Samsung patent boosts AI server memory efficiency without adding GPUs

What happened

Samsung Electronics has disclosed a patent application targeting GPU memory shortages when multiple users simultaneously submit requests to AI servers. By calculating data-transfer costs between GPUs and CPUs and minimizing unnecessary transfers, the technology is designed to enable AI servers to process more user requests quickly within limited memory capacity, without adding GPUs. This is the disclosure of a patent application, i.e., a proposed technical solution, not an implemented product or capacity change.

Impact and signal

Affected components and suppliers

Source and time

Source: DigiTimes Asia · 10/21/2026 · View original source