Phison has unveiled aiDAPTIV, a new memory extension technology designed to help everyday PCs run larger AI models locally without requiring large amounts of RAM. The technology was introduced at Computex 2026 through a collaboration with Intel.

The platform combines Intel Core Ultra Series 3 processors with Phison’s memory extension architecture, enabling consumer computers to handle more demanding AI workloads. As AI models continue to grow in size, memory capacity has become one of the biggest challenges for local AI computing.

Figure 1. Storage Cards.

To address this issue, aiDAPTIV extends working memory beyond traditional DRAM by using high-performance NAND flash storage. This approach provides additional memory resources, allowing larger AI models to operate efficiently on systems with limited physical RAM. Figure 1 shows storage cards.

According to Phison, internal tests showed that a 26-billion-parameter language model could run on a PC equipped with just 16 GB of RAM when aiDAPTIV was enabled. Without the technology, the same workload required 32 GB of RAM. The platform also includes features such as KV cache reuse, which allows AI applications to retain context from previous interactions. This reduces repetitive processing, improves efficiency, and supports smoother long-running AI sessions.

As AI increasingly moves from cloud servers to personal devices, technologies like aiDAPTIV could make advanced AI more accessible. By lowering memory requirements and enabling larger models to run locally, the innovation has the potential to deliver faster, more private, and more cost-effective AI experiences.

Reference:

1. https://interestingengineering.com/ai-robotics/phison-aidaptiv-runs-26b-model-on-16gb-ram

Cite this article:

Keerthana S (2026), 26-Billion-Parameter AI Models Now Run on Standard PCs, AnaTechMaz, pp.461