AI reasoning is reshaping the NAND cycle? What Sandisk means for memory chip stocks
- Key Takeaway: AI inference is pushing storage demand beyond traditional compute resources into high-capacity NAND flash. Long-term customer agreements and new technologies (such as HBF and QLC) may reduce NAND cycle volatility, but will not eliminate its cyclicality.
- Key Elements:
- AI inference requires large-scale, fast-access, and cost-efficient storage, positioning NAND as an AI capacity tier alongside HBM and DRAM, rather than merely a traditional commodity storage product.
- Sandisk projects enterprise data center flash demand to reach 1.2 ZB by 2030 and is developing High Bandwidth Flash (HBF) purpose-built for AI inference, targeting near-HBM bandwidth at roughly eight times the capacity.
- Sandisk has signed eight long-term agreements totaling approximately $93.9 billion, covering about two-thirds of FY2028 output; Samsung has also indicated that long-term agreements could account for 60%-70% of future memory sales, improving demand visibility.
- Due to its high capacity and cost efficiency, QLC NAND may impact AI storage sooner than HBF; Samsung's BV-NAND architecture can improve storage density by approximately 58%.
- AI increases long-term bit demand and improves supply discipline, but NAND remains capital-intensive. Capacity expansion and technology upgrades could lead to oversupply—the cycle will not disappear, though its amplitude may narrow.
Key Takeaways
AI inference is extending the storage investment thesis from GPUs and HBM to high-capacity storage. Growing flash demand, long-term customer agreements, and new technologies like HBF could elevate NAND's importance in AI infrastructure and potentially dampen the volatility of traditional storage cycles.
Takeaway
- AI inference requires more than just compute power; large-scale deployment also demands ever-increasing, rapidly accessible, and cost-efficient storage capacity.
- NAND is progressively becoming the AI capacity layer alongside HBM and DRAM, rather than just a traditional commodity storage product.
- Sandisk projects enterprise data center flash demand to reach 1.2 ZB by 2030 and is developing High Bandwidth Flash specifically designed for AI inference.
- Storage vendors like Sandisk and Samsung are adopting longer-term customer agreements, which could improve demand visibility and reduce the risk of over-expansion.
- AI won't make the NAND cycle disappear. The key question is whether stronger bit demand and better supply discipline can reduce the amplitude of future cycles.
Throughout the AI boom of recent years, the semiconductor investment thesis followed a relatively clear path: more AI models require more GPUs, GPUs require more HBM, and increasingly dense compute clusters demand faster networking and more power. Storage has always been important but has rarely received the same level of attention as these other areas.
AI inference is gradually changing this allocation of resources.
As AI shifts from training large models to serving billions of queries, enterprise AI agents, multimodal applications, and real-time workloads, infrastructure must not only perform computations but also continuously store and repeatedly access vast amounts of data.
This presents a new question for the storage industry: Could NAND become the next major capacity bottleneck in AI infrastructure?
Recent developments from Sandisk, Samsung, Kioxia, and other storage vendors suggest the answer is increasingly leaning toward yes. Consequently, the NAND investment thesis is shifting from short-term flash price increases toward more structural AI-driven storage demand.
Why Does AI Inference Require More Storage Capacity?
AI training and AI inference place different kinds of pressure on infrastructure.
Training frontier large language models requires massive compute clusters and extremely high-speed memory access, which is why GPUs and HBM were the first to become critical bottlenecks in the AI supply chain.
Inference occurs after a model is deployed. Whenever an AI system answers a question, searches a knowledge base, analyzes an image, or executes an agent task, it may need to read model parameters, embeddings, cached data, user context, and increasingly massive multimodal datasets.
As the number of queries increases, the volume of data that needs to remain economically accessible grows in tandem.
Storing all this data in HBM is not practical. HBM offers extremely high bandwidth but has relatively limited capacity and a much higher cost per bit than NAND. Traditional NAND sits at the other end of the spectrum, offering vast capacity at a lower cost but with lower bandwidth and higher latency.
This leaves a significant architectural gap between the two.
