en.Wedoany.com Reported - On June 5, NVIDIA CEO Jensen Huang confirmed that SK Hynix, Samsung Electronics, and Micron Technology have all been certified to supply HBM4 high-bandwidth memory chips for NVIDIA's AI accelerators. HBM4 is a key supporting memory product for NVIDIA's next-generation AI platform, Vera Rubin. The simultaneous inclusion of all three memory leaders in the supply chain indicates that the high-end memory supply landscape for the next-generation AI computing platform has entered a more defined phase of mass production competition.
The role of HBM in AI accelerators has evolved from a traditional "supporting memory" to a critical factor determining system performance, power consumption, and supply cadence. Training and inference for large-scale AI models require continuous movement of massive data between GPUs, CPUs, networks, and storage. The peak performance of computing chips can only be fully realized with a high-bandwidth, low-latency, and low-power memory system. Compared to its predecessor, HBM4 continues to improve stacking density, transmission bandwidth, and energy efficiency. For next-generation AI platforms like Vera Rubin, it must simultaneously meet requirements for packaging integration, thermal management, yield stability, production ramp-up, and customer validation. SK Hynix has long held an advantage in the HBM market, Samsung Electronics is accelerating its efforts to capture the next-generation product window, and Micron Technology has publicly disclosed that its HBM4 36GB 12-layer product is targeting NVIDIA Vera Rubin and has entered mass production and shipment. With all three manufacturers qualified for supply, NVIDIA gains broader supply options for this critical memory component, reducing constraints on overall platform volume caused by a single supplier's capacity, yield, or delivery schedule.
Vera Rubin has entered full-scale production and is scheduled to begin delivery in the third quarter of this year. As the platform progresses, the HBM4 supply cadence will directly impact the deployment speed of next-generation AI servers, AI factories, and large-scale inference infrastructure.
The impact of this confirmation on the global memory supply chain is primarily reflected in two aspects. First, orders for high-end AI memory will continue to concentrate among leading manufacturers with advanced packaging integration capabilities, process technology, and customer validation experience, further raising the entry barrier for the HBM market. HBM4 not only requires improved DRAM die performance but also necessitates co-optimization with the logic base die, advanced packaging, GPU platform, and system-level cooling. Instability in any of these areas can affect overall system performance and delivery. Second, with all three manufacturers simultaneously entering NVIDIA's supply system, the competition for market share will shift from "passing certification" to "who can ramp up yield and capacity faster, who can provide more stable delivery, and who can secure an early position in the next-generation HBM4E and HBM5." For SK Hynix, the advantage lies in its existing HBM supply experience and customer stickiness; for Samsung Electronics, the key is to regain influence in the AI memory market through next-generation products; for Micron Technology, entering the Vera Rubin supply chain helps enhance its presence in high-end AI memory and expand its business opportunities tied to the NVIDIA platform. As AI data centers continue to expand, HBM is no longer just one product line for memory manufacturers but a core variable in their profit structure, capital expenditure, and advanced process roadmap.
The market will subsequently focus on the actual supply shares of the three manufacturers, the speed of HBM4 yield ramp-up, advanced packaging capacity coordination, and the delivery cadence of the Vera Rubin platform. NVIDIA's confirmation that all three memory manufacturers are qualified for HBM4 supply helps stabilize supply expectations for the next-generation AI platform in the short term. Over a longer horizon, HBM4 competition will continue to drive technological advancements in the memory industry around bandwidth, energy efficiency, stacking, thermal management, and packaging integration.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
