D-Matrix joins NVLink Fusion, sidesteps custom-rack deployment headache

d-Matrix, the XPU maker focused on inference, announced yesterday it is adopting NVIDIA's NVLink Fusion to connect its next-generation Raptor processors to NVIDIA's AI platform. The move places the company on the ecosystem-partner roster with direct access to the MGX rack architecture, Spectrum-X networking, and NVIDIA's proven supply chain — instead of building that entire stack from scratch.

The incentive is straightforward. Inference demand is surging while capital, time, and energy remain finite, co-founder and chief executive Sid Sheth said. By plugging Raptor into the liquid-cooled MGX platform via NVLink Fusion, d-Matrix can integrate its silicon into infrastructure already deployed in the field, giving customers a faster, lower-risk path to ultralow-latency inference at scale. Building the XPU is only the first step; deploying at "AI factory" scale demands networking, rack architecture, power, cooling, software, and a supply chain — each layer adding cost, time, and risk.

Through NVLink, d-Matrix plans to attach its XPU to a single high-bandwidth, low-latency scale-up domain. Those racks will operate alongside NVIDIA GPU-based systems such as Vera Rubin NVL72 for disaggregated inference. In parallel, the company is integrating NVIDIA Vera CPUs, ConnectX-9 SuperNIC cards, BlueField-4 DPU units, and Spectrum-X Ethernet. Together with NVLink and MGX, the combination provides a proven foundation for dedicated inference deployments running side-by-side with NVIDIA systems inside flexible, unified AI factories.

NVIDIA's full AI-factory platform now spans Vera Rubin NVL72, Groq 3 LPX, Vera CPU racks, Vera BlueField-4 STX storage, and Spectrum-6 SPX Ethernet. The system is designed to be fully fungible — able to run any workload, model, or model architecture with the best performance-per-watt and lowest cost per token. NVLink Fusion is the key that lets customers match the right compute to each task within a shared platform, and lets silicon entrepreneurs such as d-Matrix scale performance, cut time-to-market, and reduce risk in deploying semi-custom AI factories.