AI reasoning triggers the NAND super cycle: demand surges by 280%. Why has SanDisk become the biggest beneficiary?

The semiconductor narrative of 2026 is undergoing a silent but profound restructuring. If the past two years saw all the spotlight on AI hardware focusing on NVIDIA's GPUs and HBM high-bandwidth memory, then entering 2026, the market’s focus is expanding into the deep supply chain—storage, especially NAND flash, is evolving from a supporting role in AI infrastructure to a decisive strategic bottleneck.

This shift in perception is driven by the scale deployment phase of AI. In the early industry days, the public discourse centered on training large models—massive computational power consumption, cluster computations spanning months—leading to the first wave of hardware centered around GPUs. But training models is just the starting point of commercializing technology. As AI moves from labs to daily interactions with millions of users, what is truly being continuously consumed is not just training compute, but storage.

Every user interaction with a large model generates vast amounts of data that need to be stored persistently. For example, a large language model with a context window of 128,000 tokens can produce about 61GB of cached data in a single conversation. The ongoing inference process—where the model runs for extended periods, continuously processing massive user interactions and generating ever-growing contextual information—means each interaction produces intermediate states, KV caches, and compliance records that require persistent storage. McKinsey’s research indicates that the sustained growth in NAND demand is mainly driven by AI model training, inference, retrieval-augmented generation, and the widespread adoption of multimodal models. The most direct evidence of this is the forecasted size of the global NAND market. TrendForce has sharply raised its 2026 global NAND flash revenue forecast to $270.6 billion, with a high annual growth rate of 280.7%; in 2027, the market could further expand to nearly $379.4 billion, maintaining a growth rate of 40.2%. A more intuitive comparison: NAND market revenue in Q1 2026 already exceeded that of all of 2023. Unlike the relatively concentrated data generation patterns during training, inference scenarios demand ongoing and compounded storage—each new user scenario adds a layer of storage requirement, rather than diminishing cyclically.

As AI shifts from training to inference, the logic of infrastructure investment is undergoing a fundamental change—NAND storage is evolving from a cost center to a value center.

NAND Demand in Inference Scenarios: From Storage to Strategic Asset

Understanding this proposition requires starting from the operational mechanism of AI inference. Traditional data center workloads typically demand storage with a "write once, read many" characteristic, but in AI inference, the situation is entirely different. Each AI model during inference needs continuous access to training datasets, model weights, KV caches, and user history data. These data cannot be fully stored in the expensive, capacity-limited DRAM and must rely on large-scale NAND flash as a persistent storage layer.

NVIDIA’s blueprint for next-generation AI infrastructure announced in early 2026—the Inference Context Memory Storage platform and BlueField 4 data processing units—clearly depict this shift. According to Korean media reports, NVIDIA explicitly states that as AI industry transitions from training to inference, the role of storage devices will rapidly grow, becoming as strategically important as compute performance. NVIDIA’s new AI architecture not only leverages high-speed memory near GPUs but also incorporates large-capacity external SSDs into the inference pipeline. Essentially, this elevates storage devices from auxiliary tools to core components of AI inference.

This results in a trend industry calls "massive compute-to-storage capital rotation." Market data shows enterprise SSD contract prices have surged up to 80% over the past three quarters, signaling the start of a structural NAND supercycle driven by AI data demands. JPMorgan’s latest research significantly upwardly revised its global storage market forecast, expecting the total market to reach $1.7 trillion by 2028, with NAND revenues rising from $71 billion in 2025 to over $400 billion in 2026. More critically, JPMorgan notes that storage chips are shifting from traditional cyclical commodities to strategic core assets in AI infrastructure.

