The Rules of the Storage Game Have Fundamentally Changed

In The Godfather, Vito Corleone says: “I’m going to make him an offer he can’t refuse.”

That line is iconic, not just for its ruthlessness.

But because it captures the essence of a certain kind of transaction: truly high-stakes negotiation isn’t about today’s purchase or the price tag. It’s about locking down the other party’s options for years to come.

If you accept this offer today, you’re acknowledging that for the foreseeable future, you’ll be sitting at the same table, allocating resources, sharing risks, and binding interests.

Put that into the context of this Micron conference, and it takes on an industry-level meaning.

AI customers now face a problem that isn’t just about whether memory is expensive. It’s about whether they can secure stable supply for years to come. GPUs can be bought, servers can be assembled, data centers can be expanded, and power and liquid cooling can be gradually configured — but if critical memory like HBM, server DRAM, and data center SSDs can’t keep up, the entire AI capital expenditure chain will stall. Without memory, compute is just idling.

So what Micron presented this time wasn’t just a great earnings report. It was a new table: customers sign multi-year agreements, lock in purchase volumes, accept price ranges, and put up cash deposits in exchange for stable supply rights over the next several years.

This Earnings Report: The Numbers Are Just the Surface

Micron’s latest quarterly revenue was $41.5 billion, up 74% sequentially and 346% year-over-year. Gross margin reached 84.9%. Guidance for the next quarter is $50 billion in revenue, with gross margins around 86%.

In any previous memory cycle, this would look like peak-cycle performance.

After the earnings release, the stock surged 15% in after-hours trading.

Someone made a great analogy: a top student scores 100 because 100 is the maximum; Micron’s 85% gross margin is because capacity is limited.

The data center segment is particularly staggering: annualized revenue surpassed $25 billion, over $100 billion for the full year; data center SSD quarterly revenue exceeded $5 billion, doubling sequentially. Global data center memory bit shipments in 2026 are forecast to double compared to two years ago — and this is just the beginning of AI-driven demand for memory.

But from my perspective, what truly changes the pricing logic of capital markets is not how much money was made this quarter. It’s the disclosure of 16 Strategic Customer Agreements (SCAs).

Behind the $100 Billion+ Long-Term Contracts: A New Power Structure

These agreements cover the period from 2026 to the end of 2030, involving 4 hyperscalers, 3 mid-sized customers, and some automotive customers. Standard tenor is 5 years, automotive is 3 years. Of these, 14 agreements are based on minimum price and minimum purchase volume, with total remaining performance obligations (RPO) around $100 billion. Customers provided approximately $22 billion in cash deposits and financial commitments, of which about $18 billion are cash deposits.

These lock in about 20% of DRAM sales and about one-third of NAND sales.

Management stated that if all target agreements are completed, about half or more of the company’s revenue will come from SCA customers. Approximately 40% of revenue is protected within price bands, with the floor price supporting gross margins still above historical peaks. One-third of capacity over the next five years is already pre-booked, with the long-term contract floor price higher than the highest historical price.

Translated into capital markets language:

Micron is attempting to transform a traditional, strongly cyclical business into an AI infrastructure asset with order visibility, cash flow guarantees, and supply right premiums.

Why the Market Has Historically Avoided Giving Storage High Valuations

This question isn’t complex.

Look at DRAM prices, NAND prices, inventory cycles, and which of Samsung, SK Hynix, or Micron expands capacity first or cuts capex first.

When the cycle is up, stock prices recover; when prices fall, profits collapse, and valuations revert to cyclical stocks. Storage companies have been undervalued for a long time — not because the market doesn’t understand the technology, but because the market doesn’t believe these profits are sustainable.

What Micron is trying to change this time is that old problem.

The significance of 16 long-term contracts isn’t that they eliminate the cycle — that judgment is too optimistic and easily disproven.

Their significance lies in compressing part of the cyclical volatility into contracts. Customers commit to buying a certain volume over the next few years; Micron commits to supplying a certain volume. Some large agreements have both price caps and price floors. Management repeatedly emphasizes a key figure: even at the floor price, Micron’s gross margins will be far above any historical cyclical peak.

This statement carries a lot of weight. In the past, when memory prices rose, investors’ first reaction was:

How long will this last?

Is another capacity expansion imminent?

Will margins start deteriorating next quarter?

But now, if the floor price already supports gross margins far above historical peaks, the market must reconsider not “whether the peak will pass,” but “whether the profit floor has been permanently raised.”

Long-term contracts don’t mean Micron will stop rising or falling. They allow the market to reassess the profit baseline.

What Does $18 Billion in Cash Deposits Signify?

In the past, the storage industry operated more like a spot market. Customers squeezed prices when they were low; suppliers captured elasticity when prices were high. Everyone knew the cycle would come, and everyone knew it would go.

