Phison Electronics today unveiled its aiDAPTIV technology at NVIDIA GTC 2026, introducing a multi-tier memory architecture that leverages Pascari SSDs as an AI memory tier for local and edge computing. The innovation addresses growing memory constraints as organizations struggle with fine-tuning and inference workloads requiring massive compute resources.

Traditional memory management systems weren't designed for AI workloads, creating bottlenecks that slow time-to-market for organizations deploying proprietary AI solutions. Phison's approach intelligently extends AI working memory across GPU memory, system RAM, and flash storage, enabling larger models and long-context inference without requiring additional GPU hardware. The technology is built on high-endurance flash optimized for sustained paging and context retention.

At NVIDIA GTC 2026, Phison demonstrated partner notebooks and workstations powered by NVIDIA GB10 Grace Blackwell processors alongside RTX 50 Series and RTX PRO 6000 Blackwell Max-Q GPUs. The showcase highlighted memory-intensive fine-tuning on large-scale models and agentic AI workflows leveraging KV cache reuse. By keeping AI workloads local while maintaining data privacy, Phison's solution helps organizations plan infrastructure investments more predictably without the rising costs typically associated with expanding AI capabilities.