d-Matrix Plans to Connect Future Raptor XPUs With NVIDIA NVLink Fusion

d-Matrix Raptor XPUs are expected to use the NVIDIA NVLink Fusion platform in the future. NVIDIA announced this on September 10, 2026, saying that d-Matrix plans to integrate its upcoming XPU generation into infrastructure incorporating NVLink for scale-up connectivity, Spectrum-X for scale-out networking and the NVIDIA MGX rack architecture.

The announcement describes a plan for future deployment, not the launch of a finished product. In the available materials, neither NVIDIA nor d-Matrix confirmed when the Raptor XPU will be introduced, its pricing, customer installations or specific rack configurations.

What the d-Matrix Raptor XPU integration is expected to include

NVLink Fusion is an NVIDIA platform designed to connect custom accelerators and partner processors with the company’s system infrastructure. In d-Matrix’s case, it is expected to include multiple layers of the data center.

• NVLink is expected to provide scale-up connectivity within a higher-performance system.
• Spectrum-X is intended to provide scale-out networking between systems.
• NVIDIA MGX is expected to provide the rack architecture for building the resulting systems.

In this case, XPU is d-Matrix’s designation for a specialized accelerator intended for AI inference, meaning the operation of already-trained models. The company is therefore not announcing that it will replace NVIDIA GPUs: the stated direction is to place its future accelerator within an ecosystem of NVIDIA networking and rack components.

Raptor follows the Corsair platform

d-Matrix has already brought its current Corsair platform into full production. However, the new announcement describes Raptor as a future XPU generation, so the Corsair information cannot be understood as confirmation of the availability of Raptor or systems built on NVLink Fusion.

The distinction is also important when evaluating the partnership. The published materials contain no independent tests of the Raptor XPU or comparison with Corsair, NVIDIA GPUs or other inference accelerators. The power consumption, throughput and latency specifications of a specific d-Matrix solution are also unknown.

NVIDIA expands its system ecosystem

For d-Matrix, using NVIDIA’s existing rack and networking infrastructure could mean that it does not need to build the entire system foundation from scratch while developing its own accelerator. NVIDIA, in turn, is expanding the use of its interconnect, networking and rack design into data centers where specialized accelerators from other manufacturers are used alongside GPUs.

For infrastructure operators, the main relevance is the possibility of combining GPUs and specialized inference XPUs within a related system environment. However, the real benefit of such a combination has not yet been demonstrated through customer deployments or publicly available benchmarks.

For its platform, NVIDIA cites up to three times lower latency and ten times higher packet speed compared with standard Ethernet. These are the manufacturer’s claims about the platform, not an independently verified result for a future d-Matrix configuration.

What is still missing

The key question will be when d-Matrix introduces the Raptor XPUs themselves and their commercial availability. System partners, specific rack designs and customer orders have not yet been confirmed. Technical comparisons will require independent measurements of latency, throughput, energy efficiency and compatibility with NVIDIA systems.

A separate d-Matrix announcement about this partnership could not be found in the available searches. The basic information about the integration plan was confirmed by a first-party NVIDIA source.

Verified and updated: 09/10/2026 15:24