Nvidia, D-Matrix team up on next-gen AI inference chips

Chip startup d-Matrix has announced it will incorporate Nvidia's chip-linking technology into its processors for use in Nvidia's data-center systems. This move comes as demand for AI continues to rise, with a shift from training AI models to running them for everyday use, known as inference. While Nvidia's high-end graphics processors dominate training workloads, d-Matrix specializes in inference.

D-Matrix's new chips, named Raptor, will connect to Nvidia's server racks through NVLink Fusion, a technology featuring connectors and specialized memory that enables custom AI chips to integrate with Nvidia's larger data-center systems. The Nvidia-compatible racks are expected to become available in 2027, with the Raptor chips finishing their final design stage by the end of this year.

The combined systems from d-Matrix and Nvidia are designed for fast, low-latency AI services like coding assistants, chatbots, and voice agents, where speed is crucial. The collaboration between the two companies has not disclosed financial details. The Santa Clara, California-based d-Matrix is also working with connectivity firm Astera Labs to develop custom solutions for efficient data flow throughout the system.

d-Matrix has had the backing of Microsoft since its $110 million financing round in 2023. Following its first AI chip release in November 2024, the startup was valued at $2 billion during its $450 million funding round last year.