The unstoppable rise of artificial intelligence is starting to take its toll, and it’s not just about energy consumption, but a real hardware resource crisis that risks reshaping the consumer electronics economy.
According to a recent note from KeyBanc analysts, the architectural changes implemented by NVIDIA in its new Vera Rubin series GPUs are expected to consume a significant portion of global memory resources this year.
In a market already characterized by limited supply, this memory ‘hunger’ from the AI chip giant is creating a cascading effect that will hit smartphone manufacturers hard, including giants like Apple and Samsung, forcing the market toward inevitable price increases.
NVIDIA Vera Rubin AI GPUs won’t settle for memory that ‘advances’
At the center of this market turmoil is a specific technical choice adopted by NVIDIA. Processing AI queries requires creating an enormous temporary memory register for building context, technically known as “KV Cache“.
Until now, in NVIDIA systems, this register was stored inside the HBM memory modules. However, with the Vera Rubin architecture, the company has introduced a new solution called Inference Memory Context Storage (ICMS), which serves as a memory resource dedicated exclusively to that purpose.
This novelty, though technically brilliant, has massive consequences on the supply chain. Estimates indicate that NVIDIA could integrate about 16 TB of NAND memory per GPU in a rack, which translates into a staggering total of 1,152 TB in a single NVL72 configuration.
KeyBanc provided a concerning context to these numbers: the new Vera CPU uses 1.5 TB of memory, a significant jump from the 512 GB of the previous Grace architecture. This implies that NVIDIA will require about 20 billion Gb of memory this year. To put this figure in perspective, such consumption equals the memory needed to produce between 100 and 150 million smartphones, i.e., just under 10% of the entire global mobile phone market.
Apple and Samsung in the crosshairs
The repercussions of this massive allocation of resources toward AI are also putting pressure on tech giants that traditionally dominate supply chains.
Apple, for example, is facing a challenging situation. According to Morgan Stanley’s assessments, although the Cupertino company has secured access to NAND resources sufficient until Q1 2026, the supplier KIOXIA is expected to raise prices once the new long-term supply agreement is signed.
Even more critical is the situation regarding DRAM memory, which constitutes a significant bottleneck. Although Apple had previously signed very favorable long-term contracts, current market conditions suggest the company will secure access to DRAM resources until early 2026 only by accepting a sequential price increase exceeding 50%.
Paradoxically, not even Samsung, one of the world’s leading memory manufacturers, is immune to this price-driven demand destruction dynamic.
Internal reports suggest tensions between the divisions of the Korean conglomerate: the DS division, which makes memory chips, has agreed to supply components to the MX mobile division only in exchange for price increases between 60% and 70%.
This internal inflation has pushed Samsung to plan a price increase for the upcoming Galaxy S26 series, estimated at between 30 and 60 dollars in selected regions, including South Korea.
