- Sandisk and SK Hynix propose high bandwidth memory with flashpoderes to handle larger AI models
- High bandwidth flash could store many more data than HBM based on DRAM for AI work loads
- NAND non -volatility energy saving could remodel the cooling strategies of the AI data center
Sandisk and SK Hynix have signed an agreement to develop a memory technology that could change the way in which the accelerators of the data handle the data at scale.
Companies are standardized by the “high bandwidth flash” (HBF), an NAND -based alternative to the traditional bandwidth of bandwidth used in the AI GPUs.
The concept is based on HBM packaging designs while replacing part of the DRAM pile with flash, exchanges some latency for a greatly increased capacity and non -volatility.
IA memory batteries to handle larger models with lower energy demands
This approach allows HBF to provide between eight and sixteen times the storage of HBM based on approximately similar costs.
Nand’s ability to retain data without constant power also brings potential energy savings, an increasingly important factor as Ia’s inference expands to strict energy and cooling limits.
For hyperscala operators who execute large models, change could help address the thermal and budgetary limitations that are already forcing the data center operations.
This plan is aligned with a research concept entitled “LLM in a flash”, which described how large language models could be executed more efficiently incorporating SSD as an additional level, relieving the pressure on dram.
HBF essentially integrates that logic in a single high bandwidth pack, potentially combining the largest SSD storage scale with the speed profile necessary for AI workloads.
Sandisk presented its HBF prototype at the Flash Memory Summit 2025, using BICS NAND and patented wafering techniques.
Sample modules are expected in the second half of 2026, with the first HBF hardware using projected HBF for the beginning of 2027.
Specific product associations have not been revealed, but the position of SK Hynix as an important memory provider for the main AI chips manufacturers, including NVIDIA, could accelerate adoption once the standards are completed.
This movement also occurs when other manufacturers explore similar ideas.
Samsung has announced storage levels of Flash supported and continues to develop HBM4 DRAM, while companies like NVIDIA remain committed to heavy designs DRAM.
If you succeed, Sandisk and SK Hynix collaboration could create heterogeneous memory batteries where drama, flash and other types of persistent storage coexist.
Via Hardware Toms
