d-Matrix는 자사의 AI 추론 서버에서 NVIDIA의 NVLink Fusion 기술을 활용할 것입니다.

AI chip startup d-Matrix announced on 10 September 2026 that it will use Nvidia's NVLink Fusion technology to connect its next-generation Raptor inference chips inside AI servers. The Raptor XPUs will plug into Nvidia's MGX rack-scale architecture, with initial availability targeted for the fourth quarter of 2027.

NVLink Fusion is Nvidia's interconnect standard that allows third-party chips to integrate into its rack and networking hardware without requiring rivals to build competing systems from scratch. Nvidia confirmed the arrangement in its own announcement, stating that "d-Matrix today announced it will use NVIDIA NVLink Fusion to connect its next-generation Raptor XPUs to NVIDIA's AI infrastructure platform."

d-Matrix is partnering with Astera Labs on the interconnect hardware needed to integrate its chips into the Nvidia stack. This arrangement positions d-Matrix as part of a broader NVLink Fusion ecosystem that already includes AWS, Arm, Intel, Fujitsu and MediaTek. The partnership enables d-Matrix to pitch its Raptor chips to AI labs, hyperscalers and neocloud operators for what the company describes as "premium-level" token-generation services.

Notably, d-Matrix's Raptor chips are designed to compete directly with Nvidia's own GPUs on inference workloads—the segment of AI computing that runs trained models rather than trains them. Yet the startup still requires Nvidia's rack and interconnect infrastructure to distribute its silicon at scale to data centres. This dynamic positions Nvidia to benefit from AI inference demand regardless of which chipmaker ultimately wins the workload.

d-Matrix is a well-funded operation: the company raised $110 million from investors including Microsoft at a time when many chip startups struggled to secure capital. This backing lends credibility to the NVLink Fusion partnership as a signal of where serious AI infrastructure investment is flowing, rather than serving as a mere marketing exercise.

Nvidia reported net income of $59.688 billion for its fiscal second quarter ended 26 July 2026, more than double the $26.422 billion it posted a year earlier. Diluted earnings per share for the quarter reached $2.46, compared with $1.08 a year prior. On 10 September, Nvidia shares closed at $218.6201, down 2.26% over the prior 24 hours, though no public information ties that movement specifically to the d-Matrix announcement. The stock remains up 1.08% over the preceding 20 trading days, having ranged between a high of $234.55 and a low of $208.48. Trading volume was roughly 41% of the 20-day average.

The broader economic backdrop remained stable. The 10-year US Treasury yield stood at 4.8% on 8 September, essentially flat from 4.78% at the previous reading, while the unemployment rate held at 4.1% in August.

Initial availability of the d-Matrix Raptor chips on Nvidia's NVLink Fusion architecture is scheduled for the fourth quarter of 2027. Nvidia has been steadily expanding its NVLink Fusion partner roster across custom silicon and connectivity, and further additions would signal how broadly the platform strategy extends beyond GPU sales.