NVIDIA’s next-generation Vera Rubin AI GPUs are shaping up to do more than boost AI performance—they could also intensify the global memory shortage at a time when supplies are already tight. A new note from KeyBanc suggests these architectural changes may soak up a massive amount of memory capacity this year, creating ripple effects across the entire tech supply chain. The consequence for everyday consumers could be straightforward: higher smartphone prices in 2026, even from brands as large as Apple and Samsung.
Vera Rubin GPUs could absorb memory equal to roughly 10% of the global smartphone market
The key issue is how AI inference workloads handle “context.” When you ask an AI model questions, it often builds and maintains a running record of context so it can respond intelligently. This temporary context record is commonly associated with something called the KV cache, and it requires substantial memory resources.
Today, NVIDIA’s AI systems largely keep this kind of working data in high bandwidth memory (HBM). With Vera Rubin, NVIDIA is shifting to a new approach called Inference Memory Context Storage (ICMS), designed as a dedicated memory resource for that AI context.
What makes this especially significant is the scale. With the new design direction, AI racks could be populated with enormous amounts of NAND storage—on the order of about 16TB of NAND per GPU in a rack, adding up to as much as 1,152TB in a single NVL72 configuration. That kind of demand doesn’t just affect the AI industry; it competes directly with the same memory supply that smartphones, PCs, and other consumer electronics rely on.
KeyBanc takes it a step further, estimating that NVIDIA’s latest Vera CPU uses about 1.5TB of memory (up from 512GB on Grace). Based on that, the firm estimates NVIDIA may require around 20 billion gigabits of memory this year—an amount it compares to the memory used in roughly 100 million to 150 million smartphones. That’s just under 10% of the total smartphone market, redirected toward AI infrastructure.
Why Apple and Samsung could feel the squeeze
If a major AI buildout pulls a large share of global memory supply into data center deployments, the resulting scarcity can push memory prices higher. And since memory is a meaningful portion of a handset’s cost, those increases can quickly show up in retail pricing.
Apple: NAND and DRAM pressures, plus a surprising materials bottleneck
On the NAND side, Apple is said to have secured enough supply through the first quarter of 2026, but there are expectations that pricing could rise once longer-term agreements are finalized.
DRAM could be an even tougher challenge. Apple previously benefited from favorable long-term DRAM contracts, but the latest outlook points to Apple being able to secure DRAM supply through Q1 2026 only if it accepts a sequential price increase of more than 50%.
Apple is also dealing with strain elsewhere in the supply chain—specifically in high-end glass fiber cloth known as T-glass, a specialized fiberglass material used in chip substrates. Substrates are essential: they act as the base for placing integrated circuits and components, while also helping with heat dissipation. T-glass, with its high silica content, can offer better thermal stability, a flatter surface for advanced wiring, and improved reliability.
Apple reportedly buys most of its T-glass from Japan’s Nitto Boseki (Nittobo), but AI-driven demand has pushed supply capacity to the limit. The company is said to be exploring alternatives and even considering lower-grade material in the interim—though switching comes with long testing cycles and potential performance or reliability implications.
Samsung: even a memory maker isn’t insulated
Samsung’s situation highlights how unusual the current memory market has become. Despite producing memory chips in-house, Samsung is still impacted by the same pricing dynamics affecting the industry. Anecdotal reports indicate Samsung’s semiconductor division has raised DRAM prices charged to its own mobile division by roughly 60% to 70%.
That kind of internal cost pressure has reportedly contributed to regional price increases for Samsung’s upcoming Galaxy S26 series, with estimates of around $30 to $60 higher in certain markets, including South Korea.
What this could mean for 2026 smartphone prices
Memory is not a minor line item for phone makers. KeyBanc estimates it represents about 20% of a smartphone’s bill of materials. If NAND and DRAM pricing continues to rise due to tight supply and surging AI demand, the simplest lever for phone brands is to raise retail pricing.
KeyBanc’s view is that major smartphone OEMs may ultimately need to raise handset prices by roughly $100 to $150 to offset the cost shocks—an increase that could dampen consumer demand, particularly in price-sensitive markets.
Bigger picture: AI infrastructure vs. consumer electronics
This is the new collision point in the tech industry: hyperscale AI buildouts require staggering amounts of memory, and that demand doesn’t occur in isolation. When a single AI platform shift can consume memory capacity comparable to tens of millions of smartphones, the effects can spread quickly—from component pricing to supply allocation decisions and, eventually, the price tag consumers see at launch.
If current forecasts hold, 2026 may be the year shoppers notice that AI’s growth isn’t only happening in the cloud—it’s also quietly reshaping what smartphones cost.
