Qualcomm moves ahead with powerful Samsung LPDDR6X memory for future AI accelerator
Qualcomm is reportedly testing new Samsung LPDDR6X memory for its upcoming AI250 accelerator. This move shows that the company is looking at a fresh and efficient path for AI servers instead of following the common trend.
Last October, Qualcomm introduced its AI200 and AI250 accelerator platforms. These products are designed for AI inference tasks in data centres. What surprised many people was the company’s decision to use low power DRAM instead of high bandwidth memory. Most modern AI servers now depend on HBM because it offers very high speed and bandwidth. However, Qualcomm appears to be focusing on power efficiency and cost balance rather than only raw speed.
According to a report from The Bell, Samsung Electronics has sent early LPDDR6X memory samples to Qualcomm. This is interesting because the standard LPDDR6 memory only received official approval from JEDEC in the third quarter of 2025. Samsung LPDDR6X is still in development, which makes this early testing even more notable.
LPDDR memory is usually found in smartphones and thin laptops. It is known for using less power while still offering good performance. Qualcomm already has strong experience with LPDDR memory in its mobile chipsets. Now, the company seems to be exploring how this memory type can work in server level AI hardware.
The AI250 accelerator, which is expected around 2027, is said to be linked with this LPDDR6X development. Industry sources believe that Qualcomm has a tight product roadmap for the AI200 series. Because of this, Samsung may have delivered early samples to support testing and planning.
There were also recent rumours that the AI200 and AI250 accelerators could use SOCAMM2 modules. In December, Samsung’s memory planning team mentioned that new low power DRAM technologies, including SOCAMM, are being developed around LPDDR6 architecture. These modules are designed with AI focused solutions in mind.
If Qualcomm successfully brings Samsung LPDDR6X into server accelerators, it could offer a different balance of performance, power use, and cost. While HBM remains the popular choice for top tier AI systems, Samsung LPDDR6X may provide a strong alternative for specific workloads, especially inference tasks that need efficiency at scale.
For now, this project is still in early stages. But it clearly shows that Qualcomm is thinking differently and aiming to build a more efficient AI future.
