What the Phison–Intel AI PC partnership is about
The Phison–Intel collaboration is a technical partnership that combines Intel AI platforms with Phison’s aiDAPTIV memory extension to run larger local AI workloads directly on PCs without depending on cloud processing. This effort targets AI PCs that need to support more advanced, long-running AI applications, including agentic workflows and Mixture-of-Experts (MoE) models that previously exceeded typical system memory limits. By pairing Intel Core Ultra Series 3 processors with Pascari aiDAPTIV Cache Memory, the platform treats high‑performance NAND flash as an effective extension of DRAM for AI PC storage. That extended working memory makes it possible to keep more model parameters and session state local, so users can move beyond simple assistants toward document analysis, multi-step automation, and private data tools that stay on-device. In short, the partnership aims to make on-device AI processing powerful enough for workloads once reserved for data centers.
How aiDAPTIV extends memory for local AI workloads
At the heart of the collaboration is Phison’s Pascari aiDAPTIV, a memory extension solution that blends traditional DRAM with extreme-endurance NAND flash. Instead of forcing OEMs to ship AI PCs with ever larger DRAM configurations, aiDAPTIV expands the effective AI working memory by caching key model data and session information on fast storage. Phison says aiDAPTIV can cut the DRAM requirement for certain local AI workloads while still supporting features such as KV cache reuse, which is important for long conversations and ongoing AI sessions. In one Phison test, aiDAPTIV enabled a 26‑billion‑parameter model to run on a system with 16 GB of DRAM, compared with the 32 GB required without aiDAPTIV in the same environment. That change directly tackles one of the biggest limits on on-device AI processing: how many parameters and tokens can fit into memory at once.
What this means for on-device AI processing and user privacy
By raising the ceiling on model size and session length, the Phison–Intel approach reshapes how users interact with AI PCs. Larger models and richer context windows can now live on the device, reducing the need to send prompts and documents to cloud services for processing. This shift improves latency because responses no longer depend on network round-trips, and it strengthens privacy by keeping sensitive data in local AI workloads that run fully on the machine. According to Phison’s CEO KS Pua, aiDAPTIV lets OEMs, developers, and end users “run more capable AI applications locally while maintaining privacy and infrastructure efficiency.” For enterprises, that makes it easier to build AI workflows around private repositories or domain-specific models, while consumers gain faster, more responsive assistants that can operate even when offline or on poor connections.
Ecosystem demos show AI PCs moving beyond simple assistants
The partnership is being brought to life through demonstrations on Intel AI PC platforms, including systems from ASUS, MSI, and Acer. At events like Computex, Phison and Intel are showing a local chat interface powered by an MoE model that would normally exceed available system memory, proving how aiDAPTIV-backed AI PC storage can keep such workloads on-device. Another demo involves a hybrid large language model routing application built on the open-source OpenClaw agent framework. It runs larger MoE models locally, but can selectively route complex requests to the cloud only when needed, which reduces cloud token usage. Software partners such as Ollama, LLMWare, TurinTech, Intel AI Superbuilder, and Intel AI Playground are also displaying real-world applications, from code optimization to enterprise RAG and agent workflows, signaling a broader shift toward richer, persistent on-device AI experiences.
Implications for the next generation of AI PCs
For the AI PC market, the Phison–Intel collaboration could change upgrade priorities. Instead of focusing solely on increasing DRAM, OEMs can pair reasonable memory configurations with smarter AI PC storage and aiDAPTIV to unlock larger models. Intel AI platforms with Core Ultra processors and OpenVINO support give developers a target environment for optimizing on-device AI processing, while Phison’s approach helps them scale model capacity without blowing out power or cost budgets. Intel’s Jim Johnson notes that more users “want to run AI locally – faster, more private and without the cost of sending everything to the cloud.” If this model of extended memory and hybrid routing proves reliable, AI PCs may become the default place where people run personal, domain-specific AI tools, turning local machines into long-running, context-aware AI workspaces rather than thin clients for remote services.
