Bernstein has brought the memory industry's long-term procurement agreements back into the spotlight in its latest report: Micron and SanDisk have signed a batch of new LTAs with purchase commitments, minimum prices, and financial guarantees, attempting to provide a floor to earnings over the next few years.
However, this floor doesn't seem as solid as it looks.
According to Micron and SanDisk's public filings and earnings call guidance, Micron has signed 16 strategic customer agreements, with 14 totaling approximately $100 billion in minimum revenue calculated at the minimum contract price, along with cash deposits and financial commitments of about $22 billion. SanDisk's three quarter contracts correspond to about $42 billion in minimum contract revenue, with five agreements totaling financial guarantees exceeding $11 billion.
The combined $33 billion guarantee from both companies does make customer cancellations more expensive. However, Bernstein's model estimates the scale of revenue that may need LTA protection in the next 3-5 years to be around $5.2 trillion. Based on this report's calculation, the existing guarantees are only equivalent to about 0.6%.
This is precisely the divergence the report aims to convey: LTAs are changing the negotiating position of memory companies and large customers, but it's more like adding a cushion to the downside cycle rather than transforming DRAM and NAND into a utility.
Large Customers Locked into Long-Term Agreements, Guarantees Starting to Show Real Money
LTAs aren't complicated. Customers commit to future purchase volumes in advance, and suppliers provide supply assurances and pricing mechanisms. If customers don't buy, they may lose the prepayment guarantee or incur other financial costs.
This time is different from the common purchasing intentions in the memory industry in the past, with a focus on financial guarantees entering the contract structure.
As of June 2026, Micron has signed 16 strategic customer agreements, including four super-large customers and three medium-sized customers. The cumulative minimum revenue of 14 agreements calculated at the minimum contract price is approximately $100 billion, with an expected cash deposit and related financial commitments of about $22 billion. This metric includes signed agreements and contracts signed after the quarter end, not entirely equivalent to end-of-period RPO on the balance sheet.
SanDisk disclosed that as of April 3, 2026, its RPO was $41.6 billion. The company's earnings call also mentioned that three contracts for the quarter provided approximately $42 billion in minimum contract revenue, five agreements totaled over $11 billion in financial guarantees, and covered over one-third of the FY27 bit supply.
The two companies have different mechanisms. Micron's guarantees are more backend-weighted. As contracts progress and customer remaining purchase obligations decrease, the proportion of guarantees to RPO will increase, with higher abandonment costs later on. SanDisk, on the other hand, has a more fixed amount guarantee, with the guaranteed amount expected to remain relatively stable throughout the contract term.
The bull case is most concerned with this point. The biggest issue in the memory industry in the past was that profits would collapse too quickly when prices fell. If major customers are willing to provide guarantees for long-term supply, suppliers can at least have a clearer revenue floor, and capital expenditures and capacity planning do not have to be entirely dictated by spot prices.
A $33 Billion Guarantee Is Not Thin, But Can't Fully Mitigate a Sharp Downturn
The scale of guarantees and the scale of revenue to be protected are not in the same magnitude.
Bernstein calculated using a model that if LTAs are to cover potential revenue for the next 3-5 years, the corresponding protection scale would be around $5.2 trillion. This figure is based on the model's assumptions and is not directly disclosed in public company filings in a similar industry-wide revenue context, and it also needs to distinguish between memory, total semiconductor revenue, and supplier sample revenue.
Even so, a 0.6% guarantee ratio still indicates one thing: LTAs cannot fully protect profits in all price scenarios.
If spot prices only moderately decline, it is not cost-effective for customers to abandon contracts. The costs of losing guarantees, damaging supply relationships, and potentially not securing scarce capacity in the future are enough to keep customers fulfilling their obligations. Demand for stable supply from AI servers, cloud providers, and data center customers is also stronger than that from ordinary consumer electronics customers.
However, when prices fall deep enough, customers will still do the math. As long as the remaining purchase volume is significant and the spot price is low enough relative to the contract floor price, customers may find it cheaper to buy from the market even if they lose the guarantee.
The backend-weighted mechanism can alleviate this issue. As the contract progresses, the remaining RPO decreases, the proportion of guarantees to remaining obligations increases, and the cost to customers of abandoning the contract is higher. The protection strength may be stronger in the later stages of the contract, and the memory cycle often requires more protection in the later stages.
