← Accueil actu | Salle du marché
2026-09-10 · Source：Unite AI
The inference chipmaker's multi-year collaboration with NVIDIA brings its next-generation Raptor XPUs into the MGX rack-scale architecture through NVLink Fusion, with first rack-integrated availability targeted for Q4 2027.
AI inference chipmaker d-Matrix announced on September 10, 2026 a multi-year product-roadmap collaboration with NVIDIA that will bring its next-generation Raptor inference XPUs into NVIDIA's MGX rack-scale architecture through NVLink Fusion. According to the company, initial rack-integrated availability of the resulting systems is expected in the fourth quarter of 2027, giving the roadmap a concrete shipping milestone roughly a year out from the announcement.
The agreement's centerpiece is an NVLink Fusion–enabled rack-level system that d-Matrix says will anchor its presence in NVIDIA's AI factory ecosystem. The company said the deal gives its XPUs entry into that ecosystem, NVIDIA's rack-scale framework for AI infrastructure at factory scale. The framing is significant: rather than pitching Raptor as a standalone alternative to NVIDIA's rack offerings, d-Matrix is wiring its inference silicon into NVIDIA's architecture and interconnect path, betting that proximity to NVIDIA's rack platform outweighs the cost of architectural dependence.
For d-Matrix, an inference-focused silicon vendor, the collaboration addresses a persistent commercial problem for alternative accelerator makers: securing a place in the rack-scale deployments that now define AI capacity planning. NVLink Fusion functions here as the connective mechanism, allowing the Raptor XPU line to participate in MGX-based racks rather than competing with them at the system boundary. The multi-year product-roadmap structure also signals that the two companies intend the integration to span successive Raptor generations, not a single product cycle.
Timing is the other variable worth watching. With rack-integrated availability not slated until the fourth quarter of 2027, the September 10 announcement is a roadmap commitment rather than a shipping product, and execution risk sits with a Raptor generation that has yet to reach the market. For NVIDIA, meanwhile, the deal extends NVLink Fusion's role as the on-ramp through which outside inference silicon plugs into its AI factory ecosystem — a notable data point for anyone tracking how the boundary between NVIDIA's platform and the broader accelerator market is being redrawn.
Mots-clés: d-Matrix · NVIDIA · Raptor XPU · NVLink Fusion · AI Inference · MGX
萬安算交所 AIXX · AIXX · Bourse compute vérifiée · Accéder à la salle