AI inference chip startup d-Matrix announced it will integrate NVIDIA NVLink Fusion into its next-generation Raptor XPUs. The integration enables d-Matrix to connect its specialized chips directly to NVIDIA’s scale-up NVLink and scale-out Spectrum-X networking, as well as the NVIDIA MGX rack architecture, facilitating deployment within unified AI factories.
By leveraging NVLink Fusion, d-Matrix aims to bypass the lengthy and costly process of designing custom scale-up networking, cooling, and rack infrastructure. The company plans to group its XPUs into single high-bandwidth domains that can operate alongside NVIDIA GPU platforms, including Vera Rubin NVL72 systems, while integrating Vera CPUs and ConnectX-9 SuperNICs.
According to d-Matrix co-founder and CEO Sid Sheth, the partnership offers customers a faster, lower-risk route to deploy ultralow-latency inference at scale. For the broader industry, NVLink Fusion demonstrates NVIDIA’s strategy to position its hardware platform as the universal ecosystem standard for third-party silicon.
Why it matters
Allows custom silicon startups to accelerate time-to-market by leveraging NVIDIA’s established rack, power, and supply chain infrastructure.
Demonstrates NVIDIA’s strategy to capture infrastructure lock-in even when data centers deploy non-NVIDIA compute accelerators.
Provides enterprise data centers with flexible, fungible architecture options to mix specialized inference XPUs with traditional GPUs.
Source: blogs.nvidia.com
