$SNDK making moves in the AI infrastructure space. They've taped out their first high-bandwidth flash chip and plan to ship samples to customers next year for AI inference devices.
This matters because inference—actually running AI models at scale—is where the real volume opportunity lives. Training gets the headlines, but inference is where the economics happen daily.
HBF is positioning as a faster, more efficient alternative to traditional storage in AI workloads. If SanDisk executes, they're carving out a niche in the broader AI memory hierarchy alongside HBM and traditional NAND.
Tape-out means the design is locked. Now it's about yield, performance validation, and whether hyperscalers actually adopt it at volume. Samples in 2026 means revenue impact is 2027 at earliest.
Watch how this plays against Micron, SK Hynix, and Samsung's memory roadmaps. The AI stack is still being built, and storage architecture is quietly becoming a bottleneck worth solving.
This matters because inference—actually running AI models at scale—is where the real volume opportunity lives. Training gets the headlines, but inference is where the economics happen daily.
HBF is positioning as a faster, more efficient alternative to traditional storage in AI workloads. If SanDisk executes, they're carving out a niche in the broader AI memory hierarchy alongside HBM and traditional NAND.
Tape-out means the design is locked. Now it's about yield, performance validation, and whether hyperscalers actually adopt it at volume. Samples in 2026 means revenue impact is 2027 at earliest.
Watch how this plays against Micron, SK Hynix, and Samsung's memory roadmaps. The AI stack is still being built, and storage architecture is quietly becoming a bottleneck worth solving.