AI Reshapes Storage Architecture: HDDs and NAND Coexist Rather Than Replace Each Other

Overview

This research report, based on an interview transcript with Robert Soderbery, former Executive Vice President of Western Digital, delves into the evolving relationship between hard disk drives (HDDs) and flash memory (NAND/SSDs) in data centers in the AI era. The key conclusion is that AI has not simply rendered HDDs obsolete; rather, it has driven storage architectures toward a more complex three-tier model. While the AI preprocessing phase has sparked a “gold rush” for high-capacity enterprise‑grade SSDs, pushing NAND prices sharply higher, the substantial price differential—exceeding 20 times—makes full‑scale NAND substitution economically unviable. By contrast, HDDs, leveraging their cost advantage in high‑capacity storage and the supply discipline inherent in an oligopolistic market, are poised to remain a critical component over the long term. The report also highlights the shifting dynamics of long‑term agreements (LTAs) and the potential of emerging technologies such as high‑bandwidth flash (HBF).

Core views

AI-Driven Storage Architecture Reconfiguration: Traditional data centers have long relied on a two-tier architecture—small-capacity, high-performance SSDs for compute and large-capacity nearline HDDs for storage—with HDDs accounting for 80–85% of total storage capacity. However, AI workloads—such as vectorization and embedding processing—demand exceptionally high performance during the pre-computation phase, a capability that conventional HDDs cannot deliver. This has fueled surging demand for high‑capacity enterprise SSDs ranging from 32 TB to 256 TB, leading to a temporary shift in new AI‑centric deployments toward NAND-based solutions and establishing a three-tier architecture: compute‑oriented SSDs, high‑capacity enterprise SSDs, and nearline HDDs. Economic Constraints Hinder NAND’s Replacement of HDDs: Despite the explosive growth in NAND demand, its price surge has widened the cost per gigabyte gap with HDDs to 20–25 times. Research reports indicate that hyperscale cloud providers typically reach a total cost of ownership (TCO) inflection point when the NAND‑HDD price differential stands at 2–3 times. Given the current magnitude of this disparity, fully replacing HDDs with SSDs remains economically unattractive. Moreover, for NAND manufacturers to displace HDDs entirely, they would need to invest roughly $150 billion in multiple process node iterations, while achieving only modest gross margins of around 30%. By contrast, maintaining higher‑margin AI‑specific storage businesses proves far more appealing from an investment perspective. Consequently, market expectations suggest a partial pullback in demand toward HDDs, particularly as AI operators refine their architectures to contain costs. HDD Oligopoly and Supply Discipline: The HDD market is tightly dominated by Western Digital (WDC), Seagate, and Toshiba. Drawing lessons from the oversupply episodes of the past decade, these vendors are deliberately avoiding indiscriminate capacity expansion. HDD technology—encompassing heat-assisted magnetic recording (HAMR), precision mechanics, and advanced materials—faces substantial physical and technical barriers to scaling production, thereby underpinning long-term supply discipline. In contrast, the recent NAND price spike stems primarily from AI-driven demand concentrated in cutting-edge, high‑capacity nodes that account for only 30–35% of overall industry capacity, creating a structural shortage. Evolution of Long-Term Agreements (LTAs): The industry is transitioning from historically one-sided LTA frameworks that constrained suppliers to truly reciprocal long-term contracts. Under the new model, customers are required to provide firm purchase commitments and financial guarantees; for instance, SanDisk disclosed a $42 billion procurement obligation, of which $11 billion is backed by financial assurances. This shift enhances supplier planning visibility and supports capital investment. Although contracts may be renegotiated during severe market downturns, such developments mark a significant maturation of industry pricing mechanisms. Emerging Technologies and the Impact of Edge Computing: Any migration from cloud‑based to edge‑computing environments strongly favors NAND, as edge devices face stringent constraints in space and power consumption—areas where NAND outperforms HDDs. Additionally, High‑Bandwidth Flash (HBF), an emerging technology featuring HBM‑like stacking and wide‑interface packaging, leverages the characteristics of AI inference—namely, streaming data flows rather than frequent writes—and, combined with NAND’s high density and low cost, holds promise as a competitive solution for certain AI workloads, albeit still in its early innovation stages.