The future AI memory stack is more likely to be a tiered system rather than a choice between HBM and NAND: HBM handles the most bandwidth-intensive workloads, DRAM manages working memory requirements, and high-performance NAND begins to take on larger data pools that still need to be located close to compute.
Sandisk is Betting on NAND Becoming AI's Capacity Layer
Sandisk is currently one of the most vocal proponents of this thesis. At its 2026 Investor Day, the company projected the enterprise data center flash memory market could reach 1.2 ZB by 2030, while also expecting revenue to grow at a mid-to-high teens annual rate between FY2028 and FY2030. These projections are referenced in Reuters.
This outlook builds on Sandisk's already rapidly growing data center business. The company is increasingly aligning its long-term strategy with AI infrastructure, rather than relying primarily on consumer electronics or the traditional enterprise storage market.
This shift is significant because AI systems require not just "faster storage," but more storage capacity per system configuration. Larger models, RAG, multimodal content, and continuously operating AI agents all contribute to the growing volume of data that must remain accessible.
If AI inference becomes one of the dominant workloads in future data centers, the demand drivers for NAND could shift from PC, smartphone, and traditional cloud storage cycles toward the data generated and used by AI itself.
This represents a fundamental change from the NAND demand structure the market understood a decade ago.
HBF Could Fill the Gap Between NAND and HBM
High Bandwidth Flash (HBF) is currently one of the most intriguing memory tier solutions.
Sandisk positions HBF as a NAND architecture designed to complement, not replace, HBM. The company states that in certain AI inference configurations, HBF aims to deliver bandwidth close to HBM while offering up to roughly eight times the capacity at a similar cost. Sandisk is also working with SK hynix to drive standardization, hoping future AI inference devices can adopt this technology. More information is available in Sandisk's HBF technical explanation.
The appeal of this technology is straightforward. AI inference often requires keeping very large models and datasets close enough to accelerators; otherwise, moving data from traditional storage tiers could become a performance bottleneck. Expanding HBM indefinitely could improve bandwidth, but cost and capacity would become enormous challenges.
HBF attempts to find a middle ground: significantly higher capacity than HBM while delivering far greater performance than traditional flash memory.
Currently, HBF remains an emerging architecture and should not be viewed as a mature, large-scale revenue market. However, it illustrates a more important industry trend: storage vendors are attempting to move NAND beyond just backend storage and bring it closer to AI compute itself.
QLC NAND May Have a More Immediate Impact Than HBF
HBF tends to attract more attention because it's a new technology, but continued innovation in traditional NAND could have a larger impact in the near term.
QLC NAND stores four bits per cell, allowing for higher capacity within the same physical area compared to TLC NAND. This makes QLC particularly attractive for hyperscale data centers that require massive storage and are highly sensitive to cost per bit.
The trade-off is that storing more bits per cell can impact endurance and performance, making QLC unsuitable for all workloads. However, as controllers, error correction, and NAND architectures continue to improve, the range of applications suitable for QLC is expanding.
AI could accelerate this transition, as inference workloads particularly value capacity, energy efficiency, and overall cost.
Sandisk is continuously developing higher-density QLC products, while Samsung and Kioxia are also boosting density and performance through wafer bonding and next-generation NAND designs. For example, Samsung's latest BV-NAND architecture can increase storage density by approximately 58% compared to the previous generation, addressing growing AI storage demand.
Therefore, growth in the AI storage market doesn't entirely depend on HBF's success. High-capacity enterprise SSDs and more efficient QLC NAND can also directly benefit from continuously rising data density.
Why Has the NAND Cycle Historically Been So Volatile?
The bigger investment question is whether AI demand can truly alter NAND's historically volatile cycles.
Historically, the issue wasn't just insufficient demand; it was the interaction between demand and supply.
When NAND supply is tight, prices rise rapidly, and manufacturers enjoy high gross margins. Higher returns attract more capacity investment, but semiconductor fab planning and construction take time, so by the time new capacity actually comes online, the initial supply gap has often narrowed.
Prices then decline, gross margins compress quickly, and manufacturers reduce investment.
Then the cycle repeats.
