The world’s most powerful consumer electronics buyer has, in a rare move, abandoned its cost-squeezing posture. Apple is reportedly in talks with Japanese memory giant Kioxia on a three-to-five-year NAND flash memory long-term supply agreement, and the contract may not include a fixed price ceiling. This 180-degree shift in procurement strategy is driven by aggressive competition for memory capacity from AI data centers, which is transferring memory chip pricing power from buyers to sellers.

South Korean media reports indicate that Apple has historically employed a “price suppression + multi-source” short-term procurement model with NAND suppliers, leveraging its massive purchasing scale to dominate negotiations. This pivot toward “volume lock-in + long-term contracts” completely breaks from the tech giant’s years-long practice. Multi-year long-term supply agreements were previously concentrated in the high-bandwidth memory (HBM) segment; Apple’s move into NAND long-term contracts signals a fundamental shift in the balance of bargaining power in the memory market.

The counterparty identity, purchase volume, and pricing of the Apple-Kioxia agreement have not been disclosed, and Apple has not confirmed the reports. Market observers view Kioxia as the most likely partner, while Kioxia and SK hynix are reportedly among the few major suppliers that have already adopted such agreement structures.

AI Squeezes Capacity as Memory Costs Approach End-Device Prices

The core reason Apple is willing to accept higher costs is that AI-driven memory capacity constraints have already transmitted cost pressure to end devices. Data center customers are willing to sign long-term contracts and pay high premiums, forcing Apple to lock in NAND supply years in advance to ensure stable memory sourcing for future flagship products.

According to TrendForce estimates, by Q3 2026, memory costs for the 256GB iPhone 18 Pro could be nearly four times higher than a year earlier. Memory’s share of the bill of materials (BOM) has climbed from roughly 10% in early 256GB Pro models to approximately 34%, and could exceed 40% in the first half of 2027.

The surge in memory prices is also reflected in the pricing of Apple’s most premium products. Apple’s first foldable iPhone (expected to be named iPhone Ultra) is set to start at approximately $2,099–$2,299 (about NT$66,000–NT$72,000), with high-capacity configurations potentially exceeding $3,000 (about NT$94,000). TrendForce notes that new model pricing is expected to rise approximately 10% to 20%, with Apple absorbing part of the cost increase to maintain market share.

Long-Term Contract Model Spreads — Samsung Plans 60–70% Capacity Allocation

Apple’s pivot is not an isolated event. TrendForce points out that Samsung Electronics, SK hynix, and Micron are likely candidates for future long-term contracts, covering LPDDR5X and LPDDR6 memory for upcoming iPhones. The market is watching whether Apple will extend its long-term contract strategy to the mobile DRAM segment.

Samsung reportedly plans to allocate 60% to 70% of total capacity to long-term contracts while maintaining production flexibility. NAND supply is more fragmented, with SK hynix expected to account for approximately 30% of foldable device NAND supply and Samsung 15%, totaling 45% between the two — leaving the remaining share as room for expanded cooperation between Kioxia and Apple.

In the foldable iPhone’s DRAM supply chain, Samsung Electronics and SK hynix are expected to jointly supply 70% of the 12GB LPDDR5 DRAM in the iPhone Ultra, with Samsung at 37% and SK hynix at 33%, while the remaining 30% is supplied by Micron.

Memory Industry Rides the Wave — Taiwanese Manufacturers Benefit

The tight memory supply environment is already reflected in industry revenue. TrendForce’s latest research shows that in Q2 2026, driven by significant increases in commodity DRAM contract prices, overall DRAM industry revenue surged 59.5% quarter-over-quarter to nearly $154.73 billion (approximately NT$4.9 trillion).

Among the three major manufacturers, Samsung Electronics — the first to ramp HBM4 mass production and shipments — posted Q2 revenue of $60.98 billion (approximately NT$1.9 trillion), up 63.4% quarter-over-quarter, ranking first with a 39.4% market share. SK hynix recorded Q2 revenue of $38.59 billion (approximately NT$1.2 trillion), up 37.9% quarter-over-quarter, with a 24.9% share. Micron’s revenue rose 65.5% quarter-over-quarter to $36 billion (approximately NT$1.1 trillion), with a 23.3% market share.

Taiwanese memory manufacturers are benefiting as well. Nanya Technology (2408.TW) posted Q2 revenue up 68.3% quarter-over-quarter to $2.612 billion (approximately NT$82 billion), Winbond Electronics (2344.TW) rose 75.8% to $998 million (approximately NT$31 billion), and Powerchip Semiconductor Manufacturing (6770.TW) saw consumer DRAM revenue surge 167.8% quarter-over-quarter to $115 million (approximately NT$3.6 billion).

Investors Need to Recalibrate the Silicon Cycle Framework

For investors, there is a significant expectations gap behind this memory rally. The market is still pricing NAND under the traditional “silicon cycle” framework, assuming that price increases will inevitably be followed by overcapacity and correction. But AI-driven capacity constraints are long-term and structural in nature, meaning the peak height and duration of this memory cycle may exceed traditional expectations.

The launch of OpenAI’s latest model, ChatGPT-6 Astra, has further reinforced market optimism about memory demand. Goldman Sachs and Morgan Stanley analysts estimate that of the projected $1.3 trillion to $1.5 trillion in AI capital expenditure for 2027, more than half will be directed toward memory. South Korea’s Kospi index has rebounded approximately 25% from its late-July low, approaching 7,000 points, with Goldman Sachs’ chief Asia equity strategist reiterating a year-end target of 12,000 points.

The structural transformation of the memory industry is spreading from the server side to the consumer electronics side. When the world’s most powerful buyer begins signing long-term contracts without price caps, the market may need to rethink where this memory supercycle ultimately ends.