Not Enough Memory to Go Around: NVIDIA Sets Its Sights on Chip Supplies Meant for Apple and Samsung

Not Enough Memory to Go Around: NVIDIA Sets Its Sights on Chip Supplies Meant for Apple and Samsung

Arkadiy Andrienko

Analysts at Citrini Research have released a forecast that makes you rethink the balance of power between consumer electronics and server hardware. According to their latest estimate, by 2027 a single server platform (NVIDIA Rubin) will consume more mobile RAM than Apple and Samsung’s combined purchases for their smartphones.

We’re talking about LPDDR memory, which has traditionally been associated with phones and laptops. But it’s precisely its high cell density, low heat output, and energy efficiency that have made it increasingly popular in new types of compute racks for AI workloads. NVIDIA’s Rubin platform in 2027 could require over 6,000 million GB of LPDDR memory; for comparison, Apple’s expected need for its iPhone lineup is estimated at 2,966 million GB, and Samsung’s at 2,724 million GB.

NVIDIA to swallow massive memory volumes
NVIDIA to swallow massive memory volumes

But NVIDIA isn’t the only player in this segment — AMD’s solutions, including MI400 accelerators, Verano processors, and MI455X chips for Helios racks, which are also designed to use LPDDR6X, will soak up plenty of memory. Key DRAM suppliers aren’t waiting for 2027; they’re already taking action. Micron has announced SOCAMM2 modules based on LPDDR6X with 256 GB capacity, aimed specifically at the server segment. SK hynix, for its part, has started mass production of 192 GB SOCAMM2 LPDDR6X modules, which will be used in NVIDIA Vera Rubin. At the same time, Samsung, SK hynix, and Micron are ramping up production capacity, but as the forecast points out, existing capacity may not be enough to cover both explosive demand from AI infrastructure and steady orders from the mobile market.

The situation sets the stage for intensified competition among buyers. When server manufacturers start claiming volumes comparable to the needs of smartphone industry leaders, it inevitably affects component availability. Apple and Samsung, while still the largest consumers of LPDDR for their phones, could face chip shortages and, as a result, higher procurement prices — not to mention smaller manufacturers. Ultimately, this dynamic could impact the final cost of devices or the amount of RAM in future models.

Micron SOCAMM2 memory stacks
Micron SOCAMM2 memory stacks

Citrini Research’s forecast captures a major turning point: memory that evolved for decades under the influence of the mobile market is now becoming a strategic resource for AI infrastructure. The open question is how quickly chipmakers can expand output to satisfy both sectors without creating price distortions.

In your opinion, how much could rising demand from AI servers affect smartphone retail prices over the next couple of years, and should we expect phone makers to start booking memory capacity in advance — to the detriment of smaller players?

    About the author
    Comments0