Sandisk CTO: Global AI Race Hinges on Memory, Not Compute; HBF Products Expected Next Year

Alper Ilkbahar, Chief Technology Officer (CTO) and Executive Vice President of Sandisk, said in a recent media interview that the global AI race is becoming increasingly memory-centric rather than compute-centric, which could exacerbate an unprecedented tight supply of memory chips. 

Ilkbahar stated that GPUs, CPUs, and raw computing power alone are no longer sufficient to handle increasingly complex AI workloads. As large language models grow in scale and intelligence, AI conversations and context windows become longer, requiring larger memory capacities for KV Cache. Additionally, mixture-of-experts models reduce computation but increase memory demands, making memory even more critical in the AI race.

With the memory supply gap yet to be alleviated, customers are proactively making pre-commitments and signing long-term purchase agreements to secure future supply.

Ilkbahar also provided an update on SanDisk's HBF technology. The company is currently designing the wafer, with HBF chip samples expected by the end of this year and the complete product with controller projected to launch next year. As AI inference workloads require greater storage capacity, HBF can deliver greater capacity and density than HBM in AI inference.