NVIDIA signs SK hynix memory and NAVER sovereign-AI deals as D-Matrix challenges its lead

NVIDIA's twin Korea deals deepen its grip on the AI hardware supply chain at both ends. The SK hynix partnership targets co-development of next-generation memory specifically tuned for AI factories — critical because memory bandwidth, not raw compute, is increasingly the bottleneck for large-model training and inference. The expanded NAVER agreement commits NVIDIA's DSX platform to building sovereign AI infrastructure, serving surging global AI demand and reflecting governments' and large platforms' push to keep AI compute domestic.

These moves arrive as competitive pressure on NVIDIA's inference dominance intensifies. CNBC reported that Microsoft-backed D-Matrix is ramping production of chips aimed squarely at the inference market, where cost-per-token economics matter most. Huang's counter is that NVIDIA already leads on low-cost inference, and the company is simultaneously pushing into the CPU market for AI inference and edge devices to widen its footprint.

The strategic logic is vertical integration through partnership: by locking in memory co-development and sovereign-cloud commitments, NVIDIA raises switching costs for would-be challengers who must replicate not just silicon but an entire ecosystem of memory, networking and software.

The broader industry mood this week tilted toward infrastructure and capacity anxiety — warnings that energy and compute, not better models, will be the real limiting factor. On Hacker News a viral post argued xAI "is looking more like a datacentre REIT than a frontier lab" (678 points), capturing skepticism that the frontier race is now an infrastructure-rental game. Readers should watch D-Matrix's production yields and whether sovereign-AI deals proliferate to more countries.