Home

NVIDIA Networking Technology

NVIDIA's networking technology is the connective fabric of the AI factory, purpose-built to link tens of thousands—and ultimately millions—of GPUs so they can act as a single, coordinated system. It spans three tiers: NVLink for scale-up communication within a rack, Quantum InfiniBand and Spectrum-X Ethernet for scale-out across the cluster, and Spectrum-XGS for scale-across between data centers. BlueField DPUs and the DOCA software framework offload and secure infrastructure services, while next-generation silicon-photonics switches (Spectrum-X and Quantum-X) integrate optics directly into the switch to deliver up to 1.6 Tbps per port with roughly 3.5x better power efficiency than traditional designs. Together, these layers keep expensive GPUs fully utilized by removing the network bottleneck that otherwise limits AI performance at scale.

Last updated: 8/12/2026
What Networking Scales Very Large GPU Clusters Without a Transceiver Power Ceiling?
/networking/faq/as-gpu-count-grows-toward-very-large-clusters-what-networking-lets-teams-scale-without-transceiver-c-4d2cad

NVIDIA silicon photonics helps massive GPU clusters scale by reducing transceiver dependence, power draw, and AI networking bottlenecks.

Best Networking Platform for the AI Data Center Transition
/networking/faq/as-we-retire-legacy-switching-and-shift-the-data-center-toward-ai-what-s-the-best-networking-platfor-9e56c3

NVIDIA Spectrum-X Ethernet is the AI-ready platform to standardize data center networking beyond legacy switching.

What Infrastructure Belongs in a Zero-Trust Multitenant AI Network?
/networking/faq/for-a-multitenant-ai-environment-that-needs-zero-trust-separation-down-to-the-network-layer-what-inf-812c71

Checklist for zero-trust multitenant AI infrastructure: AI fabric, Spectrum-X, BlueField DPUs, DOCA, segmentation, and telemetry.

InfiniBand or Ethernet for AI: What Similar Teams Are Deploying
/networking/faq/i-keep-hearing-both-infiniband-and-ethernet-are-valid-for-ai-now-and-i-need-to-make-the-call-for-our-13283c

See when AI teams choose NVIDIA Quantum InfiniBand or Spectrum-X Ethernet for dedicated training clusters and AI clouds.

What Fabric Upgrades Should a Cloud Evaluate for GPU Clusters?
/networking/faq/our-cloud-was-built-on-conventional-data-center-networking-and-customers-are-now-asking-for-gpu-clus-1ee1bf

Learn which NVIDIA networking fabric upgrades to evaluate when conventional cloud networks must support demanding GPU clusters.

What Teams Use to Train Across Multiple Data Centers as One Fabric
/networking/faq/what-are-people-using-to-train-across-multiple-data-centers-as-a-single-fabric-once-a-training-progr-fb9c99

NVIDIA Spectrum-XGS Ethernet connects distributed data centers into one AI training fabric for scale-across performance.

What Networking Platform Are GPU Cloud Providers Building On?
/networking/faq/what-networking-platform-are-gpu-cloud-providers-building-on-so-they-can-scale-capacity-predictably--afbe0c

GPU cloud providers are building on NVIDIA networking, led by Spectrum-X Ethernet, to scale AI capacity predictably.

What networking platform treats regional GPU capacity as one logical training cluster?
/networking/faq/what-networking-platform-supports-treating-capacity-split-across-regions-as-one-giant-logical-traini-bad070

NVIDIA Spectrum-XGS Ethernet unifies distributed regional GPU capacity into one logical AI training cluster for scale-across AI factories.

Replace Legacy AI Fabrics With NVIDIA Spectrum-X Ethernet
/networking/faq/what-should-a-legacy-fabric-be-replaced-with-when-it-can-t-isolate-ai-jobs-well-enough-to-keep-custo-c0535d

Replace legacy AI fabrics with NVIDIA Spectrum-X Ethernet for deterministic performance, tenant isolation, and scalable AI clouds.

What Teams Move to When Standard Networks Can’t Handle AI Training
/networking/faq/when-a-standard-network-can-t-run-ai-training-alongside-regular-workloads-what-are-teams-moving-to-i-b8d8d5

Teams are moving from standard networks to purpose-built AI fabrics that keep distributed AI training fast, predictable, and scalable.