The AI era is different. For large model companies, cloud providers, and AI server customers, failing to secure memory isn’t just a cost issue — it’s a strategic pacing issue. Models need longer context, inference needs lower latency, agents need larger-scale invocation, and data centers need to shift from training to continuous inference. Memory has evolved from “a cost item in a server” to the fundamental constraint on whether an AI system can run at all.

Micron stated plainly on the call: AI system performance, at the architecture level, depends on the performance and capacity of the memory subsystem. You can use GPUs, ASICs, or CPUs, but ultimately you can’t bypass memory bandwidth, capacity, and low power.

That’s why customers are willing to put down deposits. $18 billion in cash deposits is not a traditional prepayment, nor does it directly count as free cash flow; it will be gradually returned to customers in the latter part of the agreement. But its signal to the market is extremely strong: customers aren’t just talking about shortages; they’re putting real money forward to secure supply.

Capital markets have a simple judgment criterion: demand expressed in words is worthless; demand in contracts is valuable. Long-term growth presented in PPTs is worthless; long-term growth that can collect deposits is valuable.

The Narrative Around Capacity Expansion Has Been Completely Reconstructed

In traditional cycles, whenever a memory manufacturer announced expansion, the market got nervous. We’ve seen it too many times: expand at the peak, then supply comes online as demand slows, prices collapse, inventory piles up, profits fall. In the past, larger capex sometimes looked like a top-cycle signal.

But this time is different. Micron can now expand capacity backed by SCAs, RPO, and customer deposits. The logic is no longer “I’m betting future demand will be good,” but “Customers have already told me via contracts and deposits that they need product for the next several years.”

Micron’s capex isn’t low going forward. The company expects Q4 capex around $10 billion, full-year capex about $27 billion, and quarterly capex in FY2027 will be higher than Q4 levels. The ID1 facility in Lehi is expected to produce first wafers in mid-2027, ID2 in late 2028; the New York wafer fab cluster has been initiated, and manufacturing and advanced packaging capabilities in Taiwan, Japan, and Singapore are advancing.

Without long-term contracts, these expansions would easily be interpreted as supply risk. With contracts, at least at this stage, expansion can be understood as building capacity to order. Capital expenditure also becomes more tolerable to institutional investors.

Supply Can’t Catch Up with Demand — This Isn’t a Short-Term Phenomenon

The rigid constraints on the supply side make the strategic value of long-term contracts even more prominent. Wafer fab construction cycles are long, specialized workers are scarce, EUV equipment is limited, HBM consumes general DRAM capacity, and some NAND capacity is being shifted to DRAM. Multiple constraints are stacking up. The memory supercycle has been confirmed — supply-demand tightness will last at least until 2027, with minor improvements in 2028, and no point in the near term where supply catches up to demand.

SK Hynix and Samsung are also ramping up capex. The Korean government is discussing large new semiconductor production facility investments with Samsung and SK Hynix, potentially accelerating some construction timelines by more than a decade. Japan is even more aggressive — it has announced a 14-year super investment plan through FY2040, totaling over ¥370 trillion (about $2.3 trillion), with AI and semiconductors as key focuses.

HBM4 12-layer yield ramp speed is twice that of HBM3E, and has already surpassed $1 billion in revenue. Technology iteration is accelerating, but the pace of capacity release is far behind the explosion in demand.

Micron itself says the tightness will persist beyond 2027; even if industry supply begins to gradually improve in 2028, it’s unclear when memory supply will truly catch up with demand. This is the core pricing basis for bulls. Conversely, if AI capex growth slows or customers’ outlook for future demand weakens, the market will immediately re-evaluate these expansion projects.

The Cycle Hasn’t Disappeared — It’s Just Taken a Different Form

Tolerance doesn’t equal blind faith.

Will Micron be free of cycles from now on? Impossible.

The storage industry cycle won’t disappear, but its form has become more complex. AI demand is strong, supply expansion is slow, HBM consumes a large amount of wafer capacity, and tightness may last longer. But as long as the industry keeps expanding capacity and capex keeps flowing, there will always be the risk of future supply-demand mismatches.

So the core judgment of this article is not “Micron will only go up from now on,” nor “the storage industry has completely left the cycle behind.”

My point is simpler: The significance of this Micron event is that it has taken a big step in moving the valuation anchor of the storage industry from spot prices toward long-term orders and supply rights.

Previously, Micron’s story was about DRAM price increases, NAND recovery, inventory digestion, and earnings recovery.

Now, Micron’s story is about $100 billion+ RPO, take-or-pay contracts, customer deposits, price floors, supply assurance, and AI strategic assets. This language is clearly more suitable for institutional investors and more likely to trigger asset revaluation.

Profits Will Continue to Migrate Downstream

Zooming out. Micron’s story is not isolated.