It is still not unconditional insurance. LTA protection depends on three numbers: where the spot price drops to, how much of the purchase obligation the customer still has, and how much collateral balance is left.
This is also at the heart of the bull-bear dichotomy. Bulls see that the memory company has finally secured a long-term commitment of real money from customers. Bears are concerned that these commitments may still not be sufficient to protect peak profits, and once the downturn is deep enough, customers will still act based on cost.
Not All Memory Demand Is Willing to Be Locked by LTA
LTA also has a practical limit: not all customers are suitable for signing long-term agreements.
American cloud providers are the most ideal targets. They have large demand, strong credit, are sensitive to the stability of AI infrastructure supply, and are also more motivated to lock in supply through long-term agreements. Micron has basically completed negotiations with American CSPs, and is still advancing with Chinese CSPs, enterprises, and some other customers.
Consumer businesses are different. The SanDisk CFO has stated that the consumer business is "more transactional," and LTA "does not apply." Mobile phones, PCs, and consumer storage channels are more accustomed to purchasing based on price and inventory cycles. When prices fall, customers naturally want to maintain flexibility rather than be locked in by a multi-year floor price.
Chinese customers may not necessarily become stable buyers under LTA. On one hand, Chinese cloud providers and end customers may be more inclined towards local suppliers. On the other hand, the expansion of local DRAM and NAND supply will also add uncertainty to long-term procurement commitments.
Bernstein estimates that 30%-50% of the overall DRAM and NAND end markets may be difficult to cover with LTAs. Even if leading suppliers lock in major U.S. customers, a considerable portion of the market will continue to operate based on spot prices, short-term orders, and cycle expectations.
As long as a sufficiently large proportion of demand remains in the spot or short-contract system, price signals will not disappear. As long as price signals exist, supplier expansions, customer destocking, and channel cancellations will continue to amplify cycle fluctuations.
Elevated Valuation Supported by AI Demand, but Peak Earnings Cannot Be Extrapolated Directly
The market is willing to give memory companies a higher valuation, partly because AI demand has changed the foundation of this cycle.
On the DRAM front, HBM demand remains strong. Bernstein's Asia team predicts that by 2027, HBM prices may increase 2-2.5 times compared to 2026, with regular DRAM commercial prices having already risen significantly, likely to remain elevated in the next 12 months. While HBM is more stable than regular memory, it shares some capacity with regular DRAM, and capacity allocation will affect other product lines.
On the NAND front, AI inference and longer context windows have also brought about new demand scenarios. Early AI training mainly consumed HBM and DRAM, but as inference, Agentic AI, and long-context applications increase, storage demand may continue to rise. It is worth noting that the description of Vera Rubin-related capacity should not be simplistically referred to as "GPU NAND capacity"; NVIDIA's official page discloses a 20.7TB HBM4 GPU memory.
In this environment, Long-Term Agreements (LTA) value is more like locking in a portion of the high-cycle revenue. With ongoing strong AI demand, suppliers can lock in some large customer purchases through long-term agreements. If prices fall, guarantees and floor prices can delay profit declines.
SanDisk stress tests also point to a similar conclusion. The Bernstein model shows that under stricter assumptions, LTAs can still keep FY29-FY30 EPS higher than scenarios without LTAs in most penetration rate scenarios, especially with stronger protection later on. However, the same set of stress tests also indicates that peak earnings cannot be easily extrapolated. In scenarios with lower operating profit margins, EPS may be significantly lower than the current run rate.
The most noteworthy conclusion from this report is not the "end of the memory cycle" but the "softening of the downward cycle."
Micron and SanDisk have secured long-term agreements and financial guarantees, indicating that large customers are willing to pay for supply certainty in the AI era. For memory companies, this will increase revenue visibility in the coming years and make the capital markets more willing to believe that the profit bottom line is higher than in the past.
The constraint is equally clear. A $330 billion guarantee can only provide partial cushioning, ensuring that consumers, Chinese clients, and some transactional needs will not all enter into a long-term agreement. Bernstein also estimates that China's DRAM market share could increase from about 8% to 16% in the coming years, with NAND possibly facing stronger supply pressure after 2028.
What the LTA truly needs to prove is not whether it can sign during an upturn, but whether customers will fulfill their commitments, guarantees will be painful enough, and suppliers will continue to maintain capacity discipline in the next downturn cycle. Until these questions are answered, it serves as a new buffer for the memory industry, not as a cycle-ending button.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
