Alper Ilkbahar, CTO and Executive Vice President of SanDisk, stated that the global AI competition is gradually shifting towards a memory-centric approach, which may further exacerbate the supply tightness of memory chips. He pointed out that relying solely on GPUs, CPUs, and raw computing power is no longer sufficient to handle complex AI workloads. With the expansion of large language models, increased reliance on KV caching, and the application of mixture-of-experts models, the importance of memory in AI systems is becoming increasingly prominent. Alper Ilkbahar predicted that High Bandwidth Flash (HBF) will become a key technology in the future, with samples of its die to be available by the end of the year and complete products expected next year. HBF offers significantly higher capacity and memory density compared to HBM while maintaining similar bandwidth, making it better suited for future AI inference needs. Currently, SanDisk has collaborated with SK Hynix to jointly establish technical standards for HBF.