FADU: Memory Shortage Unlikely to See Notable Improvement in Short Term, Gen6 Shipping in 2H, Gen7 Sampling in 2028

From CPU-Centric to AI-Native: FADU 2.0 Redefines Its Technology Roadmap

FADU Chief Business Officer Tony Kim recently stated that the company has established "FADU 2.0" as its core development strategy, strengthening its enterprise SSD controller technology and extending its product portfolio from traditional CPU-centric data centers to AI computing environments. In his opinion, AI demand is rapidly shifting from training to inference. The explosive growth of AI agents, long-context processing, and complex reasoning applications is forcing data centers to access massive amounts of data more frequently and at higher speeds. In particular, the frequent read/write operations of intermediate inference data such as KV Cache mean that SSDs are no longer merely storage devices but have become a critical component of AI computing architectures.

Gen6 Controllers Shipping in 2H 2026, Gen7 Targeting 2028

Storing all KV Cache data in GPU memory, which is costly and limited in capacity, is not cost-effective. Some data can be offloaded to high-performance SSDs. Therefore, next-generation enterprise SSDs must not only deliver high bandwidth, low latency, and consistent performance but also prioritize energy efficiency.

FADU's product roadmap extends from PCIe Gen5 to Gen6 and Gen7. The Gen6 controller completed chip-out in the first half of this year, with shipments expected to begin in the second half. Among them, the FC-6161 delivers sequential read speeds of up to 28.5GB/s while consuming under 7W at peak performance.

FADU has set a Gen7 target of 57GB/s transfer performance, with a completely redesigned architecture tailored to AI inference, high IOPS, KV Cache, and near-GPU data processing demands. The company is currently discussing specifications with customers, with sample validation expected to begin in 2028.

Memory Supply Gap Unlikely to Close in Short Term, Long-Term Outlook Depends on New Capacity and AI Demand

Kim noted that since new capacity requires significant time from investment and fab construction to mass production, the memory shortage is unlikely to see notable improvement until 2027. However, greater uncertainties may be seen from 2028 to 2029. As new capacity from mainland Chinese manufacturers such as CXMT and YMTC gradually comes online, market supply-demand dynamics could shift. The pace of AI demand growth will also be a key factor influencing future supply-demand trends.