Storage Risks: Can "Long-term Protocols" Truly Be "Executed Long-term"?

A long-term contract is not only the commercial backbone of the AI boom, but may also be the hidden risk for the next downturn.

The AI boom has spawned a wave of large-scale long-term supply agreements. Chip manufacturers, cloud computing companies, and AI developers alike use these contracts to showcase “unprecedented revenue visibility” to investors. However, according to recent analysis from The Wall Street Journal, while these contracts may appear unbreakable during boom times, their actual binding effect in the face of reversing demand warrants careful scrutiny.

For example, the history of how similar contracts were widely waived when supply and demand reversed during the COVID-19 pandemic shows that these agreements are often much more fragile than they seem.

Memory Market: The “Most Extreme Example” of Long-term Contracts

The memory chip industry is the most typical example of this trend.

The explosive growth of autonomous AI agents has driven up the demand for memory, as such applications are highly dependent on memory resources. This has prompted the memory industry—a sector historically defined by intense price wars and strong cyclicality—to begin shifting towards a more stable model.

The three memory giants—Samsung Electronics, SK Hynix, and Micron Technology—are all currently reporting record profits, and anticipate that supply shortages will persist through 2028. SK Hynix just went public in New York this month, and a company executive stated in an April analyst call that long-term contracts are helping improve the market’s perception of the entire memory industry.

Micron has been especially proactive. Its “strategic customer agreements” are usually five years long and include take-or-pay clauses—meaning the buyer must pay whether or not they actually take delivery. Last month, Micron CEO Sanjay Mehrotra said in the company’s earnings call that these agreements will contribute over half the firm’s revenues going forward.

The capital market has responded directly in share prices: since the start of the year, Micron’s stock price has tripled, SK Hynix’s has risen by a similar margin, and Samsung’s has roughly doubled.

The “Weak Point” of Contracts: Who Enforces Them When Demand Drops?

The problem is: While long-term contracts fuel prosperity in upcycles, can they truly act as constraints in downcycles?

The Wall Street Journal’s analysis suggests the answer is likely no, for a simple reason:

First, if demand declines before the contract expires, chip manufacturers are reluctant to force shipments on customers—because chips that customers cannot use end up piling up in warehouses, and when demand returns, customers will clear out existing stock before buying more, thereby delaying the revenues of chip manufacturers.

Second, forcing shipments could damage long-term customer relationships, especially if competitors offer more flexibility. In that case, companies that insist on enforcing the contract would be at a disadvantage.

History has examples. The chip shortage during the COVID pandemic also led to a wave of long-term contracts, but once the shortage turned to oversupply, many of those agreements were renegotiated or postponed, with customers receiving large waivers.

Microchip Technology, a microcontroller chip producer, launched a “preferred supplier program” in 2021 that required clients to sign long-term commitments. But as the supply-demand balance reversed a few years later, the program was simply suspended. CEO Steve Sanghi plainly stated in November last year: “We’re not going to force customers to buy anything they don’t need.”

This statement is almost a perfect snapshot of the industry's reality in a downward cycle.

The Risk Has Spread Across the Entire AI Supply Chain

This risk isn’t confined to the memory market; it runs through the whole AI supply chain.

The chain looks roughly as follows: AI developers (like OpenAI) sign compute contracts with cloud providers (like Oracle, CoreWeave); cloud companies then sign procurement deals with AI chip manufacturers; chipmakers contract TSMC for fabrication; and TSMC then signs long-term equipment procurement contracts with Dutch company ASML.

Every link relies on the demand materializing at the next link.

The amounts involved are enormous. Oracle signed a massive cloud contract with OpenAI last year. By the end of last quarter, its “remaining performance obligations” (i.e., undelivered contract volume) reached $638 billion. Oracle CFO Hilary Maxson told analysts last month that this figure “gives us excellent visibility into future revenue growth, with all supported by long-term contractual commitments from customers.”

Data shows contract dependence has surged over the past year. Since mid-2025, the four largest AI spenders—Google, Microsoft, Amazon, and Oracle—have seen their collective revenue backlog swell by more than $1 trillion, more than doubling in total volume.

The Bank for International Settlements Warns

Such “visibility” can quickly become blurred.

The Bank for International Settlements (BIS) pointed out in its annual economic report this month that supply chain shortages in various AI segments may be amplifying overinvestment—“because companies are trying to lock in future capacity through long-term contracts, but those contracts in turn make them more vulnerable to shocks if demand falls short of expectations.”

In other words, lenders and investors who provide capital to companies based on these long-term contracts may face unexpected losses once demand cools.

The bigger the contract and the longer the chain, the more severe the knock-on effect when any link breaks down.