Nvidia talks about compute pricing power — it wants to transform from a chip supplier into an AI industrialization infrastructure company. Vera Rubin isn’t just a new GPU; it’s packaging GPU, CPU, NVLink, networking, storage, liquid cooling, and power management into an AI factory system.

TSMC talks about advanced manufacturing fabs. Micron this time wants to talk about advanced memory supply rights. It doesn’t have Nvidia’s software ecosystem or TSMC’s strong manufacturing moat, but it sits right at another bottleneck of AI infrastructure: AI needs more and more memory, and advanced memory supply can’t be quickly replicated.

Profit migration flows layer by layer along the industry chain. Phase one: hyperscaler capex flows to Nvidia, GPUs become the first valuation anchor of the AI bull market. Phase two: Nvidia delivers more AI systems, HBM and high-end storage become new bottlenecks, and Micron, SK Hynix, and Samsung capture profit elasticity. Phase three: storage manufacturers are forced to increase capex to lock future orders, and equipment and materials companies stand at the entrance of the next order wave.

The simplest and most effective saying in investment in recent years: Follow the money.

Micron signed many long-term contracts, locked future prices, and tried to escape the cycle — but this also means prices are locked, losing some profit elasticity from price increases. It has to meet market expectations through capacity expansion. If every chip company wants to expand, who decides whether capacity can actually be delivered? The answer is realistic: lithography, etching, deposition, metrology, inspection, and advanced packaging equipment.

If the three major memory makers expand capacity combined with sovereign AI investments from China, Japan, and the US, global wafer fab equipment (WFE) spending could reach $170-180 billion — the equipment industry moves from “cyclical recovery” to “supercycle pricing.” According to institutional forecasts, LRCX profit growth could hit 60-90%, AMAT 45-65%, KLAC 35-55%. Without significant valuation compression, theoretical stock price upside could correspond to 40-100%.

Short, Medium, and Long Term: What to Watch

Short term, stock prices will naturally fluctuate. Earnings and guidance were very strong, and there may be post-expectation profit-taking. A $160 billion end-of-quarter rebalancing sell order could bring short-term pressure. Traders will watch if good news is fully priced in, if options structures amplify volatility, and if funds temporarily rotate away from the AI hardware theme.

Medium term, the market will watch three things: Can SCAs continue to expand? Can RPO and deposits keep growing? Can HBM/AI revenue continue to climb? As long as these three lines are delivering, Micron is not just a cyclical rebound — it’s validating a business model revaluation.

Long term, the real question is only one: Can these $100 billion+ long-term orders transform Micron from a “memory price elasticity company” into an “AI infrastructure supply rights company”? If yes, Micron’s valuation system will be rewritten — the market will give it not just the cyclical PE multiple corresponding to the next cyclical peak, but an asset revaluation reflecting higher revenue visibility, stronger cash flow quality, and a more stable profit floor. If no, it will eventually go back to the old script: price increase, expansion, oversupply, valuation compression.

But at least from this conference call, Micron has put a question before the capital markets: Are you still going to use the old cyclical framework to price a memory company holding $100 billion+ long-term contracts, customer deposits, and AI supply rights?

Final Thoughts

The earnings report was just the entrance. The $100 billion+ long-term contracts are the main feature.

I want to add a few less “investment-focused” thoughts.

Every wave of technology, as it reaches its middle stage, presents a similar turning point:

The market shifts from chasing “who has the sexiest story” to asking “whose position is most irreplaceable.”

The internet era was about bandwidth and servers; the mobile era was about baseband chips and display panels; the AI era now turns to memory and advanced packaging. Technology changes, but the underlying law of industrial competition hasn’t changed — what’s truly scarce is never imagination, but the physical things in the world that can’t be replicated with code.

What Micron did this time is essentially to fix a physical scarcity into contract form, and then let the capital market reprice that scarcity. It’s similar to the logic TSMC used to lock in Apple and Nvidia with advanced process technology — you’re not selling a product; you’re selling a structural position that says “you can’t do it without me.”

In the AI era, many companies talk about the future. But what Micron did this time, the most impactful thing, is that it wrote part of its future revenue, capacity, and customer relationships into contracts ahead of time.

This is not your typical cyclical recovery.

It’s the first serious attempt by the storage industry to bind the future into a long-term contract that the capital market can reprice.

And for us observers, perhaps the most valuable lesson is a simpler one:

In an era where everyone talks about how AI will change the world, the ones who truly make money are often those who control the physical resources necessary for AI to run.

Algorithms can be open-sourced, models can be iterated, but wafer capacity, HBM yield rates, and the physical limits of advanced packaging — those have no shortcuts.

The point where technological revolution translates into commercial value is never at the most dazzling place, but at the point that cannot be bypassed.

Let’s keep that in mind.