TL;DR
Bernstein focuses on Micron and SanDisk's new LTA, noting that long-term purchase agreements are improving memory revenue visibility.
The two companies disclosed a combined RPO of approximately $142 billion and financial guarantees of about $33 billion, both significantly below the model protection threshold.
· LTA increases the cost of contract abandonment for large clients, while retail consumers, Chinese customers, and spot demand will still exhibit cyclical fluctuations.

Bernstein has brought long-term procurement agreements in the memory industry back into focus in its latest report: Micron and SanDisk have signed new LTAs with purchase commitments, minimum pricing, and financial guarantees, aiming to establish a floor for profitability over the coming years.

This floor doesn't appear as thick as it looks.

According to public filings and conference calls from Micron and SanDisk, Micron has signed 16 strategic customer agreements, with a combined minimum revenue of approximately $100 billion based on minimum contract prices, and related cash deposits and financial commitments totaling approximately $22 billion. SanDisk’s three current-quarter contracts correspond to approximately $42 billion in minimum contract revenue, and five agreements combined provide financial guarantees exceeding $11 billion.

The combined $33 billion in collateral from the two companies has indeed made it more expensive for large clients to default. However, the Bernstein model estimates that the revenue requiring LTA protection over the next 3–5 years could reach approximately $5.2 trillion. According to this reporting口径, existing collateral represents only about 0.6%.

This is precisely the divergence the report highlights: LTA is shifting the negotiating position of memory companies and large customers, but it’s more like providing a buffer during a downturn than transforming DRAM and NAND into utilities.

Large clients are locked into long-term agreements, and guarantees are beginning to turn into real money.

LTA is not complicated. The customer commits in advance to purchasing volumes over the coming years, and the supplier provides supply assurance and a pricing mechanism. If the customer does not purchase, they may lose their prepaid guarantee or incur other financial costs.

This time, unlike typical purchase intentions in the memory industry, the focus is on the inclusion of financial guarantees in the contract structure.

Micron has signed 16 strategic customer agreements as of June 2026, including four hyperscale customers and three mid-sized customers. The aggregate minimum revenue under these 14 agreements, calculated at minimum contract prices, is approximately $100 billion, with expected cash deposits and related financial commitments totaling approximately $22 billion. This figure includes both signed agreements and those signed after the quarter-end and does not fully equate to the RPO as of the balance sheet date.

SanDisk disclosed that, as of April 3, 2026, the RPO amounted to $41.6 billion. The company’s earnings call also noted that three contracts for the quarter provide approximately $42 billion in minimum contract revenue, while five agreements collectively offer over $11 billion in financial guarantees and cover more than one-third of FY27 bit supply.

The two companies have different mechanisms. Micron's guarantee places greater emphasis on backend weighting; as the contract progresses and the customer's remaining purchase obligations decrease, the ratio of the guarantee to RPO increases, resulting in higher penalties for abandonment in later stages. SanDisk's guarantee is closer to a fixed-amount guarantee, with the guaranteed amount expected to remain relatively stable throughout the contract period.

Micron: 16 agreements, RPO of approximately $100 billion, guarantees of approximately $22 billion; SanDisk: 5 agreements, RPO of approximately $42 billion, guarantees exceeding $11 billion.

The bulls value this most. The memory industry's biggest historical issue has been that profits collapse too quickly when prices fall. If major customers are willing to pay guarantees for long-term supply, suppliers can at least secure a clearer revenue floor, reducing the need to base capital expenditures and capacity planning entirely on spot prices.

$33 billion in collateral is substantial, but it's not enough to withstand a deep downturn.

The scale of collateral and the scale of income to be protected are not on the same level.

Bernstein's model estimates that to cover potential revenues over the next 3 to 5 years, the LTA would require a protection scale of approximately $5.2 trillion. This figure is based on the model used in the report; public company filings do not directly disclose similar industry-wide revenue figures, and it is necessary to distinguish between memory, total semiconductor revenue, and supplier sample revenue.

Even so, a collateral ratio of 0.6% indicates that LTA cannot cover profits under all price scenarios.

If the spot price only declines modestly, it is not economical for customers to default. The costs of losing collateral, damaging supply relationships, and potentially being unable to secure scarce capacity in the future are sufficient to encourage customers to continue fulfilling their contracts. AI server, cloud provider, and data center customers also have stronger demands for stable supply than typical consumer electronics customers.

But when prices fall deep enough, customers still do the math. As long as the remaining purchase volume is substantial and the spot price is low enough relative to the contract floor price, customers may find it cheaper to buy on the market—even at the cost of forfeiting their guarantee.

The backend weighting mechanism can alleviate this issue. As the contract progresses, the remaining RPO decreases, causing the ratio of collateral to outstanding obligations to rise, thereby increasing the cost for clients to abandon the contract. Protection strength may be stronger in the later stages of the contract, and memory cycles often require greater protection during these later stages as well.

