{
 "id": "1.06",
 "slug": "nics-supernics",
 "url": "/parts/compute-silicon-and-packaging/nics-supernics",
 "name": "NICs / SuperNICs",
 "qualifiers": null,
 "layer": {
  "id": 1,
  "name": "Compute silicon & packaging"
 },
 "parent_id": null,
 "children": [],
 "summary": "Network interface cards, including Nvidia's higher-bandwidth \"SuperNIC\" variants, that connect a server's CPU and GPUs to the cluster network.",
 "description_md": "NICs move data in and out of a server node; SuperNICs are a higher-bandwidth variant built specifically to keep pace with GPU-to-GPU traffic in an AI cluster rather than conventional data center networking loads. They sit between the server CPU and GPUs (Server CPUs, Merchant GPUs / accelerators) on the board and the switches and cabling in Layer 3, and their bandwidth and port count are set largely by what the GPU generation on the same board needs to feed and be fed at.\nSupplier concentration here is meaningful — a small number of vendors, several of them also selling the GPUs themselves, supply the bulk of AI-cluster NIC and SuperNIC silicon — which gives an investor a fairly direct read-through from GPU shipment trends to NIC demand without needing a separate forecast. The qualification bar is set by interoperability with the switch and optics layer above it (Layer 3) as much as by the NIC silicon alone, so generational transitions tend to move in lockstep with the broader cluster networking refresh rather than on an independent cycle.",
 "ai_delta": "Training and inference workloads depend on moving large tensors between accelerators at high bandwidth and low latency, which pulls NIC and SuperNIC speed and count up in step with cluster and GPU generation size.",
 "bottleneck_status": null,
 "scale_tiers": [
  "enterprise",
  "hyperscale_building",
  "ai_campus",
  "gw_class"
 ],
 "density_modes": [
  "any"
 ],
 "scale_notes": "At least one NIC ships per server regardless of scale; port count and speed step up as cluster size and GPU generation move from enterprise toward campus- and gigawatt-class builds.",
 "flow": [
  "compute"
 ],
 "companies": [
  {
   "company_id": "amd",
   "name": "AMD",
   "role": "manufacturer",
   "confidence": "inferred",
   "ownership_type": "public",
   "ticker": "AMD",
   "product_note": "Pensando/Pollara",
   "note": null,
   "source_url": null
  },
  {
   "company_id": "broadcom",
   "name": "Broadcom",
   "role": "manufacturer",
   "confidence": "inferred",
   "ownership_type": "public",
   "ticker": "AVGO",
   "product_note": null,
   "note": null,
   "source_url": null
  },
  {
   "company_id": "intel",
   "name": "Intel",
   "role": "manufacturer",
   "confidence": "inferred",
   "ownership_type": "public",
   "ticker": "INTC",
   "product_note": null,
   "note": null,
   "source_url": null
  },
  {
   "company_id": "marvell",
   "name": "Marvell",
   "role": "manufacturer",
   "confidence": "inferred",
   "ownership_type": "public",
   "ticker": "MRVL",
   "product_note": null,
   "note": null,
   "source_url": null
  },
  {
   "company_id": "nvidia",
   "name": "Nvidia",
   "role": "manufacturer",
   "confidence": "inferred",
   "ownership_type": "public",
   "ticker": "NVDA",
   "product_note": "ConnectX",
   "note": null,
   "source_url": null
  }
 ]
}