What the Phison–Intel Collaboration Changes for AI PCs
Phison and Intel’s collaboration is an effort to remove memory and storage limits that prevent local AI workloads from running larger, more capable on-device AI models directly on AI PCs. By pairing Intel Core Ultra Series 3 processors with Phison’s Pascari aiDAPTIV memory extension technology, the partners aim to let users run AI applications that would normally require more system DRAM or cloud resources. This shift matters because AI PCs are moving from simple assistant tasks to heavier jobs like document analysis, multi-step workflow execution and private data protection. These newer workloads demand larger models, persistent session state and fast access to memory. Instead of pushing everything to the cloud, the Intel AI platform plus Phison’s aiDAPTIV turns high-performance NAND into effective working memory, pushing AI PC storage to the center of AI performance.
How aiDAPTIV Extends Memory Using Storage
At the heart of the collaboration is aiDAPTIV, which extends effective AI working memory by spreading it across system DRAM and Phison’s high-endurance NAND flash via Pascari aiDAPTIV Cache Memory. In practice, this means the system treats part of AI PC storage as a fast extension of RAM, tuned specifically for AI access patterns such as key–value (KV) cache reuse. According to Phison, aiDAPTIV enabled a 26B-parameter model to run on a system with 16GB of DRAM, instead of the 32GB needed without aiDAPTIV in the same test setup. That difference shows how storage-aware caching can relax strict DRAM requirements for on-device AI models. For OEMs, this design unlocks higher-tier AI performance on mid-range configurations, while developers get more headroom to ship applications that once required costly memory upgrades or cloud backends.
Local AI Workloads Move Beyond Simple Assistants
The collaboration comes as AI PCs evolve from basic assistant-style features into platforms for complex local AI workloads. Users now expect on-device AI models to summarize long documents, run multi-step workflows and guard private data without sending everything to remote servers. Phison and Intel position their solution as a way to support larger Mixture-of-Experts (MoE) models, longer-running AI sessions and agentic AI workflows that depend on fast, ample memory. KS Pua, Phison’s CEO and founder, notes that AI PCs are becoming “platforms for more sophisticated local AI workloads” that push memory capacity and responsiveness. By combining aiDAPTIV with the Intel AI platform and tools like the OpenVINO toolkit, ISVs can tune their software for local execution first, and reserve cloud usage for rare, high-complexity requests rather than every interaction.
From Cloud Dependence to Hybrid AI on the Intel AI Platform
The Phison–Intel effort signals a broader shift away from total dependence on cloud-based AI processing toward hybrid models where local AI workloads handle everyday tasks. At Computex, the companies plan demos of a local chat interface running a MoE model that would normally exceed system memory, showing aiDAPTIV’s extended working memory in action. Another demo uses OpenClaw, an open-source AI agent framework, to route requests between local and cloud large language models, cutting cloud token usage while preserving a fallback path for difficult queries. Ecosystem partners such as Ollama and LLMWare see this as key for enterprise GenAI workflows like RAG, agents and domain-specific models that benefit from keeping data close to the user. In this architecture, AI PC storage is no longer passive; it becomes an active, tuned part of the on-device AI pipeline.
Why Storage is Now a Critical AI PC Component
Traditional PC design treated storage and memory as separate tiers, but local AI workloads blur that line. High-performance NAND flash, when controlled by AI-aware firmware like aiDAPTIV, can behave as an extension of DRAM for AI tasks that demand large context windows and persistent state. Michael Chiang, co-founder of Ollama, notes that memory is the limiting factor for many powerful models on client hardware, and that aiDAPTIV on Intel AI PC platforms could let people run far larger models locally than their hardware usually allows. With demos spanning partners such as TurinTech, Intel AI Superbuilder and Intel AI Playground, Phison is arguing that AI PC storage is now as important as CPU and GPU choices. As more on-device AI models arrive, users may shop not only for compute performance, but also for storage tuned to AI, where caching strategy determines what their PC can handle offline.
