The Silicon Valley firm’s new Raptor processors will plug directly into Nvidia data-centre hardware via NVLink Fusion to target low-latency AI inference workloads.
Artificial intelligence startup d-Matrix announced on Thursday that it will adopt Nvidia’s chip-linking technology, enabling its processors to operate directly within the semiconductor giant’s data-centre systems as demand for AI services continues to intensify.
As artificial intelligence workloads shift from model training to daily deployment—a process known as inference—d-Matrix is positioning its hardware to capture processing market share. While Nvidia’s high-end graphics chips dominate training tasks, d-Matrix specialises in inference operations.
The startup’s new Raptor chips will connect to Nvidia server racks using NVLink Fusion, a proprietary technology featuring specialised connectors and memory designed to integrate custom AI chips into Nvidia’s broader infrastructure. The Raptor processor is expected to complete its final design phase by the end of this year, with compatible server racks scheduled for market release in 2027.
D-Matrix stated that the integrated systems are engineered for high-speed, low-latency applications, including coding assistants, chatbots, and voice agents. Financial terms of the collaboration were not disclosed.
To maintain rapid data flow across the system, the Santa Clara-based firm is also partnering with semiconductor connectivity provider Astera Labs.
D-Matrix has drawn backing from Microsoft since a $110 million funding round in 2023. After shipping its debut AI chip in November 2024, the startup achieved a $2 billion valuation following a $450 million capital raise last year.
