Nvidia는 d-Matrix에 투자할 계획인 것으로 전해졌으며, 소식통은 d-Matrix를 경쟁 관계의 AI 칩 스타트업이라고 설명했다.

Nvidia plans to invest in d-Matrix, an AI inference chip startup that sells processors competing with its own, according to The Information. Citing three people with knowledge of the deal, the report says the investment is part of a push to make Nvidia's technology work with chips from other designers. Neither Nvidia nor d-Matrix has confirmed the investment, and neither has disclosed its size or timing.

The reported stake would follow similar arrangements with Marvell, Intel and Amazon. Nvidia and Marvell announced a strategic partnership on March 31 that connects Marvell's custom chips to Nvidia's AI factory ecosystem through NVLink Fusion, and Nvidia has said it also invested in Marvell. Amazon Web Services (AWS), meanwhile, said in December that it is designing its Trainium4 chips to work with NVLink Fusion.

First unveiled in May 2025, NVLink Fusion is the interconnect technology that lets partners build semi-custom AI infrastructure. Non-Nvidia accelerators plug into Nvidia's rack designs and networking, so a data center can mix processor types without a separate rack architecture for each.

d-Matrix announced on September 10 that it will use NVLink Fusion to connect its next-generation Raptor processors to Nvidia's platform. Its racks are meant to work alongside Nvidia GPU systems such as Vera Rubin NVL72 for disaggregated inference, in which different hardware handles different stages of a workload. "Demand for inference is soaring, but capital, time and energy remain finite," said Sid Sheth, d-Matrix's cofounder and CEO. Reuters reported that Microsoft has backed the startup since a $110 million round in 2023.

The logic is about staying central as customers diversify. Rivals' chips that connect through Nvidia's networking and racks still sit inside Nvidia's infrastructure. Customers can add specialized hardware without abandoning their existing systems, and Nvidia keeps its role in the stack. The move also hedges against the inference market fragmenting into purpose-built chips, a shift Nvidia has addressed with its Groq 3 LPX inference accelerator.

d-Matrix is a sizable bet. Data Center Dynamics notes that it closed a $275 million Series C in November 2025 at a $2 billion valuation, and that The Information separately reported in July that it was seeking new financing at a $5 billion valuation. Reuters' figures differ slightly, citing $450 million raised last year at the same $2 billion valuation.

Nothing in the reporting mentions the Gulf. The relevant point is indirect: Gulf Cooperation Council (GCC) governments and cloud providers building AI capacity buy racks and networking from this ecosystem, and a more open Nvidia platform would widen the chip options they can slot into one design. That remains a possibility, not a confirmed offer. Nvidia's RTX Spark and Surface Laptop Ultra launch shows the same company pushing its stack from data centers down to Windows PCs.