AI won't automatically eliminate this mechanism. Even with very strong end-user demand, if manufacturers ultimately over-expand, supply can still outpace demand.
What could genuinely change is the degree of demand visibility suppliers have before making investment decisions.
Long-Term Agreements Could Change Supply Discipline
Sandisk is progressively moving customers toward multi-year agreements rather than relying primarily on spot transactions.
Reuters reported after Sandisk's latest earnings that the company has signed eight long-term agreements with six customers, totaling approximately $93.9 billion in contract value, with an average contract duration of about four years. Roughly half of FY2027 output is expected to be covered by these agreements, increasing to about two-thirds by FY2028. These figures are referenced in Reuters.
The importance of this model is that it improves visibility for both suppliers and customers.
Hyperscalers can lock in future storage capacity in advance, while Sandisk gains more committed demand visibility before deciding the scale of capacity expansion.
Sandisk is not alone in adopting this approach. Samsung has also indicated that long-term customer agreements could account for 60% to 70% of future memory sales, suggesting a broader business model shift may be occurring across the storage industry.
If suppliers increasingly base expansion decisions on multi-year committed demand rather than aggressively adding capacity in response to rising spot prices, the storage industry could potentially reduce the risk of the extreme oversupply seen in the past.
This won't eliminate the cycle, but it could change its severity.
What Does This Mean for Sandisk, Micron, and Other Memory Chip Stocks?
Growing AI storage demand won't benefit all storage companies in exactly the same way.
Sandisk is one of the most direct representatives of NAND and enterprise flash memory. The company's investment thesis now combines AI-driven bit demand, multi-year customer commitments, QLC, and HBF, making SNDK particularly sensitive to whether the market believes current NAND profitability can persist.
Micron's product portfolio spans DRAM, HBM, and NAND, exposing it to both ends of the AI memory hierarchy: high-priced, high-bandwidth memory close to accelerators, and higher-capacity flash memory. Strong AI demand has also improved supply visibility across multiple Micron products.
SK hynix's primary role in the AI investment theme is HBM, but its collaboration with Sandisk on HBF is noteworthy because it indicates that a leading HBM manufacturer also sees a potential place for a NAND-based high-capacity memory tier in future AI inference architectures.
Kioxia provides another direct window into NAND. As AI workloads drive demand for high-capacity storage, the company is accelerating production of next-generation flash memory, and the AI boom has significantly improved its operations and valuation.
Western Digital's core business has shifted toward hard disk drives rather than NAND following the Sandisk spin-off. But if the volume of AI-generated data continues to grow rapidly, it could support multiple storage tiers simultaneously. Flash is better suited for high-performance, low-latency workloads, while HDDs retain a cost advantage for massive cold data storage.
The broader conclusion is that AI storage is not a single-product trade.
Sustained growth in AI data could simultaneously drive demand for HBM, DRAM, NAND, enterprise SSDs, and high-capacity HDDs, with different technologies serving distinct positions within the storage hierarchy.
AI May Change the NAND Cycle, But Won't Make It Disappear
Describing the current storage rally as the industry permanently exiting its cyclical nature would be overly optimistic.
NAND remains a highly capital-intensive industry. Even without building entirely new fabs, technology upgrades can increase the number of bits produced per wafer; competitors may also add supply due to high gross margins, and AI capital expenditure itself cannot sustain the same growth rate indefinitely.
Recent stock price volatility reflects this risk. Both Sandisk and Western Digital experienced significant declines after reporting strong earnings in August, as market expectations were already so high that even optimistic outlooks failed to satisfy investors.
Therefore, a more reasonable structural view should converge on this:
AI can make long-term NAND bit demand stronger and more predictable, while multi-year customer agreements and stricter supply discipline could reduce the severity of future storage cycles.
This distinction is crucial.
Reducing NAND cycle volatility doesn't require storage prices to rise forever. What's truly needed is for suppliers to maintain reasonable profitability even when prices return to normal fluctuation, rather than repeatedly swinging from supply shortage to severe oversupply.
If this change does occur, the valuation frameworks for companies like Sandisk may ultimately shift as well.
