Learn News Ask the Data AI Agents
About About us Methodology Data & API Disclaimer
Options
DATA & API

Free read-only JSON access to the site's cached data.

Dark mode

Guided View
New to chips, data centers and model jargon? We explain key terms as you browse, in plain English. Same data, with the help built in.

Expert View
You already know the stack. Just the data: clean, fast and compact, with no extra explanations. This is the default view.

Interface language
‹ Back
Field notes

The unglamorous truth of big AI: the network decides how much of your billion-dollar cluster you actually get to use. The war to follow is copper empire versus open Ethernet — NVIDIA's private fabrics against everyone else's shared standard.

Start with these three

N& NVLink & NVSwitch NVIDIA's deepest moat, cast in copper. The reason 72 GPUs can act as one. UE Ultra Ethernet The industry's collective answer, published as one open spec in 2025. 8& 800G & 1.6T Optics Where the money quietly goes — clusters consume these by the hundred thousand.

Interconnects & fabrics

9
N& NVLink & NVSwitch NVIDIA's scale-up fabric: what makes 72 GPUs one computer. IN InfiniBand The low-latency training fabric NVIDIA acquired with Mellanox. 8& 800G & 1.6T Optics The pluggable transceivers AI clusters consume by the hundreds of thousands. CP Co-Packaged Optics Light brought onto the switch package — the next power saver. PE PCI Express The universal socket: how accelerators, NICs and CPUs meet. R( RoCE (RDMA over Ethernet) The transport trick that lets ordinary Ethernet move GPU memory. SX Spectrum-X Ethernet NVIDIA's answer to losing Ethernet: an AI-tuned Ethernet of its own. UA UALink The open scale-up standard: everyone-but-NVIDIA's answer to NVLink. UE Ultra Ethernet The industry consortium rebuilding Ethernet for AI — 1.0 spec shipped 2025.

Networking companies

14
AN Arista Networks The Ethernet switch leader of the AI back-end network. ANET AM Amphenol Connectors and cable assemblies in every server, switch and rack. APH AL Astera Labs PCIe retimers and CXL connectivity inside nearly every AI server. ALAB CA Celestial AI The Photonic Fabric: optical links from any chip to any memory. CI Ciena Coherent optics linking AI data centers across cities and oceans. CIEN Cisco The networking incumbent pushing its own silicon into AI fabrics. CSCO CO Coherent One of the largest makers of the optical transceivers AI clusters consume by the million. COHR CO Corning The fiber-optic cable itself — glass demand doubled by AI campuses. GLW CT Credo Technology Active electrical cables — the copper that wires racks together. CRDO FA Fabrinet The contract manufacturer behind NVIDIA's optical transceivers. FN LI Lightmatter Photonic interconnect — moving light into the chip package itself. LU Lumentum Lasers and photonics for the optical layer of AI networks. LITE Nokia Optical transport and data-center routing — with NVIDIA as a 2025 investor. NOK TC TE Connectivity The other connector giant wiring AI racks and power paths. TEL