Google DeepMind has endorsed SK hynix’s High Bandwidth Flash (HBF) as the memory technology to lead the next-generation AI era, as the AI market shifts from training to inference. SK hynix announced on Aug. 7 that it discussed HBF with representatives from Google DeepMind and SanDisk at the memory technology event ‘FMS 2026’ held in Santa Clara, California, on Aug. 6. HBF, which stacks NAND flash to increase storage capacity and bandwidth, supports AI chips like GPUs in processing massive data efficiently. Google and SanDisk formed a consortium with SK hynix to announce standard specifications for HBF at the event. Researcher Xiaoyu Ma of Google DeepMind said during a panel discussion that the need for HBF will grow as data required by large language models for inference tasks significantly increases, making fast response speeds difficult with existing memory systems. Ma added that HBF’s high capacity and high bandwidth will secure high power efficiency. Regarding concerns that NAND cells could be damaged faster than GPUs or HBM if writing occurs frequently, SanDisk Vice President Rajiv Nagabirava stated such concerns would be overcome if HBF takes on a predominantly data-reading role. SK hynix Vice President Eui-cheol Lim proposed a model using HBF jointly with HBM in the reading-centric AI inference stage.