Scale-up fabric (intra-cluster GPU-to-GPU interconnect)
BOTTLENECKNVLink and NVSwitch are the only fabric in volume production that ties hundreds of GPUs into one coherent memory space, and no open alternative ships today.
NVLink and NVSwitch fabrics tie GPUs within one training domain into shared memory space. Larger coherent clusters let bigger models train faster. NVIDIA captures full-stack margin; the UALink consortium is pursuing an open alternative.
What the evidence shows
Nvidia’s NVLink is the predominant scale-up technology; 2026 is the first year credible alternatives reach market.
chipstrat.comNVLink’s proprietary nature prevents plugging an NVIDIA GPU into an AMD Infinity cluster.
intuitionlabs.aiWho controls it
Where it sits in the stack
Takes in: Switch ASIC wafers, copper/optical cables, backplanes, NVLink connector assemblies
Sends on: Unified GPU domain enabling model parallelism at scale-up cluster level
Related nodes
Full scorecard, owner shares, supply edges and the full tracked roster are in the desk letter.
GET THE BRIEFING