d-Matrix to Integrate Raptor XPUs with NVIDIA AI Factories via NVLink Fusion

On September 10, 2026, AI inference chipmaker d-Matrix announced a multi-year product-roadmap collaboration with NVIDIA. [1] Under this agreement, d-Matrix will integrate its next-generation Raptor inference XPUs into NVIDIA’s MGX rack-scale architecture using NVLink Fusion. [1] Initial rack-integrated availability of these systems is expected in the fourth quarter of 2027. [1]

According to d-Matrix, this integration into NVIDIA’s AI factory ecosystem will allow AI labs, hyperscalers, and neoclouds to provide premium-tier token services with ultra-low latency. [1] During a press briefing, d-Matrix CEO Sid Sheth noted that utilizing NVLink Fusion and MGX allows the company to integrate Raptor XPUs into a widely deployed, liquid-cooled architecture. [1] Sheth stated this approach reduces the time and risk associated with bringing their inference silicon to large-scale deployment. [1]

NVIDIA confirmed the partnership in a September 10, 2026, blog post, adding d-Matrix to its roster of NVLink Fusion ecosystem partners. [1] NVIDIA CEO Jensen Huang stated that NVLink Fusion allows partners to integrate custom silicon with NVIDIA's existing technologies, including advanced packaging, networking, and rack-scale systems. [1] Huang also noted that this integration path widens the accelerator options available for customers constructing new AI factories. [1]

This partnership highlights NVIDIA's strategy to maintain its dominance in AI infrastructure by opening its proprietary NVLink interconnect to specialized third-party accelerators like d-Matrix's inference chips. By integrating directly into the MGX architecture, d-Matrix gains a significant fast-track to enterprise and hyperscaler deployments, bypassing the need to develop custom, stand-alone rack solutions. This move could increase pressure on alternative interconnect standards, as NVIDIA consolidates its ecosystem. What the sources don't address: How the pricing and specific performance benchmarks of these hybrid d-Matrix/NVIDIA racks will compare to NVIDIA's own native inference solutions.