Report: NVIDIA may relax HBM4 specification requirements due to capacity and yield constraints faced by Samsung and SK Hynix

The global AI computing power leader NVIDIA is facing severe tests from the realities of the supply chain. As the mass production timeline for the next-generation AI accelerators approaches, NVIDIA may be forced to adjust its procurement strategy due to dual pressures on capacity expansion and yield improvement from major memory chip suppliers, relaxing its technical specifications for fourth-generation high bandwidth memory (HBM4) to prioritize supply stability over extreme performance.

According to ZDNet, although Samsung Electronics recently announced that HBM4 is about to enter mass production, slightly ahead of competitors SK Hynix and Micron, the overall market supply dynamics still depend on NVIDIA's final procurement decisions. Current indications suggest that even industry leaders face significant challenges in ensuring large-scale supply while meeting NVIDIA's set maximum specification of 11.7Gbps.

This potential strategic adjustment reflects the current tension in the semiconductor supply chain. If NVIDIA insists on the highest performance standards, the mass production of its next-generation AI chip codenamed Rubin may be hindered by shortages of key components. The market currently widely expects NVIDIA to adopt a more pragmatic approach, procuring the highest specification HBM4 while simultaneously sourcing slightly lower-spec versions to balance performance needs with supply chain security.

This shift is crucial for investors, as it not only relates to the launch rhythm of NVIDIA's new products but will also directly reshape the competitive landscape of the memory chip market. With the shortage of memory chips expected to be more severe this year than last, the ability to achieve stable deliveries under the relaxed specifications will be key for suppliers like Samsung and SK Hynix in competing for market share.

Samsung has certification advantages, but mass production yield remains a concern

Although Samsung Electronics is leading in the qualification testing for NVIDIA's HBM4, this does not mean that supply issues have been resolved. The market focus has shifted from merely shipping speed to overall supply capability. Reports indicate that if NVIDIA strictly enforces the 11.7Gbps specification standard, Samsung's current capacity may struggle to meet the large-scale production demands for the Rubin chip.

The constraints mainly come from yield and capacity. As of this month, Samsung's 1c DRAM yield is estimated to be around 60%, and when accounting for advanced backend packaging processes, the effective yield will further decline. Additionally, Samsung's monthly production capacity for 1c DRAM at the end of last year was about 60,000 to 70,000 wafers, which is insufficient to cover NVIDIA's entire demand for HBM4. Although Samsung is pushing for new investments and production line transitions, the ramp-up in capacity will take time, making it difficult to significantly increase supply in the short term.

SK Hynix faces performance compliance challenges, hardware optimization is still ongoing

As another core player in NVIDIA's HBM supply system, SK Hynix is also facing technical obstacles. According to ZDNet, although SK Hynix has secured about 60% of the allocation for NVIDIA's HBM4, its products have encountered difficulties in achieving performance levels at the 11Gbps mark during early reliability assessments. Reports indicate that SK Hynix is currently working to address these performance shortcomings through hardware improvements, but a final solution has yet to be confirmed. This technological bottleneck further increases the risk of NVIDIA's reliance on the supply of top-spec products, forcing it to reassess its technical requirements from different suppliers.

A dual-track procurement strategy may become a certainty to ensure the stability of memory chip supply

Given the aforementioned realities on the supply side, NVIDIA is highly likely to adopt a flexible procurement strategy. According to industry sources cited by ZDNet, NVIDIA will not only procure top-tier HBM4 at 11.7Gbps but may also parallelly procure slightly lower-spec versions such as 10.6Gbps.

This strategic adjustment aims to reduce the technical pressure on suppliers, making it more likely for the three major manufacturers—Samsung, SK Hynix, and Micron—to achieve scalable supply. The report further states that in the context of an increasingly severe shortage of memory chips, NVIDIA's moderate relaxation of HBM4 performance standards is widely seen as a necessary measure to ensure a stable supply. This is crucial for the timely delivery of the next-generation AI infrastructure