d-Matrix Joins Nvidia's AI Infrastructure Ecosystem with NVLink Fusion Integration for Raptor XPU

d-Matrix Embraces Nvidia Infrastructure for Raptor XPU Deployment

AI chipmaker d-Matrix announced it will integrate Nvidia's NVLink Fusion chip-linking technology and MGX rack designs into its next-generation Raptor XPU platform, joining a growing roster of partners adopting Nvidia infrastructure for specialized AI workloads1

1

2

2

. The Santa Clara-based startup, valued at $2 billion after raising $450 million last year, specializes in AI inference—the process of running trained AI models for everyday use2

2

. By leveraging NVLink Fusion, d-Matrix sidesteps challenges associated with scaling custom chip architectures across large compute clusters, enabling customers to deploy Raptor XPUs using the same racks and NVSwitch fabrics as Nvidia systems1

1

.

Technical Capabilities of Raptor XPU and Memory Bandwidth Advantages

The Raptor XPU features 32 GB of ultra-fast 3D-stacked DRAM capable of delivering 100 TB/s of memory bandwidth—roughly 4.5 times the memory bandwidth of Nvidia's Rubin GPU1

1

. By the end of 2027, d-Matrix expects to offer rack-scale XPU deployment systems with up to 144 Raptor accelerators connected by a single all-to-all NVLink fabric1

1

. Each rack will provide approximately 2.3 TB of memory capacity—sufficient for models exceeding four trillion parameters at 4-bit precision—and about 7.2 petabytes per second of peak aggregate memory bandwidth1

1

. This in-memory compute platform achieves these specifications by bonding compute logic atop DRAM stacks, delivering memory bandwidth closer to SRAM performance while maintaining higher capacity than traditional HBM approaches1

1

. Memory bandwidth represents the biggest bottleneck for AI inference, directly impacting token generation speed in applications like coding assistants, chatbots, and voice agents2

2

.

Strategic Integration with Nvidia AI Factory Ecosystem

d-Matrix plans to integrate Nvidia Vera CPUs, NVSwitch appliances, BlueField-4 DPUs, ConnectX-9 SuperNICs, and Spectrum-X Ethernet networking alongside its Raptor accelerators1

1

4

4

. The collaboration provides access to MGX rack designs, validated supply chains, power infrastructure, and liquid-cooled architecture without requiring separate rack systems for each processor type3

3

4

4

. Sid Sheth, cofounder and CEO of d-Matrix, stated during a press briefing: "Demand for inference is soaring, but capital, time and energy remain finite. With NVLink Fusion and MGX, we can integrate our Raptor XPUs into a broadly deployed, liquid-cooled architecture, giving customers a faster, lower-risk path to deploy and scale ultralow-latency inference"3

3

4

4

. The Nvidia-compatible racks are expected to be available in 2027, with Raptor chips scheduled to complete their final design stage by the end of this year2

2

. d-Matrix is also partnering with Astera Labs to build custom connectivity solutions ensuring high-throughput data flow throughout the system5

5

.

Nvidia's Expanding Walled Garden Ecosystem Beyond GPUs

Nvidia stands to gain significantly beyond licensing revenues from NVLink Fusion adopters like d-Matrix1

1

. The company generates revenue from multiple infrastructure components even when not selling GPUs directly, as partners integrate Vera CPUs, networking equipment, and other Nvidia technologies1

1

. MediaTek, Marvell, Qualcomm, Arm, Fujitsu, and Amazon Web Services have already adopted NVLink Fusion interconnects1

1

. Nvidia has invested billions in incentives to attract partners into its AI factory platform—spending $3.5 billion on MediaTek and $2 billion on Marvell1

1

. Jensen Huang, founder and CEO of Nvidia, noted: "NVLink Fusion enables partners to integrate custom silicon with NVIDIA's deep ecosystem of NVLink, advanced packaging, rack-scale systems and networking technologies"5

5

. This strategy allows hyperscalers and AI labs to combine specialized inference chips with general-purpose GPUs within unified AI factories, optimizing different workloads across the same infrastructure4

4

. Microsoft has backed d-Matrix since its $110 million financing round in 2023, and the startup shipped its first AI chip in November 20242

2

.