AI chip startup d-Matrix will adopt NVIDIA’s NVLink Fusion technology to integrate its artificial intelligence processors directly into NVIDIA’s data-centre systems, as demand for computing infrastructure for AI inference continues to grow. The collaboration will allow d-Matrix’s Raptor processors to connect with NVIDIA’s larger data-centre systems and is expected to bring the companies’ combined solutions to the market in 2027.
The development comes as the AI infrastructure market increasingly focuses on inference, the process of running trained AI models to generate responses for users. While NVIDIA’s graphics processing units have traditionally played a major role in AI model training, d-Matrix has built its technology around inference workloads. Its Raptor chips are designed for applications where fast response times and low latency are important.
Under the collaboration, d-Matrix’s Raptor chips will use NVIDIA’s NVLink Fusion technology to connect directly with NVIDIA’s data-centre systems. The technology includes specialised connectors and memory designed to allow custom AI processors to work within NVIDIA’s server infrastructure. This gives chip companies such as d-Matrix a way to integrate their processors into larger NVIDIA-based systems rather than operating them as completely separate infrastructure.
The combined systems are being targeted at AI applications where speed is particularly important. These include chatbots, coding assistants and voice agents, all of which require AI models to process and respond to user requests quickly. The move reflects the growing demand for specialised hardware capable of handling inference workloads efficiently as AI applications become more widely deployed.
d-Matrix expects the final design stage of its Raptor chips to be completed by the end of 2026. The NVIDIA-compatible systems incorporating the processors are expected to become available in 2027. The companies have not disclosed the financial terms of their collaboration.
The partnership also expands d-Matrix’s work with other technology companies to develop the infrastructure required for its AI processors. The Santa Clara, California-based startup is working with connectivity company Astera Labs to develop customised high-speed data paths across the systems. The objective is to support fast movement of data between the different components of the AI infrastructure.
The focus on inference reflects a broader shift in the AI computing market. As more AI models move from development and training into everyday applications, the infrastructure required to run those models for users is becoming increasingly important. AI services such as coding tools, conversational assistants and voice-based agents can require large amounts of computing capacity while also placing a premium on response times.
d-Matrix has positioned itself around this inference market with specialised AI processors. The company shipped its first AI chip in November 2024 and has continued developing hardware aimed at running AI workloads. Its latest collaboration with NVIDIA gives its processors a path into NVIDIA’s broader data-centre ecosystem through NVLink Fusion.
The startup has also attracted backing from major technology investors. Microsoft has supported d-Matrix since the company’s $110 million financing round in 2023. The startup was valued at $2 billion when it raised $450 million in 2025. These investments have supported its efforts to develop and commercialise specialised AI computing technology.
For NVIDIA, the collaboration expands the potential use of its data-centre infrastructure by allowing custom AI chips to connect through NVLink Fusion. For d-Matrix, the technology provides an opportunity to place its inference-focused processors within NVIDIA-compatible server systems and target customers looking for specialised AI computing capabilities.
The partnership comes as AI infrastructure companies increasingly explore specialised processors alongside conventional GPUs. While GPUs remain central to many AI workloads, inference requirements can differ from model-training workloads, creating opportunities for processors designed specifically around speed, efficiency and low-latency AI execution.
With Raptor’s final design expected by the end of 2026 and NVIDIA-compatible systems targeted for 2027, the collaboration represents a significant step in d-Matrix’s effort to scale its inference technology. The company’s work with NVIDIA and Astera Labs will focus on connecting specialised processors and high-speed data infrastructure to support the next generation of AI applications.
