d-Matrix to plug inference chips into Nvidia racks via NVLink Fusion

Chip startup d-Matrix will adopt Nvidia's chip-linking technology so that its processors can sit directly inside the semiconductor giant's data-center systems, the company said on Thursday, as demand for artificial intelligence continues to climb.

AI workloads have increasingly moved from training models to running them in everyday use, a phase known as inference. Nvidia's expensive graphics processors remain dominant in training, while d-Matrix focuses on inference.

The startup's new chips, named Raptor, will connect to Nvidia server racks through NVLink Fusion, a technology that provides connectors and specialised memory so custom AI chips can be integrated into Nvidia's broader data-center systems. Racks compatible with Nvidia are expected to be available in 2027, and the Raptor chips are due to finish their final design stage by the end of this year.

According to d-Matrix, the combined systems are intended for fast, low-latency AI services such as coding assistants, chatbots and voice agents, where response speed is critical. The company did not disclose the financial terms of the collaboration.

The Santa Clara, California-based startup is also working with connectivity firm Astera Labs to build custom solutions aimed at ensuring rapid data flow across the system.

Microsoft has backed d-Matrix since its $110 million financing round in 2023. The startup shipped its first AI chip in November 2024 and was valued at $2 billion when it raised $450 million last year.