NAND Flash contract prices were forecast to rise 70% to 75% quarter over quarter in Q2 2026 as AI-server demand redirected supply toward enterprise storage, according to TrendForce’s second-quarter memory pricing survey. The increase followed an 85% to 90% first-quarter forecast and left consumer-device manufacturers facing higher costs before substantial new production capacity can arrive.
The mechanism behind the shortage is visible in where the output is going. Enterprise SSDs absorbed 48% of global NAND bits in Q2, up from 26% a year earlier, as AI inference increased demand for high-capacity server storage. Counterpoint Research expects servers to consume more than half of all NAND bits by the end of 2026.
That shift helps explain why the price increase cannot be reduced to a temporary rebound in smartphone or PC demand. Suppliers are directing more NAND output toward enterprise SSDs, while the investment decisions needed to add meaningful capacity will take years to affect production.
Enterprise SSDs Are Taking Nearly Half of NAND Supply
Training large AI models concentrates attention on accelerators and high-bandwidth memory, but inference creates another storage requirement. Models serving large numbers of users need rapid access to datasets, embeddings and cache data, increasing demand for high-capacity enterprise SSDs alongside conventional server memory.
Counterpoint said the migration from training toward inference pushed enterprise SSDs to 48% of NAND bit shipments during Q2. The resulting shortage left less supply available for consumer products and drove consumer NAND selling prices to record levels.
The effect has reached retail storage and device pricing. IDC said manufacturers including Lenovo, Dell, HP, Acer and Asus had warned customers about higher PC costs, with some price adjustments and contract resets reaching 15% to 20%. IDC’s downside scenarios put the potential increase in PC average selling prices at 4% to 8%, depending on how long the shortage lasts.
Smartphone manufacturers face a similar problem because memory can account for 15% to 20% of the bill of materials in a mid-range handset. Higher NAND and DRAM costs leave manufacturers choosing between raising retail prices, reducing storage and memory specifications, or accepting lower margins.
NAND and DRAM Are Tight for Different Reasons
The two memory markets are connected, but the supply mechanisms should not be treated as identical. High-bandwidth memory uses DRAM production capacity, so increased HBM output directly restricts the wafers and cleanroom space available for conventional DRAM used in PCs, smartphones and general-purpose servers.
NAND pressure is driven more directly by enterprise SSD allocation, restrained capital spending and suppliers’ preference for higher-value storage products. TrendForce has also said some manufacturers are directing resources toward the more profitable DRAM market, limiting the speed at which NAND capacity can expand.
Conventional DRAM contract prices increased approximately 93% to 98% in Q1, according to TrendForce’s post-quarter industry analysis. The firm then forecast another 58% to 63% increase for Q2 as suppliers continued prioritizing servers and HBM.
The concentration of DRAM production gives those allocation decisions broader consequences. Samsung, SK Hynix and Micron account for close to 90% of the market, leaving device manufacturers with few alternative sources when the three producers prioritize AI-related products. The same supply pressure has supported the investment case around Micron as its revenue and margins rise with memory prices.
China’s CXMT is adding conventional DRAM production, but it does not yet provide a near-term substitute for the leading HBM suppliers. Its expansion, including the capacity plans discussed around its Shanghai market debut, is more relevant to commodity DRAM than to the advanced HBM products driving AI accelerator deployments.
Price Growth Is Slowing, but Prices Are Not Falling
TrendForce’s July outlook forecast that NAND contract-price growth would moderate to 10% to 15% in Q3. The slowdown was attributed to record price levels, weaker consumer demand and resistance from PC and smartphone manufacturers that had reached the limits of what they could absorb.
That is a deceleration from the 70% to 75% Q2 forecast, not a reversal. The distinction matters because a lower rate of increase can coexist with record prices, reduced transaction volumes and limited availability for buyers outside long-term supply agreements.
Spot-market figures also require care. A Bloomberg-reported increase of nearly 700% applied to selected legacy DRAM spot products rather than the entire DRAM market or the contract prices paid by large manufacturers. Spot prices can move more sharply because they cover incremental supply outside negotiated contracts.
Sourceability separately reported that NAND prices had risen 246% from the beginning of 2025 through December, attributing the figure to Kingston. That measure should not be combined directly with TrendForce’s quarterly contract-price ranges because the periods, products and transaction channels differ.
New Capacity Cannot Respond Quickly
Memory manufacturers cannot resolve the shortage simply by increasing output at existing facilities. Cleanroom construction, equipment installation, process qualification and yield improvement can take several years before a new fab reaches volume production.
TrendForce said meaningful NAND capacity expansion was unlikely before late 2027 or 2028. IDC also expects the memory shortage to persist well into 2027, with 2026 DRAM and NAND supply growth remaining below historical levels at 16% and 17%, respectively.
Most near-term supply growth must therefore come from process migrations that increase the number of bits produced from each wafer. Those gains help at the margin, but they do not match the increase in enterprise storage demand when AI servers are taking a rapidly growing share of total output.
The delay also limits the speed at which producers can respond to high prices without creating another oversupply cycle. Memory companies spent years cutting output and capital expenditure after the previous downturn, making them cautious about approving capacity that may enter production after market conditions have changed.
SanDisk’s Forecast Is Aggressive but Has Supply Data Behind It
SanDisk used its 13 August Investor Day to estimate that NAND industry revenue would exceed $300 billion in 2026 and reach $500 billion in 2027, with supply remaining allocated beyond that year. The company’s forecast is substantially more aggressive than a normal cyclical recovery and should be treated as management guidance rather than independent market consensus.
However, the supporting operating figures explain why investors gave the forecast weight. SanDisk generated approximately $8.97 billion in fiscal fourth-quarter revenue, up 51% sequentially, with a non-GAAP gross margin of 84.6%. Management attributed roughly two-thirds of the sequential revenue increase to pricing and one-third to higher volume.
The company also said demand from customers was growing faster than its available supply. Its longer-term agreements covered about half of expected fiscal 2027 bit shipments and approximately two-thirds of fiscal 2028 shipments, giving it more revenue visibility than memory producers typically have during a price cycle.
Those disclosures extend the supply thesis examined before SanDisk’s Investor Day and after options markets priced a large move around its results. They also help explain the read-across to Western Digital’s storage business, even though the companies now carry different direct exposures following their separation.
The central question is no longer whether memory prices have risen. It is whether enterprise demand can remain strong enough to absorb production gains before new facilities reach volume output. With enterprise SSDs already taking 48% of NAND bits and major capacity relief still years away, suppliers retain pricing power even as consumer resistance slows the rate of increase.
