d-Matrix adopts Nvidia's NVLink Fusion for its Raptor AI inference chips
- News
- Rocks on Galaxy
- Tech
- 11 Sep, 2026
- 0
AI inference specialist d-Matrix announced on September 10 a collaboration with Nvidia: its next-generation Raptor chips will integrate into Nvidia server racks using NVLink Fusion technology.
NVLink Fusion: plugging custom chips into Nvidia's AI factory
NVLink Fusion is Nvidia's platform for letting third-party chipmakers integrate their processors (XPUs) directly into its data-center ecosystem. Per Nvidia, it delivers 3x lower XPU-to-XPU latency than off-the-shelf Ethernet, 10x higher packet rates, and 3 TB/s of all-to-all bandwidth per XPU via sixth-generation NVLink.
Raptor, a purpose-built inference chip
Raptor is a memory-centric inference chip designed from the ground up for NVLink Fusion and Nvidia's MGX rack architecture. It is being actively evaluated at AI hyperscalers and frontier labs, and is backed by more than 100 patents. Its tape-out (final design stage) is expected before the end of 2026.
Availability in 2027
Initial availability of Raptor XPUs integrated into the Nvidia MGX rack is expected in Q4 2027. Demand for inference compute is surging as AI models shift from training to everyday use — the market d-Matrix targets, alongside the Nvidia GPUs that dominate training.