How compatibility will work

d-Matrix said its new Raptor accelerators will connect directly to NVIDIA server racks through NVLink Fusion. The technology combines specialized memory and interconnects, allowing custom accelerators to be added to NVIDIA’s larger data-center system. [1 · Reuters]

The company expects to complete Raptor’s final design before the end of 2026, with compatible racks expected in 2027. The financial terms of the collaboration have not been disclosed; separately, d-Matrix brought in Astera Labs to develop high-speed data pathways. [1 · Reuters]

Expert commentary

d-Matrix specializes in the computing needed to run already-trained models, known as inference. It is targeting Raptor at low-latency tasks such as coding assistants, chatbots and voice agents. [1 · Reuters]

the partnership shows that NVIDIA is prepared to retain control over server-rack architecture even when an accelerator from another manufacturer performs a particular computation. For d-Matrix, the main test will be not just chip performance but the timely release of a complete system in 2027. [1 · Reuters]

Sources

  1. Reuters — d-Matrix, Raptor and NVLink Fusion — September 10, 2026