Bernstein highlights High-Bandwidth Flash (HBF) as a potential game-changer for AI infrastructure, addressing memory capacity and bandwidth constraints .
Key takeaways:
- HBF combines HBM-like read bandwidth with 8-16x greater capacity at lower cost per gigabyte, making AI inference more efficient .
- Could reduce GPU memory needs and allow larger AI models with fewer GPUs, cutting computing and power costs .
- Significant opportunity for NAND manufacturers: HBF could consume ~6.5x more NAND wafer capacity per bit than conventional TLC NAND due to higher-performance SLC flash and advanced packaging .
This signals a new demand vector for the NAND sector amid expanding AI workloads.
Could HBF be the solution to the Memory bottleneck?
INVESTING.COMNov 8, 10:10 PM UTC
6