For years, Apple was one of the most feared buyers in the global component supply chain. Its volume of iPhones, iPads, Macs, and wearable devices allowed it to negotiate prices, demand priority, and push suppliers to compete for its business, which in turn granted these suppliers prestige and stability within Cupertino’s industrial machinery. In the DRAM memory market, however, that advantage is starting to erode.
The change isn’t due to a decline in Apple’s power, but rather the rise of an even more voracious buyer: artificial intelligence infrastructure. NVIDIA, Google, Amazon, Microsoft, Oracle, and other major cloud providers are consuming data center memory at a scale that shifts the balance of power from consumer devices to AI servers. Memory is no longer only negotiated with future iPhone models in mind but now revolves around computing platforms capable of consuming enormous quantities of HBM, DDR5, LPDDR, and enterprise SSDs.
This results in a silent reordering of the technological hierarchy. Apple remains a massive customer, but in certain memory segments, it no longer sets the rules as comfortably. Its priorities are shifting: less obsession with the lowest price and more urgency to ensure supply.
From Negotiating Discounts to Securing Wafers
The nature of these agreements is better understood by looking at how supply contracts used to function. According to Korea JoongAng Daily, before the AI era, memory supply contracts were shorter, less binding, and largely based on trust. Major buyers like Apple could breach or reduce purchases within annual agreements without significant consequences. Now, the market is moving toward longer, more stringent contracts of up to five years, as everyone wants to reserve capacity before someone else does.
The position of Samsung Electronics and SK hynix has shifted. For years, they relied heavily on the smartphone, PC, and consumer electronics cycles. Today, they sell to AI clients willing to pay higher prices, enter into longer agreements, and commit to massive volumes to feed data centers. If hyperscalers want to secure memory for their AI clusters, they can now compete directly with Apple for components that were once more tightly associated with phones and laptops.
The most notable case is HBM, the high-bandwidth memory that accompanies AI accelerators. Samsung, SK hynix, and Micron are dedicating more effort to these products because of higher margins and near-guaranteed demand from NVIDIA, AMD, and major cloud providers. However, this shift has collateral effects: as more production capacity is allocated to HBM and server memory, less remains for conventional DRAM, consumer LPDDR, or NAND for devices.
Reuters reported earlier in 2026 that Samsung and SK hynix warned of tighter supply for PCs and phones due to the AI boom, with Apple acknowledging that memory price impacts will increasingly affect its business. The pressure isn’t limited to a single category: it affects RAM, storage, server modules, and components linked to data centers.
| Before the AI Era | During the AI Boom |
|---|---|
| Apple could push prices | Apple competes with hyperscalers for supply |
| Shorter, more flexible contracts | Longer, binding agreements |
| DRAM mainly aimed at PCs and mobile | More capacity allocated to HBM, servers, and enterprise SSDs |
| iPhone volume was a major leverage | AI data centers consume memory at massive scale |
| The goal was to buy cheaply | The goal is not to run out of components |
LPDDR Is No Longer Just Mobile Memory
One of the most interesting aspects of this shift is in LPDDR. Traditionally, LPDDR memory has been associated with smartphones, tablets, and ultraportable laptops due to its low power consumption. For years, Apple has been one of the largest buyers of this type of memory. But AI is breaking that association.
NVIDIA has designed its Vera Rubin platform with an architecture that uses LPDDR5X in SOCAMM modules. According to NVIDIA’s technical documentation, Vera can integrate up to 1.5 TB of LPDDR5X per socket, with 1.2 TB/s bandwidth, using SOCAMM modules to improve service, fault isolation, and availability in AI factories. This changes demand scales: a server CPU may need so much LPDDR that it makes a smartphone’s consumption look tiny.
NVIDIA also confirmed that its first Vera Rubin instances will arrive in 2026 from providers like AWS, Google Cloud, Microsoft, and OCI. This isn’t just experimental architecture without clients; it arrives with the same players already competing for chip capacity, advanced packaging, networking, energy, and memory.
This directly impacts Apple. If higher-capacity, more reliable LPDDR is being designed with AI servers in mind, smartphone manufacturers lose some of their natural priority. Korea JoongAng Daily reports that the most advanced LPDDR is already being routed into SOCAMM formats for servers, integrated with AI CPUs, and that Apple is forced to purchase large volumes without the separate discounts it once negotiated.
