China's AI Storage Bottleneck Highlights: HBM and Advanced Packaging May Become Core Challenges in the Next Three Years

Recently, discussions on the Chinese social platform X regarding the core bottlenecks for China's AI industry over the next three years have been heating up. Participants generally believe that high-bandwidth memory (HBM), DRAM, optical interconnects, and advanced packaging technology will be key factors constraining AI development. Developments at companies such as Micron (MU) and NVIDIA (NVDA), as well as storage product price increases and long-term contract signings, are all seen as industry bellwethers.

Industry Discussion Background and Focus

In related topic posts, multiple industry practitioners noted that the demand for high-performance storage in AI model training is growing exponentially. As a crucial supporting memory for GPUs and accelerators, HBM supply tightness is not a new topic. The discussion pointed out that if HBM production capacity cannot match demand over the next three years, the efficiency of AI inference could be directly impacted. Meanwhile, DRAM prices show clear signs of recovery, with some manufacturers beginning to sign long-term supply agreements to lock in costs.

Optical interconnects and advanced packaging technology were also repeatedly mentioned. These technologies play an important role in interconnects within data centers, reducing latency and improving energy efficiency. However, the relevant supply chain is still heavily reliant on overseas suppliers, and domestic companies still need time to catch up in terms of technology accumulation and large-scale production.

Storage Market Dynamics Analysis

From a global perspective, fluctuations in storage chip prices are closely linked to AI demand. Recent market data shows that HBM product quotes are trending upward, mainly due to increased procurement by leading companies like NVIDIA. The earnings expectations of storage manufacturers such as Micron have also attracted attention. Long-term contract signings have become the norm, reflecting both downstream AI companies' concerns about supply and upstream companies' caution in capacity planning.

Participants in China's AI industry chain are closely tracking these changes. Some believe that domestic storage companies have opportunities to make breakthroughs in HBM substitution, but technical barriers and yield issues still need to be overcome. The optical interconnect field may accelerate deployment through international cooperation.

Potential Impact and Industry Chain Outlook

If the above bottlenecks persist, AI training costs may rise further, putting greater pressure on small and medium-sized model developers. For Chinese companies, this is both a challenge and an opportunity to promote self-sufficiency and control. Progress in advanced packaging technology may alleviate some memory bottlenecks, but overall supply chain resilience still requires multi-party collaboration to improve.

Market participants remind that such discussions are mostly industry observations and not definitive predictions. Storage price trends will be influenced by multiple factors including the macroeconomy and the pace of capacity expansion.

Conclusion

The discussion on AI storage and HBM bottlenecks reflects the cautious attitude of the industry toward the technology supply chain. Over the next three years, developments in areas such as HBM and optical interconnects will directly impact the competitiveness of China's AI industry. Enterprises and policymakers are likely to continue focusing on capacity deployment and technological innovation to address potential challenges.