Latest Search
Quote
| Back Zoom + Zoom - | ||||
|
Sandisk Corporation (SNDK.US) CTO: Global AI Race Hinges on Memory Rather Than Compute; HBF Chip Under Design for Launch Next Year
Recommend 21 Positive 35 Negative 22 |
||||
|
||||
|
Memory manufacturer Sandisk Corporation (SNDK.US) has seen its share price surge about 5.7 times year to date and more than 35 times over the one-year period since May last year. In an interview with Nikkei Asia, Alper Ilkbahar, Chief Technology Officer (CTO) and Executive Vice President of Sandisk Corporation, said the global AI race is becoming increasingly "memory-centric" rather than focused on compute power, a trend that could exacerbate the unprecedented global tightness in memory chip supply. He noted that relying solely on graphics processing units (GPU), central processing units (CPU) and raw computing power is insufficient to handle increasingly complex AI workloads. Ilkbahar said several trends are making memory more critical. First, as large language models become bigger and more intelligent, they require more memory to operate effectively. Second, systems such as OpenAI's ChatGPT, Google's Gemini and Anthropic's Claude are increasingly dependent on so-called key-value (KV) cache. In simple terms, KV cache functions like AI's short-term memory, enabling it to retain previous user inputs and earlier parts of a conversation without recomputing everything from scratch. This approach significantly enhances speed and efficiency, but as AI conversations and context windows lengthen, KV cache demands substantially greater memory capacity. Another factor is the adoption of mixture-of-experts models, which reduce computation but increase memory requirements. This is why the AI race is being described as memory-centric. Ilkbahar added that customers are now proactively making advance commitments and signing long-term procurement agreements to secure future supply. He believes the next major development will be High Bandwidth Flash (HBF). The company is currently designing related chips, with HBF memory dies expected to be sampled by the end of this year and full products equipped with controllers scheduled for launch next year. Sandisk Corporation stated that as AI inference workloads require larger memory capacity, HBF can play a greater role. The company said HBF offers significantly higher capacity and memory density than High Bandwidth Memory (HBM), while maintaining comparable bandwidth, making it better suited to future AI inference needs. Sandisk Corporation has partnered with SK Hynix to jointly formulate technical standards for HBF. (da/u) Auto-translated by AI This article was automatically translated by AI, the original language version should be considered the authoritative version. AASTOCKS.com Limited does not guarantee its accuracy or completeness and accepts no liability for any damages or losses arising from the use of this translation. More Details
AASTOCKS Financial News |
||||
Copied to Clipboard
 |
||||
