Long-term supply agreements signed by Micron and SanDisk are giving the memory-chip industry better revenue visibility, but they are not strong enough to end the sector’s boom-and-bust cycle, according to Bernstein’s latest assessment of the market.
The firm estimates that Micron and SanDisk have a combined $142 billion in remaining performance obligations tied to long-term agreements, or LTAs. Those contracts represent a meaningful shift in how major memory suppliers are trying to secure demand before the next downturn. Yet the financial guarantees attached to those deals total about $33 billion, covering only a small fraction of the much larger revenue base that could be exposed to falling prices over the next three to five years.
Bernstein estimates that the broader pool of memory revenue potentially needing protection could reach about $5.2 trillion over that period. On that basis, the combined guarantees from Micron and SanDisk would cover roughly 0.6% of the total exposure. That means the agreements can soften the blow from weaker market conditions, but they are unlikely to fully protect chipmakers if memory prices fall sharply.
The findings underline a central tension in the memory business. The industry is enjoying powerful demand from artificial intelligence infrastructure, cloud computing and data center expansion. At the same time, DRAM and NAND remain cyclical products, with prices still heavily influenced by supply additions, inventory swings and spot-market buying behavior.
For Micron and SanDisk, the new contracts represent progress. They create a firmer base for production planning, capital spending and customer relationships. But the scale of remaining uncovered demand shows that the traditional memory cycle has not disappeared.
Micron and SanDisk lock in major contracts
Micron has signed 16 strategic customer agreements that together represent about $100 billion in minimum contract revenue. Those agreements are backed by roughly $22 billion in deposits and financial guarantees.
SanDisk has five major contracts that secure about $42 billion in minimum contract revenue. Those deals are supported by more than $11 billion in financial guarantees, with coverage extending into fiscal 2027.
Together, the figures show that large memory manufacturers are moving more aggressively to lock in demand from key customers. These agreements are designed to reduce the uncertainty that has historically made the memory business difficult to forecast.
In past cycles, buyers often reduced purchases when demand slowed or when prices dropped quickly. That behavior left suppliers with excess inventory, weaker pricing power and sudden earnings pressure. LTAs are meant to reduce that risk by creating penalties for customers that walk away from committed purchases.
Still, Bernstein’s view is that these contracts should be seen as partial protection rather than a full hedge. The agreements raise the cost of cancellation, but they do not remove the economic incentive for customers to switch to cheaper spot purchases if market prices fall far enough.
Different contract structures aim for the same goal
Micron and SanDisk use different LTA structures, though both are intended to make contract termination costly during weak pricing periods.
Micron’s model places more weight on guarantees in the later years of the agreements. As purchase obligations decline over time, the guarantee-to-obligation ratio rises. That structure can increase the cost of cancellation in later contract stages, when customers might otherwise be more tempted to renegotiate or exit.
SanDisk uses fixed-amount guarantees that remain steadier throughout the contract term. That approach provides a more consistent financial penalty if a customer chooses not to honor its obligation.
The practical effect is similar. Both models are designed to discourage customers from abandoning committed supply during periods of weak pricing. The agreements make it more expensive to default, while also giving customers more secure access to supply during periods of tight availability.
This is especially important in memory markets, where buyers often face a trade-off between price flexibility and supply certainty. When supply is abundant, short-term purchases can look attractive. When supply is tight, customers with long-term agreements may have better access to critical components.
Protection weakens in a severe downturn
Bernstein’s model suggests that LTAs are likely to work best during moderate price declines. In that environment, customers may continue to honor agreements because walking away would mean losing deposits, guarantees or access to reliable long-term supply.
The calculus changes if prices fall sharply. If spot-market memory becomes cheap enough, some customers may decide that forfeiting guarantees is still less costly than continuing to buy under older contract terms. That is the key limitation of the current LTA structure.
The effectiveness of these agreements depends mainly on three variables: how far market prices decline, how much of the purchase obligation remains, and how large the residual guarantee is at the time of potential default.
Contracts tend to offer stronger downside protection in later stages when the guarantee-to-obligation ratio is higher. In simpler terms, if a customer has less remaining product to buy but still faces a meaningful penalty for walking away, the contract becomes harder to abandon.
