d-Matrix has announced a concrete infrastructure direction for its next-generation Raptor XPUs: using NVIDIA NVLink Fusion to connect them with NVIDIA’s AI infrastructure platform. The sourced claim is about intended deployment architecture, not demonstrated performance or completed deployment.
What the evidence says
NVIDIA Blog reports that inference-chip maker d-Matrix will use NVLink Fusion to connect its next-generation Raptor XPUs to NVIDIA infrastructure. The primary account presents the company as joining a growing group of ecosystem participants.
The source says NVLink Fusion allows XPU and CPU developers to concentrate on processor design while using NVIDIA infrastructure for deployment at AI-factory scale. It also identifies the NVIDIA MGX ecosystem as a route to mature, validated rack designs and supporting supply-chain, power, and cooling infrastructure.
Taken strictly as sourced fact, the announcement establishes adoption intent and the proposed connection between Raptor XPUs, NVLink Fusion, and NVIDIA’s infrastructure platform. It does not establish deployment completion, operating results, or comparative performance.
Operator implications
Operator analysis: the significant move is not simply the selection of an interconnect. The source frames processor development as only one part of rack-scale deployment, placing rack design, supply chain, power, and cooling alongside silicon as relevant operational concerns.
For infrastructure teams evaluating this direction, the announcement suggests a useful diligence boundary: separate the processor’s stated role from the surrounding platform required to deploy it. The relevant questions would cover how the Raptor XPU is expected to fit into validated rack designs and how the named infrastructure components will support that intended deployment.
The adoption also suggests that d-Matrix plans to rely on an existing infrastructure ecosystem for parts of rack-scale integration. That is an architectural signal, but the available evidence does not show how the integration will perform in operation or what implementation work remains.
Limits and open questions
This is a single primary account and is not independently confirmed. The source describes an announced plan, so it cannot by itself verify that systems have been deployed or that the intended integration has achieved any operational outcome.
Unknowns include deployment timing, system configurations, implementation status, performance, power behavior, cooling requirements, supply availability, and validation results. The evidence also does not provide comparisons with other deployment approaches. Until further evidence appears, the defensible conclusion is limited: d-Matrix says it will connect its next-generation Raptor XPUs to NVIDIA infrastructure through NVLink Fusion and expects to draw on the MGX ecosystem for rack-scale deployment support.