Supply-side dynamics are equally noteworthy. Samsung Electronics and SK Hynix have adopted proactive supply control strategies, aiming to keep NAND wafer output at lower levels while prioritizing capex for high-margin HBM products. This structural contraction on the supply side, combined with rapid demand growth, is reshaping NAND industry pricing logic. Historically, NAND flash suffered from oversupply and price volatility due to lower technical barriers, but the rapid penetration of enterprise QLC SSDs and high-bandwidth NAND technologies is redefining profit structures. An industry insider states that for current NAND flash, maintaining prices and restoring profitability is more rational than blindly expanding production.

Market Landscape: The Battle of Six Giants, Samsung Leading but Uncertain

According to the Q1 2026 NAND flash market tracking report from Counterpoint, the global NAND market revenue hit a record $46 billion, up 246% year-over-year and 90% quarter-over-quarter. Enterprise SSDs accounted for 43% of total NAND market volume, with expectations to surpass 60% by the end of 2026.

In terms of competitive landscape, Samsung holds the top spot with 29% market share; SK Hynix is second with 18%. The fight for third place is fierce—Kioxia holds 14%, Micron, SanDisk, and Yangtze Memory all at 13%, forming a tight pack. Yangtze Memory was the standout performer this quarter, with revenue up about 445% YoY, increasing its market share from 8% to 13%. According to Counterpoint’s research director Hwang, if Yangtze Memory successfully IPOs, it will have the capacity to expand production and scale further, potentially surpassing Kioxia and Micron to become the third-largest NAND manufacturer globally.

Notably, Kioxia has also performed strongly in the capital markets, with its stock price hitting new highs and briefly surpassing Toyota’s market cap to become Japan’s second-largest listed company. In fiscal 2025, Kioxia’s revenue reached ¥2.34 trillion (~$17k), up 37% YoY, with net profit of ¥876.2 billion (~$23.4k), a 93.4% increase.

Samsung, Micron, and SK Hynix have crossed the trillion-dollar market cap threshold. Industry insiders believe that data storage, as the foundational resource for AI systems, is becoming a key pillar of the AI era. Previously viewed as cyclical and volatile, storage chips are now entering a longer, structurally high-growth phase.

From a technological route perspective, major players are accelerating the deployment of high-performance NAND products for AI inference scenarios. Kioxia has launched high-speed SSDs targeting AI workloads, aiming for up to 10 million IOPS, and plans to develop a 5TB high-bandwidth flash module prototype in 2025 with 64GB/s bandwidth, attempting to directly connect NAND to GPU memory buses. Samsung continues to iterate its 9th-generation V-NAND, focusing on increasing I/O speeds and storage density. SanDisk showcased its redesigned Optimus series consumer and enterprise SSDs at CES 2026, covering content creators, gamers, and AI PCs.

All these paths point in the same direction: NAND flash is being redefined from a "cheap storage semiconductor" into a core component of AI infrastructure. As NAND’s role expands from simple data archiving to storing and managing large-scale intermediate data and contextual information during AI inference, its strategic value undergoes a fundamental transformation.

SanDisk (SNDK): The Pure-Play NAND Titan’s Independent Dividend and AI Opportunities

In the supercycle of NAND, SanDisk (ticker: SNDK) is one of the most talked-about stocks. On February 24, 2025, SanDisk officially completed its spin-off from Western Digital, becoming an independent, publicly listed pure flash memory company. This timing is highly strategic—post-spin, SanDisk is riding the wave of the generative AI-driven "storage supercycle," free from the valuation drag of HDD business over the past decade.

Fundamentally, SanDisk is experiencing explosive growth. According to recent financials, in Q3 2026, SanDisk’s revenue surged 251% YoY, with EPS exceeding analyst consensus by over 63%. For the first nine months of fiscal 2026, adjusted EPS reached $31.32, with full-year estimates around $62.82. Wall Street analysts’ consensus for FY2027 EPS has already risen to approximately $175.

Supply-side tightness also underscores market demand for SanDisk’s products. Reports indicate that all capacity for SanDisk in 2026 has been sold out, with 2027 capacity rapidly booked. Customers are shifting from quarterly price negotiations to multi-year agreements to lock in supply, which could significantly reduce the cyclicality of SanDisk’s valuation.