However, during the early stages of a contract, large outstanding purchase obligations can still create pressure if market prices move sharply lower. A customer may view the remaining commitment as too expensive relative to the open market, even after factoring in penalties.
That is why Bernstein describes the current protections as buffers, not cures. They can help stabilize revenue, but they cannot completely override the economics of a deep industry downturn.
Cloud customers offer the strongest demand base
The structure of memory demand also limits how far LTAs can reshape the industry.
U.S. cloud service providers are considered the most dependable base for these agreements. Their demand is large, recurring and closely tied to long-term infrastructure plans. These companies need stable access to memory components for data centers, AI servers and enterprise computing platforms.
For large cloud customers, supply reliability can matter as much as price. A shortage of high-performance memory can delay server deployments, reduce data center efficiency and disrupt customer-facing services. That makes long-term agreements more attractive.
Other parts of the market are less suited to LTAs. Consumer electronics companies, Chinese customers and smaller enterprise buyers often prefer flexible purchasing arrangements tied more closely to current market pricing. These buyers are more likely to use spot purchases or short-term contracts, especially when demand visibility is limited.
Bernstein estimates that these more flexible customer groups represent roughly 30% to 50% of the DRAM and NAND market. As long as such a large share of demand remains outside long-term commitments, spot prices will continue to play a major role in production decisions, purchasing behavior and inventory adjustments.
That means the memory sector is likely to remain cyclical even if LTAs become more common.
AI demand is changing the baseline for memory
Artificial intelligence continues to strengthen underlying demand for memory, particularly in high-bandwidth memory, known as HBM, and in NAND storage used for inference workloads.
HBM has become a critical component in AI accelerators because it provides the bandwidth needed to move large volumes of data quickly between processors and memory. Demand has risen as data center operators expand AI training and inference capacity.
Bernstein projects that HBM pricing in 2027 could be two to two-and-a-half times higher than in 2026. That would reflect tight supply, strong customer demand and the technical complexity of producing advanced memory stacks.
However, higher HBM pricing does not automatically translate into profit growth across every memory segment. HBM production shares capacity with standard DRAM, meaning manufacturers must allocate equipment, wafers and production resources between product categories. A stronger HBM market may support margins, but it can also create trade-offs in standard DRAM output.
NAND demand is also expected to benefit from AI growth. As AI systems evolve toward agent-based workloads, long-context processing and heavier retrieval requirements, storage needs could increase. More advanced AI applications often require larger data pools, faster access to stored information and more persistent memory infrastructure.
Bernstein cautions that storage demand should not be directly compared with GPU memory specifications. For example, new system architectures may require more NAND capacity, but that does not mean storage needs can be measured in the same way as publicly listed GPU memory configurations, such as Nvidia’s 20.7 TB HBM4 system figures.
The distinction matters because AI demand affects different parts of the memory stack in different ways. HBM supports high-speed computation, while NAND supports data storage and retrieval. Both may benefit from AI growth, but their pricing and supply dynamics are not identical.
Data center spending adds pressure to supply chains
The broader infrastructure buildout is adding another layer of pressure to the memory market.
Data center capital spending is widely expected to remain elevated in 2026 as cloud platforms, AI developers and enterprise technology companies continue expanding computing capacity. Some market estimates cited in the source material place 2026 data center capital spending on track to approach $650 billion by the end of the year.
That spending wave has already tightened availability across several hardware categories. Memory components, networking systems, power equipment and AI server parts have all faced periods of constraint as large buyers compete for supply.
Raw memory costs have also risen sharply since early 2025, according to the source material, with some categories up more than 200%. Such increases have created challenges for smaller data center operators, independent compute providers and companies that lack long-term supply arrangements with major chipmakers.
Dell’s recent move to lift its AI server revenue forecast to $60 billion for the coming fiscal year illustrates the scale of corporate demand. Strong orders from large server makers can absorb significant component supply before smaller buyers reach the market.
This environment gives memory suppliers more leverage in contract negotiations. When customers are worried about shortages, they may be more willing to accept long-term commitments, deposits and guarantees. That helps explain why Micron and SanDisk have been able to secure large LTAs during the current cycle.
