JEDEC has confirmed that the LPDDR6 standard will support 512GB SOCAMM2 modules, doubling LPDDR5X capacity, with commercialization targeted for 2028–2029, or potentially earlier. The specification is positioned for agentic AI datacenters, emphasizing higher density, lower power use, and Processing-in-Memory (PIM) capabilities. NVIDIA’s Vera CPU and AMD’s Verano CPU are cited as adopters of the LPDDR ecosystem, while Samsung, SK Hynix, and Micron are advancing toward mass production, according to EETimes.

TSMC is ramping up volume production of its 2nm process, but the per-wafer price is expected to be markedly higher than 3nm as costs rise and pricing increases in gradual steps. That shift could raise foundry costs for key customers such as Apple and NVIDIA. A South Korean report says Samsung, helped by its GAA roadmap and greater pricing flexibility, is emerging as a potential alternative—particularly for non-flagship chips in areas like automotive, robotics and edge AI.