엔비디아는 경쟁 추론 칩 공급업체 디매트릭스에게 데이터센터 랙을 개방하여 경쟁 가속기가 동일한 인프라 생태계에 통합될 수 있도록 했다.

Nvidia has opened its high-speed interconnect technology to d-Matrix, a Microsoft-backed AI chip startup, allowing the company to plug its inference processors directly into Nvidia-powered racks using NVLink Fusion—a capability expected to launch in 2027. The move extends Nvidia's interconnect standards to external partners, potentially establishing it as an industry benchmark while creating new revenue streams for Nvidia beyond GPU sales, including fees from networking hardware, CPUs, and software services.

The arrangement maintains Nvidia's stronghold in data center infrastructure even as it invites competition in the inference segment. While rivals could capture meaningful inference workloads, Nvidia's ecosystem positioning—collecting revenue across multiple product layers—insulates the company from pure chip commoditization. The strategy strengthens Nvidia's role as the backbone of AI infrastructure rather than merely a processor vendor.

Market sentiment reflects cautious optimism. Nvidia shares (NVDA) trade at $212.06, up 0.52% intraday but closer to their 52-week low of $165.17 than their high of $235.75. Microsoft (MSFT), d-Matrix's backer, is priced at $498.47, down 1.37% today and below its 52-week high of $542.07. Marvell Technology (MRVL), a competing chipmaker, stands at $222.63, up 1.74% but significantly below its 52-week peak of $316.43, suggesting divided investor sentiment across the semiconductor space.

The infrastructure trend is evident elsewhere: Nvidia's latest AI supercomputer, the VR200 NVL72, allocates 70% of its $4 million bill of materials to Vera Rubin GPUs. Broadcom, meanwhile, reported AI semiconductor revenue up 221% year-over-year to $16.7 billion, with total revenue rising 85.5% to $29.59 billion and free cash flow reaching $13.66 billion—underscoring the ongoing capital intensity and profitability of AI infrastructure plays across the sector.