Samsung is already punting LPDDR6X memory samples to Qualcomm.
A Korean outlet, the Bell, reported that it had confirmed that Samsung had shipped next-generation LPDDR6X parts to Qualcomm.
Samsung has already wrapped up its LPDDR6 work and is expected to start mass production in the second half of 2026.
The first LPDDR6 parts are tipped to run at 10.7Gbps, with 21 per cent better efficiency than LPDDR5.
Samsung is also expected to crank out improved variants later that push speeds to 14.4Gbps and more.
LPDDR6X is the spicier take on LPDDR6, meant to stretch the standard further once the industry stops arguing about the details. JEDEC has not finalised the LPDDR6X specification yet, so the real numbers are still subject to a committee moving at committee speed.
The samples sent to Qualcomm are expected to feed into its AI accelerator chip called the “AI250”.
That part is meant to follow this year’s AI200, which also leans on LPDDR rather than the pricier memory stack.
The idea is to tackle AI inferencing with LPDDR, in the same rough lane as Troubled Chipzilla’s Crescent Island GPUs based on Xe3P chips.
Rivals such as NVIDIA, AMD, Huawei and the rest of the data centre crowd tend to use HBM, which is faster but costs more, guzzles power and is a pain to build.
HBM production is harder than DDR thanks to packaging, validation and testing, while LPDDR cuts some of that grief and keeps the bill down.
Qualcomm is expected to pack as much as 768GB of LPDDR on AI200, while AI250 is expected to push LPDDR6X capacity beyond 1TB.
Even with samples flying about, LPDDR6X is still a way off, with realistic timing pegged at late 2027 or early 2028.
