Computex 2026: Phison Collaborates with Intel to Bring Larger Local AI Workloads to Intel AI PC Platforms

Key Takeaways

• Phison's aiDAPTIV lets 26B‑parameter model run on 16 GB DRAM
• Collaboration targets Intel Core Ultra AI PCs with OpenVINO support
• Extends effective AI memory using high‑endurance NAND flash cache
• Enables larger MoE models, agentic AI, and private on‑device workloads
• OEMs ASUS, MSI, Acer will demo aiDAPTIV at Computex

Pulse Analysis

The PC industry is at a turning point as generative AI moves from cloud‑only services to on‑device execution. Memory capacity has been the primary bottleneck for running sophisticated models such as Mixture‑of‑Experts (MoE) or multi‑step agents on client hardware. Phison's aiDAPTIV addresses this constraint by creating a cache layer that merges system DRAM with ultra‑fast, endurance‑rated NAND flash, effectively expanding the working memory pool without a hardware redesign. This approach aligns with the broader trend toward edge AI, where latency, bandwidth costs, and data sovereignty are increasingly critical.

Technically, aiDAPTIV leverages Phison's Pascari controller to offload portions of the model’s key‑value cache and intermediate tensors to flash storage, while maintaining the speed needed for real‑time inference. In internal benchmarks, the technology enabled a 26‑billion‑parameter model—typically requiring 32 GB of RAM—to run on a 16 GB system, delivering comparable latency and throughput. Integration with Intel's Core Ultra Series 3 processors and the OpenVINO toolkit ensures developers can tap into existing software stacks, simplifying deployment for ISVs and OEMs. The partnership also includes joint performance validation and reference designs, accelerating time‑to‑market for AI‑enhanced laptops and workstations.

For the market, the collaboration promises tangible business benefits. Enterprises can keep sensitive data on‑premise, reducing exposure to cloud‑related privacy risks and lowering recurring inference costs. OEMs gain a differentiator that lets them ship thinner, cheaper devices without sacrificing AI capability, potentially expanding the addressable market for AI‑centric PCs. As competitors explore alternative memory architectures, Phison and Intel’s joint solution positions them as early leaders in the emerging on‑device generative AI segment, setting a benchmark for future hardware‑software co‑designs.