[Caitong Tang Jia] Overseas memory makers abandon or scale down #self-developed CXL controllers, benefiting Montage Technology and Fortune Tech's demand growth

English summary

CXL is a high-speed interconnect standard for the AI computing era, primarily addressing the memory wall pain point: traditional servers can only carry a fixed-capacity local DDR5 memory. In AI large model inference and KV Cache scenarios, memory utilization is only 50%-65%, leading to severe hardware waste. The CXL bus enables memory pooling, allowing computing nodes to attach independent memory expansion pools and flexibly allocate memory on demand, significantly reducing total data center hardware costs. From an industry trend perspective, #2026 is the starting point for CXL large-scale deployment, and #it is expected that 2027 will mark the first year of CXL commercial scale adoption. 1. Marginal changes: According to Korean media reports, recently #Samsung Electronics, SK Hynix, and Micron Technology have all scaled down or abandoned their self-developed CXL controller commercialization plans. Among them, Micron has closed its self-developed division and adopted PrimeMass's solution; SK Hynix terminated its self-developed project and shifted the team to PIM (processing-in-memory); Samsung removed its self-developed controller from the commercialization roadmap, retaining only frontier R&D. 2. ...