Author: hoidya | 0xU
1/ What exactly is the storage industry?
The storage industry is primarily composed of three core products: DRAM, NAND, and HBM. Together, they form the data memory system for all digital devices. Whether it’s a smartphone, computer, or data center, all rely on this foundational infrastructure to enable temporary data processing and long-term storage.
Functionally, DRAM is used for temporary data storage during operation, meeting high-speed read and write demands during computation. NAND is used for long-term data storage, serving as a persistent memory layer for devices. HBM is a newer architecture developed for high-performance computing environments to address bandwidth bottlenecks between GPUs and computing units.
From a system architecture perspective, the storage industry is not an independent component separate from computing systems, but rather the foundational dependency layer of all computing systems. Any computing task must first "read data," then perform "computation," and finally "write back the results." Therefore, storage is one of the fundamental constraints in the computing process, not an optional module.
Over the past two decades, demand in this industry has primarily come from three sources: consumer electronics (mobile phones and PCs), enterprise servers, and internet infrastructure. These demands share common characteristics: high dispersion, delayable update cycles, and limited individual demand scale. As a result, the market has long classified it as a typical cyclical semiconductor industry.
2/ Why has storage long been regarded as a cyclical industry?
The storage industry has long exhibited strong cyclicality due to an imbalance in supply and demand dynamics. Demand is typically tied to consumer electronics cycles and enterprise IT spending cycles, while supply is driven by wafer fab investments, which introduce significant time lags.
When demand rises, prices surge rapidly, prompting manufacturers to expand production. However, since capacity construction typically takes 12 to 24 months, new supply often peaks after the demand turning point, leading to a sharp price decline. This mechanism creates a classic boom-bust cycle.
Between 2010 and 2022, this cyclical structure was particularly evident. For example, the DRAM industry repeatedly experienced rapid declines from high-margin states into losses, followed by rebounds as new demand recovered. These fluctuations have led the market to consistently view the storage industry as a cyclical asset class characterized by high volatility and low predictability.
At this stage, the industry’s pricing mechanism is essentially inventory-driven. Prices rise when inventory decreases and fall when inventory accumulates, while demand itself acts more as a triggering variable than a structural one.
3/ What was the demand structure before AI?
Before the emergence of artificial intelligence, storage demand was primarily driven by consumer electronics and traditional internet infrastructure. Consumer electronics are characterized by long update cycles and relatively predictable demand—for example, smartphone replacement cycles typically range from two to three years. Server and enterprise storage, on the other hand, are more dependent on the rhythm of IT capital expenditures, also exhibiting strong cyclical patterns.
In this structure, storage is offered as a standardized product, with prices primarily determined by supply and demand rather than long-term commitments from large individual customers. As a result, the market exhibits strong spot characteristics, allowing price signals to quickly reflect changes in inventory and production capacity.
In other words, before AI, the demand structure in the storage industry was fragmented and lacked long-term rigid constraints—this formed the fundamental basis for its cyclical nature.
4. Why has AI completely transformed the structure of storage demand? (From commodities to infrastructure)
Past storage demand was driven by consumer electronics (smartphones, PCs) and was essentially "delayable consumption." However, AI introduces a completely different demand function: it is a continuous computing system, with memory usage growing linearly or even superlinearly with model size.
Taking AI data centers as an example, during training and inference, GPUs are not limited by computational power but by memory bandwidth, which directly drives the essential demand for HBM. Industry data shows that demand for high-bandwidth memory in AI servers is growing at a rate far exceeding that of traditional DRAM, leading to long-term allocation of HBM production capacity—and even full pre-sales through 2026.
More critically, supply-side changes are occurring: since HBM offers significantly higher profit margins than traditional DRAM, manufacturers are actively reallocating wafer capacity from DDR4/DDR5 to HBM production. This structural crowding-out effect is causing a “non-demand-driven shortage” in traditional DRAM and NAND.
The market has shown extreme signals: spot prices for certain DRAM and NAND products have risen 15–20% within the quarter, with instances of "intraday price adjustments."
5. How was storage priced in the past?
Between 2010 and 2022, the pricing mechanism in the storage industry was highly typical, following a standard semiconductor cycle model:
Prices are driven by inventory cycles, not demand structure.
When inventory declines → prices rise → manufacturers increase production → supply exceeds demand → prices collapse.
The core constraint of this mechanism is "production capacity lag (1–2 years) + demand deferability."
For example, during the previous cycle, the DRAM industry frequently experienced significant quarterly fluctuations in profits, even shifting from high margins to losses and quickly reversing again.
But this mechanism has been disrupted in the AI era, due to two variables changing simultaneously:
- First, demand has shifted from decentralized consumption to centralized procurement.
- Second, supply shifts from "free-market expansion" to "profit-driven allocation (HBM priority)."
The result is: cyclical fluctuations still exist, but price elasticity has been structurally compressed.
6/ What structural changes are currently occurring?
The key shift in the current (2024–2026) storage market is not price increases, but a transition from a "spot market" to a "contract allocation system."
First is the crowding-out effect of HBM. Due to significantly higher per-wafer profits compared to DDR4/DDR5, Samsung, SK hynix, and Micron are all prioritizing the allocation of production capacity toward HBM. Industry data shows that HBM is rapidly rising from a low single-digit share to a structural level of over 40% of DRAM revenue.
This structural adjustment has led to two outcomes:
- First, traditional DRAM supply is contracting.
- Second, NAND has entered a state of passive shortage.
Meanwhile, the market has entered an extreme supply-demand state: DRAM industry revenue grew 17.1% year-over-year in Q2 2025, but this growth was driven by rising prices and supply constraints, not by a surge in demand.
Even more extreme signals come from the supply side: industry lead times have extended from the normal 8–12 weeks to 39–52 weeks, with some automotive-grade memory exceeding 70 weeks.
This signifies a key structural change: memory is no longer a "freely tradable commodity," but has become a "rationed resource."
This will create a positive feedback loop:
Price increases → Manufacturers reduce spot supply → Buyers lock in orders early → Further reduction in spot liquidity → Price continues to rise.
7/ Who benefits from this structure?
The profit structure of the storage industry is undergoing a significant shift.
Layer 1: Supply Side (Samsung / SK hynix / Micron)
These companies are transitioning from "cyclical manufacturers" to "AI infrastructure providers." SK hynix's leadership in HBM has gradually positioned it as a structural pricing power holder, with its DRAM market share rising to approximately 38%.
Layer 2: Demand Side (Microsoft / AWS / Google)
These companies lock in future supply through long-term contracts, essentially engaging in "time arbitrage": using current capital expenditures to secure future AI computing and memory costs.
Layer 3: AI model companies (such as OpenAI)
They operate between cash flow pressure and hashing power demand, forming a closed loop through financing → capex → renewed supply locking.
The key change is that pricing power is shifting from the "market" to the "contract structure."
8 / Risks and Falsification Conditions
This round of the "AI memory supercycle" has at least three clear falsification conditions:
First, if AI capex enters a contraction cycle (as hyperscalers reduce investment intensity), the current demand structure will quickly become distorted, since memory demand is heavily dependent on AI compute expansion.
Second, if the HBM technology path is replaced (e.g., by new memory architectures or compute-memory fusion), the current HBM premium will be compressed, leading to a reallocation of capacity back to DRAM/NAND.
Third, if the capacity expansion cycle accelerates again (e.g., Samsung or SK hynix re-enters aggressive expansion), the current supply constraints will reverse into a period of oversupply within one to two years.
In other words, this structure is predicated on:
The growth rate of AI demand exceeds the speed of capacity expansion plus the speed of technological substitution.