Nvidia is expanding its AI server ecosystem by bringing chip startup d-Matrix into its NVLink Fusion platform. The move will allow d-Matrix to connect its upcoming AI inference processors directly with Nvidia’s data center infrastructure.
d-Matrix plans to use Nvidia’s NVLink Fusion technology with its next-generation Raptor chips. Raptor is being developed mainly for AI inference, which is the process of using a trained AI model to produce answers, images, recommendations, or other results. This type of computing is becoming increasingly important as companies deploy more AI services such as chatbots, voice assistants, and AI agents.
NVLink Fusion gives companies a way to connect their own processors with Nvidia’s wider AI infrastructure. It is designed for semi-custom AI systems where different CPUs, GPUs, and specialized AI accelerators can work together using Nvidia’s high-speed interconnect technology. This gives cloud providers and AI companies more flexibility when building large data center systems instead of relying on only one type of processor.
For d-Matrix, the partnership could make it easier to bring its specialized inference technology into large-scale Nvidia-based server environments. The company’s Raptor architecture focuses heavily on memory performance, placing compute closer to memory to reduce the amount of data that must move between different parts of the system. This approach is designed to improve speed and energy efficiency for demanding generative AI workloads.
d-Matrix is also working with Astera Labs on technology that will help move data quickly across the system. The final design of the Raptor chips is expected to be completed by the end of 2026, while servers using the technology are expected to become available in 2027. Financial details of the Nvidia and d-Matrix arrangement have not been disclosed.
The collaboration shows how Nvidia is making its AI infrastructure more open to specialized processors from other companies. Instead of competing only through its own GPUs, Nvidia is building an ecosystem where custom AI chips can connect to its networking, server, and rack-scale technologies.
This could become increasingly important as AI workloads become more specialized. Training large AI models remains highly demanding, but inference is also growing quickly as businesses put those models into everyday applications. Specialized chips such as d-Matrix’s Raptor may give companies more choices for improving performance, power efficiency, and cost while still using Nvidia’s broader data center infrastructure.