It is still not unconditional insurance. The level of protection provided by LTA depends on three factors: how far the spot price has fallen, how much of the purchase obligation remains, and how much collateral balance remains.

RPO decreases over time, while the collateral/RPO ratio increases; if the spot ASP falls significantly below the contract floor price, customers may still choose to default.

This is at the heart of the bull-bear divergence. Bulls see that memory companies have finally secured long-term, real-money commitments from customers. Bears worry that these commitments are still insufficient to protect peak profitability, and if the downturn is deep enough, customers will still act based on cost.

Not all memory demands are willing to be locked by LTA.

LTA also has a practical limit: not all customers are suitable for signing long-term agreements.

U.S. cloud service providers are the ideal targets. They have high demand, strong creditworthiness, are sensitive to the stability of AI infrastructure supply, and are more motivated to lock in supply through long-term agreements. Micron has largely completed negotiations with U.S. CSPs and is continuing to engage with Chinese CSPs, enterprises, and select other customers.

The consumer business is different. SanDisk’s CFO once stated that the consumer business is “more transactional,” and LTAs “don’t apply.” Mobile phones, PCs, and consumer storage channels are accustomed to purchasing based on price and inventory cycles; once prices fall, customers naturally prefer to maintain flexibility rather than be locked into multi-year floor prices.

Chinese customers may not necessarily become stable buyers of LTA. On one hand, Chinese cloud providers and end customers may prefer local suppliers. On the other hand, the expansion of domestic DRAM and NAND supply could introduce uncertainty into long-term procurement commitments.

Bernstein estimates that 30% to 50% of the overall DRAM and NAND end markets may remain uncovered by LTAs. Even if leading suppliers secure major U.S. customers, a significant portion of the market will continue to operate based on spot prices, short-term orders, and cyclical expectations.

The end-market breakdown for DRAM/NAND shows demand from Chinese CSPs, enterprise servers, consumer/PC, smartphones, and others, beyond U.S. CSPs, with approximately 30%-50% of the market potentially uncovered by LTAs.

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, suppliers increasing production, customers reducing inventory, and distributors canceling orders will still amplify cyclical fluctuations.

AI demand supports valuation, but peak profitability cannot be directly extrapolated.

The market is willing to assign higher valuations to memory companies, as AI demand has altered the bottom shape of this cycle.

On the DRAM side, demand for HBM remains strong. Bernstein’s Asia team forecasts that HBM prices could rise 2 to 2.5 times in 2027 compared to 2026. Conventional DRAM commercial prices have already surged significantly and are likely to remain high over the next 12 months. Although HBM is more stable than standard memory, it shares part of the production capacity with conventional DRAM, and capacity allocation can impact other product lines.

On the NAND side, AI inference and longer context windows also create new demand possibilities. Early AI training primarily consumed HBM and DRAM, but as inference, agentic AI, and long-context applications increase, storage demands may continue to rise. It is important to note that references to Vera Rubin-related capacity should not be simplistically stated as “NAND capacity on GPUs”; NVIDIA’s official page discloses 20.7 TB of HBM4 GPU memory.

In this environment, LTA’s value lies in locking in a portion of high-demand revenue. With AI demand remaining strong, suppliers can secure purchases from major clients through long-term agreements. If prices decline, guarantees and floor prices can help slow the decline in profitability.

SanDisk stress tests also point to similar conclusions. The Bernstein model shows that, under more stringent assumptions, LTA still enables FY29–FY30 EPS to exceed the no-LTA scenario in most penetration scenarios, with particularly stronger protection in later periods. However, the same set of stress tests also indicates that peak profitability cannot be simply extrapolated; under lower operating margin scenarios, EPS could fall significantly below current run rates.

The SanDisk FY29-FY30 EPS sensitivity table shows a wide range of EPS across different ASPs and LTA penetration rates; even under improved LTA scenarios, peak profitability cannot be guaranteed.

The most valuable insight in this report is not that "the memory cycle has ended," but that "the downward cycle may be softened."

Micron and SanDisk securing long-term agreements and financial guarantees demonstrate that major customers are willing to pay for supply certainty in the AI era. For memory companies, this enhances revenue visibility over the coming years and makes it easier for capital markets to believe that profit margins will be higher than in the past.

The limitations are equally clear. A $33 billion guarantee can only provide partial cushioning, and consumers, Chinese customers, and some speculative demand will not all enter long-term agreements. Bernstein also estimates that China’s DRAM market share could rise from around 8% to 16% over the next few years, and NAND may face stronger supply pressures after 2028.

What LTA truly aims to prove is not whether contracts can be signed during an upcycle, but whether customers will honor their obligations, whether guarantees will be sufficiently stringent, and whether suppliers will maintain production discipline during the next downturn. Until these issues are resolved, LTA is a new buffer for the memory industry—not a button to end the cycle.