The striking figure: each Vera CPU may require up to 1.5 TB of LPDDR5X, compared to just a few gigabytes in a typical smartphone. While hundreds of millions of phones continue to sell, AI infrastructure requires much more memory per system. When these demands meet, memory providers gain increased leverage.
Memory Pricing Becomes Part of Product Design
For Apple, this impact extends beyond margins to product strategy. If memory prices rise significantly, the company can respond in several ways: absorb costs, raise prices, keep base configurations leaner, reserve more memory for Pro models, or tweak storage options to maintain margins.
Apple has long managed these tensions, using memory and storage as commercial levers: base models with adjusted specs, capacity jumps at premium prices, and clear segmentation across tiers. But a supply crunch limits that flexibility. If availability becomes a critical issue, the entire product lineup could be affected.
This is especially true for upcoming iPhone, Mac, and personal AI device cycles. An iPhone with more local AI features demands more RAM. A Mac aimed at local models or advanced creative tools also needs more memory. If, precisely when Apple needs more memory, the market is competing with NVIDIA and hyperscalers, its negotiating room shrinks.
KB Securities, cited by Korea JoongAng Daily, estimates that DRAM and NAND prices could see significant year-over-year hikes in 2026, as AI data center operators take a larger share of memory shipments. While these remain forecasts, they align with retail trends: skyrocketing DDR5 kits, more expensive SSDs, and electronics manufacturers pre-ordering to avoid later price hikes.
It also clarifies why Apple might need to alter its supplier relationships. In a comfortable market, the big buyer pressures prices downward. In a constrained market, it reserves supply. The strategic difference: in the first scenario, lower prices matter; in the second, uninterrupted supply is paramount.
The New Power of Samsung, SK Hynix, and Micron
The paradox is that AI is strengthening memory manufacturers just when many believed that value lay primarily with chip designers. NVIDIA commands much attention, but its platforms depend on HBM, LPDDR, packaging, interconnects, and production capacity. Without enough memory, even the best architecture remains incomplete.
Samsung, SK hynix, and Micron are thus positioned more strongly. They can prioritize higher-margin products, secure long-term deals, and negotiate with now more balanced power dynamics. SK hynix leads in many AI HBM contracts; Samsung aims to regain ground; Micron has intensified its focus on advanced server memory.
This doesn’t mean Apple will run out of memory. Its scale, financial health, and historical relationships remain formidable. But it no longer dominates the top tier alone. NVIDIA, Google, AWS, Microsoft, and other buyers are willing to pay to stay in the next-generation AI race.
The effect might eventually reach consumers: if memory becomes more expensive and scarce, devices could cost more, offer less base capacity, or see delays in certain upgrades. Users might not notice the chip war directly, but they’ll feel it when laptops with more RAM cost more, or when base storage options stagnate, or specific configurations take longer to arrive.
Memory has ceased to be a discrete component; it’s now a strategic resource. During the smartphone decade, Apple represented consumer device purchasing power. In the AI era, that power is shared with data centers. And when a GPU or server CPU can consume as much LPDDR as thousands of phones, even Apple must negotiate differently.
Frequently Asked Questions
Why is Apple losing negotiating power in memory?
Because DRAM and NAND manufacturers now have AI clients with huge orders, long-term contracts, and greater willingness to pay. Apple remains a large buyer but no longer dominates demand as before.
What’s NVIDIA’s role in LPDDR memory?
NVIDIA’s Vera Rubin platform uses LPDDR5X in SOCAMM modules for AI servers. A Vera CPU can support up to 1.5 TB of LPDDR5X, increasing pressure on a memory type previously more linked to mobile and laptops.
Could this crisis cause iPhone or Mac prices to rise?
Yes, Apple might need to absorb costs, adjust configurations, or charge higher prices if memory costs stay high. While it normally protects margins via product segmentation, availability is now also an issue.
Who benefits most from these changes?
The biggest winners are memory manufacturers like Samsung, SK hynix, and Micron—especially for high-margin AI products like HBM, advanced LPDDR, server DRAM, and enterprise SSDs.
vía: koreajoongangdaily