Smaller buyers remain exposed to spot prices
The benefits of long-term agreements are not evenly distributed across the market.
Large cloud platforms and major technology companies can negotiate directly with suppliers and reserve future capacity. Smaller data operators often buy through distributors, system builders or the spot market. That leaves them more exposed to price spikes and shortages.
Companies that operate computing farms for blockchain validation, high-performance computing or specialized data processing may face higher replacement costs if memory shortages intensify. These operators depend on reliable access to servers, graphics processors, memory boards and storage devices. When large corporate buyers absorb a greater share of available supply, smaller buyers can face longer lead times and higher prices.
The same dynamic can affect traders in digital assets linked to physical computing networks. Tokens connected to decentralized compute, storage or hardware-intensive infrastructure may become more sensitive to equipment availability and component costs. If hardware becomes more expensive, network expansion can slow and operating margins can tighten.
However, the relationship between chip supply and digital asset markets is indirect. Token prices are influenced by many factors beyond hardware costs, including liquidity, protocol design, trading sentiment, interest rates and broader risk appetite. The main takeaway from the memory market is that physical infrastructure remains a key constraint for compute-heavy business models.
LTAs may cover only part of future supply
The source material cites reporting that long-term contracts currently cover roughly one-third of expected supply output for the next fiscal cycle. If that estimate is accurate, a large share of factory capacity is already committed to major customers before open-market buyers place orders.
That could make the spot market more volatile. When a greater portion of supply is locked into long-term agreements, buyers without contracts compete for a smaller pool of available chips. If demand remains strong, those buyers may pay higher prices. If demand weakens, the spot market could still fall quickly because flexible buyers may delay purchases.
This split market is important. LTAs can make revenue more predictable for suppliers, but they may also concentrate supply access among the largest customers. Smaller companies may still face the full force of memory-price volatility.
For chipmakers, the benefit is clearer. Even if spot prices weaken, contracted customers may continue purchasing under agreed terms. That can support factory utilization and provide more confidence for capital spending decisions. But because guarantee coverage is limited relative to the total market, suppliers still face risk if a downturn spreads across customer segments.
SanDisk stress tests show partial earnings support
Bernstein’s stress tests for SanDisk suggest that its LTAs could help keep fiscal 2029 and 2030 earnings per share above non-LTA scenarios under most modeled conditions. The support appears strongest in later contract stages, when guarantees provide more protection relative to remaining obligations.
That finding supports the broader view that LTAs can improve downside resilience. They can reduce the severity of earnings declines compared with a fully unprotected market.
But the same stress tests also show that peak profitability remains vulnerable. If market prices or margins fall sharply from current levels, LTAs may not fully protect earnings. They can preserve some revenue, but they cannot force customers to pay above-market prices indefinitely if the economic gap becomes too large.
This is especially relevant in memory, where supply additions can quickly change pricing conditions. New capacity, weaker end demand or inventory corrections can all push prices lower. Once a downturn begins, customer behavior often shifts rapidly from securing supply to minimizing cost.
China’s production growth could add future pressure
Another risk is rising production from China.
Bernstein expects China’s share of DRAM production could roughly double in coming years, increasing from about 8% to around 16%. If that expansion continues, it could add more supply pressure to the global market.
The impact may be especially important after 2028, when additional capacity could affect NAND pricing and broader memory supply conditions. More production from Chinese manufacturers could intensify competition, particularly in mainstream memory products where price sensitivity is high.
For established suppliers, LTAs may help offset some of that pressure by tying major customers to committed volumes. But they cannot fully prevent pricing pressure if global supply grows faster than demand.
The ultimate test will come in the next downturn. That is when the market will learn whether customers honor their obligations, whether financial guarantees are large enough to discourage contract exits, and whether suppliers maintain discipline on production.
For now, Micron and SanDisk have secured stronger commitments from key customers and improved visibility into future revenue. The agreements mark a meaningful change in the memory industry’s commercial structure. But with only $33 billion in guarantees set against a much larger exposure model, substantial risk remains outside LTA coverage.
The memory cycle has gained a new cushion. It has not found a cure.
To navigate similar market downturn risks, explore strategies for managing volatility across cyclical tech and crypto sectors.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