Zacks Research’s latest forecast further reinforces this outlook: it expects SanDisk’s revenue and profit growth in FY2027 to both exceed 100%, with analyst consensus EPS estimates being raised by 76.1% over 60 days. Charles Schwab even lists SanDisk as one of the core drivers of S&P 500 earnings growth in 2026.

The underlying logic supporting this rapid growth is NAND’s irreplaceable role in AI inference scenarios. Each deployment of AI inference workloads requires multiple times the NAND storage compared to traditional enterprise workloads. As cloud providers and tech giants continue expanding inference infrastructure, enterprise SSD demand is being systematically amplified. JPMorgan projects that in 2026, the enterprise SSD market will exceed 500 exabytes (EB), accounting for 43% of total NAND demand, and is expected to grow at a 52% CAGR to over 1100EB in the next two years.

Of course, the rapid market rise also raises valuation questions. Based on 25 analyst 12-month price targets, the average target price for SanDisk is about $1,398, with a most optimistic target of $3,250, but a minimum of only $63—showing wide divergence. Some, like Morningstar, suggest current prices may already contain bubble risks. But from fundamental trends and structural industry shifts, SanDisk’s growth prospects seem underappreciated.

Gate Stock Trading: An Investment Gateway Connecting Digital Assets and Storage

Against the backdrop of continuously rising storage demand driven by AI inference, investing in US stocks related to the NAND supply chain has become increasingly strategic. Gate platform officially launched stock trading services on June 1, 2026, aiming to create a unified trading hub connecting crypto assets with mainstream financial products. Users can directly trade US stocks and ETFs in USDT within their Gate accounts.

This product suite features several notable aspects. First, Gate’s stock trading service is integrated with a compliant brokerage infrastructure (partnering with US broker Alpaca), supporting over 10,000 stocks and ETFs on NYSE and NASDAQ, covering core sectors like tech, semiconductors, and finance. When users buy, they hold actual shares, with full shareholder rights including dividends and stock splits, consistent with traditional brokerage models.

Second, in terms of product structure, Gate stocks share the same account system as crypto assets. Users can settle and fund with USDT, and even trade fractional shares as small as 0.01, with a minimum investment of $1—significantly lowering the capital barrier. Transaction fees can be as low as 0.023%, with no funding rates, swaps, or overnight fees.

For users interested in storage industry investments, Gate stock trading offers a convenient channel to track related stocks—from upstream NAND giants like SanDisk (SNDK), Micron (MU), to storage industry players like Western Digital (WDC), Kioxia (listed on Tokyo Stock Exchange)—all accessible on one platform.

More broadly, the launch of Gate stock trading signals a strategic move: as crypto and traditional finance continue to converge, a super-financial platform integrating cryptocurrencies, stablecoins, and mainstream securities is taking shape. For users wanting to participate in both crypto ecosystems and storage industry cycles, Gate provides a compliant, transparent, and streamlined gateway.

Conclusion

The story of AI is no longer just about GPUs. As inference deployment scales, NAND flash is becoming the second fundamental infrastructure bottleneck after compute. From surpassing 2023’s annual revenue in Q1 2026 to enterprise SSD prices soaring 80% in three quarters, and TrendForce raising its NAND market forecast to $270.6 billion—these data points point to a clear trend: storage demand is underestimated, and this underestimation is being corrected.

In this structural shift, SanDisk, as a pure NAND play, with 251% quarterly revenue growth, full capacity sell-out, and far exceeding industry profit expectations, offers a forward-looking indicator of AI inference storage demand strength. Meanwhile, the launch of Gate stock trading provides a compliant, accessible channel for crypto-native users to directly participate in this unfolding storage supercycle.

As industry witnesses, AI inference is no longer just about computation—storage is writing its own growth narrative. The cognitive leap from "AI is just GPU" to "storage is a strategic asset" may just be beginning.