NVIDIA surpasses smartphone makers in LPDDR use

Nvidia is on track to become the biggest user of low-power DRAM, or LPDDR, overtaking smartphone makers that historically dominated that market, according to a May 18 report by Chosun citing Hana Securities and Citrini Research. The shift is tied to AI servers, where LPDDR is moving from a mobile-memory niche into a larger role alongside high-bandwidth memory used with graphics processors. The report said Nvidia’s AI servers are projected to consume 3.144 billion gigabytes of LPDDR in 2026 and 6.041 billion gigabytes in 2027. That would put Nvidia above Apple and Samsung Electronics combined next year, based on the estimates cited by Chosun. ### Why is Nvidia using a memory type associated with smartphones? LPDDR has traditionally been used in smartphones and laptops because it is designed for lower power consumption than conventional server memory. Reuters reported in November 2025 that Nvidia had decided to shift some AI servers to LPDDR from DDR5 to reduce server power costs. Counterpoint Research said at the time that the change could create a demand surge the industry was not equipped to absorb because each AI server uses far more memory than a handset. (chosun.com) The Chosun report said LPDDR is now emerging as a core memory component for AI servers, not just mobile devices. It described LPDDR as the memory that coordinates the entire server, while high-bandwidth memory remains attached to GPUs and has been central to the AI semiconductor market. ### What is the number that stands out most in the new estimates? (investing.com) Hana Securities and Citrini Research estimated that Nvidia’s AI servers will use 6.041 billion GB of LPDDR in 2027. The same estimates put Apple at 2.966 billion GB and Samsung Electronics at 2.724 billion GB, meaning Nvidia alone would exceed the two smartphone makers combined. (chosun.com) Chosun also said Nvidia could account for 36% of total LPDDR supply capacity next year. That figure, if realized, would mark a significant concentration of demand in one buyer category, with AI infrastructure taking a larger share of a market long tied to consumer electronics. That framing comes from the estimates cited in the report. (chosun.com) ### Which Nvidia product is driving the jump? Nvidia’s next-generation Vera Rubin platform is central to the forecast, according to Chosun. The platform combines the Rubin GPU with the Vera CPU, and the report said the Vera CPU is equipped with 1.5 terabytes of LPDDR. Chosun compared that with a typical smartphone’s 12 to 16 GB, putting the server-side memory load at roughly 90 times a handset’s capacity. (chosun.com) That hardware design helps explain why AI infrastructure can absorb mobile-style memory at a scale that rivals consumer-device makers. Reuters reported that Counterpoint viewed Nvidia’s LPDDR pivot as placing the company on the scale of a major smartphone maker, calling it a supply-chain shift that would be hard to absorb quickly. (chosun.com) ### Who else is adding pressure to LPDDR supply? Google and AMD were also cited by Chosun as increasing their use of LPDDR in AI hardware. The report said that as more AI chip companies adopt LPDDR, AI servers are becoming the largest source of LPDDR demand, surpassing smartphones. Reuters reported in November that Samsung Electronics, SK Hynix and Micron were already dealing with shortages in older DRAM categories after prioritizing high-bandwidth memory for AI accelerators. (investing.com) Counterpoint said then that supply tightness at the lower end of the memory market risked spreading upward if manufacturers diverted more factory capacity to LPDDR. (chosun.com) ### What does this mean for phones and laptops? Chosun said the immediate concern is supply availability, not a product launch. The report said that if AI server companies lock in large LPDDR volumes in advance, supply for smartphones and laptops could tighten. A tech-industry source quoted by Chosun said memory price increases were already affecting PCs, consoles and smartphones, and that AI server demand could intensify that pressure after 2027. (investing.com) Counterpoint’s earlier forecast was more specific on pricing. Reuters reported that the firm expected server-memory prices to double by the end of 2026 as Nvidia’s shift to LPDDR reshaped demand. ### What comes next in the timeline investors and suppliers are watching? (chosun.com) 2027 is the date embedded in the most aggressive demand forecast, with Vera Rubin named as the Nvidia platform behind much of the projected increase. Samsung Electronics, SK Hynix and Micron are the suppliers most directly exposed to any reallocation of capacity toward LPDDR, according to Reuters and Chosun. The next concrete milestone in this story is whether those 2027 demand estimates begin to show up in supplier capacity plans and customer allocations over the coming quarters. (investing.com) (chosun.com)