Samsung Ships LPDDR6X to Qualcomm for Next-Gen AI250 Accelerator

Samsung’s HBM4 announcement grabbed attention recently, but the company appears to be moving just as aggressively in the LPDDR space. According to Wccftech, citing Korean outlet The Bell, Samsung Electronics has reportedly shipped LPDDR6X memory samples to Qualcomm.

The report claims Samsung has already completed development of standard LPDDR6 memory. Mass production is expected to begin ahead of a second-half 2026 launch window, positioning LPDDR6 as the next major step beyond LPDDR5.

Samsung pushes LPDDR6X toward AI accelerators

Initial LPDDR6 speeds are said to reach 10.7 Gbps, with Samsung targeting 21% higher efficiency compared to LPDDR5. Future LPDDR6 variants could scale up to 14.4 Gbps or even higher, depending on final configurations and platform support.

LPDDR6X, described as an advanced evolution of LPDDR6, remains under development. Official specifications have not yet been finalized by JEDEC, with more technical details expected later this year.

Qualcomm AI250 rumored to adopt LPDDR6X

Qualcomm is expected to use LPDDR6X in its upcoming AI250 accelerator, which will succeed the current AI200. The AI200 already supports up to 768GB of LPDDR memory, and industry rumors suggest the AI250 could push capacity beyond 1TB using LPDDR6X.

This approach differs from the high-bandwidth memory strategies used by major data center players. Companies such as NVIDIA, AMD, and Huawei typically rely on HBM for top-tier AI accelerators, prioritizing maximum bandwidth and performance.

Qualcomm’s strategy resembles Intel’s Crescent Island Xe3P-based GPU designs in terms of memory philosophy. Instead of using expensive HBM stacks, the design leans on large pools of LPDDR to balance capacity, cost, and power efficiency.

LPDDR as a cost-focused alternative to HBM

HBM delivers extreme bandwidth but introduces higher cost, greater power draw, and more complex packaging requirements. Ongoing DRAM shortages have also tightened HBM supply, affecting availability for high-end AI systems.

LPDDR offers a more cost-effective option for certain AI deployments, especially where massive memory capacity matters more than peak bandwidth. As AI workloads diversify, not every accelerator needs HBM-class throughput.

That context makes Samsung’s LPDDR6 and LPDDR6X developments strategically important. If Qualcomm’s AI250 successfully scales beyond 1TB of LPDDR6X, it could create a new performance tier between mainstream accelerators and ultra-premium HBM-based systems.

LPDDR6X is unlikely to reach mainstream availability until late 2027 or early 2028. Until then, Samsung’s sample shipments signal early ecosystem alignment with key partners.

At the same time, broader memory market dynamics continue to shift. Some manufacturers are reportedly considering Chinese-made memory alternatives, while suppliers increasingly focus on short-term contracts due to uncertain demand and pricing volatility.