
10 Sep 2026
NVIDIA blog: d-Matrix to connect Raptor XPUs via NVLink Fusion
NVIDIA published that inference chipmaker d-Matrix will use NVLink Fusion to connect its next-generation Raptor XPUs to NVIDIA’s AI infrastructure platform, including MGX racks and Spectrum-X.
Another specialized inference-chip vendor is now on NVIDIA’s NVLink Fusion and MGX rack path — a named way to sit custom silicon beside GPU factories. Schedule and performance stay the companies’ framing.
NVIDIA published “d-Matrix Adopts NVIDIA NVLink Fusion for Rack-Scale XPU Deployment.” d-Matrix will use NVLink Fusion to connect its next-generation Raptor XPUs — custom inference processors, chips built to run already-trained models — to NVIDIA’s AI infrastructure platform. NVLink Fusion is NVIDIA’s high-speed link for mixing those custom chips into its racks. That NVIDIA blog is the primary.
NVIDIA says the path includes NVLink scale-up, Spectrum-X scale-out, the MGX rack architecture, and the broader NVIDIA AI platform — framed as a faster, lower-risk route from custom silicon to large-scale deployment. That framing is NVIDIA’s.
Sid Sheth, d-Matrix cofounder and CEO, quoted on NVIDIA’s page: “Demand for inference is soaring, but capital, time and energy remain finite.” He said NVLink Fusion and MGX let the company integrate Raptor XPUs into a broadly deployed, liquid-cooled architecture for a faster, lower-risk path to deploy and scale ultralow-latency inference.
NVIDIA says NVLink Fusion lets silicon companies connect custom XPUs and CPUs into the NVIDIA stack. d-Matrix plans a single high-bandwidth, low-latency scale-up domain, and racks that can work alongside GPU systems such as NVIDIA Vera Rubin NVL72 for disaggregated inference — splitting the work across machines. That is NVIDIA’s description of the plan.
Planned integrations named on the same post: NVIDIA Vera CPUs, ConnectX-9 SuperNICs, BlueField-4 DPUs (data processing units), and Spectrum-X Ethernet. Treat that as NVIDIA’s description of d-Matrix’s plan — not a shipped customer deployment count. Product SKU lists are in Sources.




