Analysis framework

Institutional investors gather first-hand industry insights through expert interviews and derive conclusions by integrating supply-and-demand frameworks with total cost of ownership (TCO) analysis. First, by reviewing the historical evolution of system architectures, they identify a new variable introduced by AI workloads—namely, pre‑compute performance requirements—thereby explaining the short‑term surge in demand for high‑capacity SSDs. Second, employing substitution‑effect analysis from economics, they compare the price‑to‑cost ratio between NAND and HDD with the TCO crossover threshold to demonstrate that full substitution is economically unviable. The underlying logic here is that substitution will only occur when the substitute’s price falls below a critical threshold and non‑monetary factors such as space and power consumption are factored in. Third, from the perspective of industrial organization theory, they examine the oligopolistic behavior of the HDD market, highlighting its strategy of controlling production capacity to sustain profit margins—a dynamic that contrasts sharply with the NAND market, which experiences cyclical volatility due to capital intensity and rapid technological iteration. Finally, by assessing the technical characteristics and application scenarios of emerging technologies such as HBF and QLC, they evaluate their potential impact on the future competitive landscape, rather than focusing solely on current market shares.

Methodology notes

• Industry/ Sector Analysis FrameworkSupply-and-Demand Framework

  Structural supply-demand imbalance

  The research report distinguishes between the distinct drivers behind price increases in NAND and HDD: NAND is experiencing a structural shortage driven by AI demand concentrated on the most advanced process nodes, which account for only 30–35% of capacity, while HDD prices are being supported by overall demand that slightly exceeds supply. This granular analysis helps to better assess price elasticity and the sustainability of these trends.

• Company Fundamentals and Financial FrameworkROIC–WACC spread

  Capital Allocation Efficiency and the Rate of Return Threshold

  Analyses indicate that for NAND vendors to displace HDDs through capacity expansion would require massive capital expenditures—on the order of $150 billion—while yielding only modest gross margins of around 30%, far below the high margins of AI‑oriented storage businesses. From an investor’s perspective, projects with low return on invested capital (ROIC) are unattractive, leaving manufacturers with little incentive to fundamentally disrupt the HDD market.

• Competition and Strategic FrameworkMoat / competitive advantage

  Oligopolistic Competition and Barriers to Entry

  The HDD market is widely regarded as a classic oligopoly, with its competitive moats rooted not only in market share but also in the barriers to capacity expansion arising from technological complexity—such as HAMR and precision manufacturing. These “soft” entry barriers enable incumbent players to coordinate supply, avert destructive price wars, and sustain long-term profitability.

• Valuation MethodPE/PEG valuation

  Periodic Profit Normalization and Valuation

  The research report values SanDisk at 11 times its four-year average EPS, while assigning WDC and Seagate a valuation of 21 times their FY28 EPS. This approach seeks to smooth out the pronounced cyclical volatility of the semiconductor storage industry, reflecting its fair value at the midpoint of the cycle rather than basing valuations solely on current peak or trough earnings.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Western Digital Corp (WDC.US)

  Benefit: As one of the HDD oligopolists, the company stands to gain from sustained demand for nearline HDDs in AI data centers and from margin expansion driven by supply discipline.

  Strengths

  HAMR technology leadership, vertical integration capabilities, and a robust balance sheet.

  Weaknesses

  The NAND business may be weighed down by cyclical fluctuations.

  Comparison

  Compared with Seagate, WDC operates a NAND business, which diversifies its revenue streams but also adds complexity.

  Risks

  The period of digesting cloud capital expenditures and the yield risks associated with the transition to HAMR technology.

• Seagate Technology PLC (STX.US)

  Benefit: Pure-play HDD players stand to gain directly from stable HDD prices and the consolidation of market share.

  Strengths

  Focuses on HDDs, boasts high operational efficiency, and is making steady progress with HAMR technology.

  Weaknesses

  The absence of hedging in the NAND business creates exposure to risks associated with a single product line.

  Comparison

  Compared with WDC, Seagate is more directly exposed to the HDD cycle; should HDDs be replaced, its downside risk would be greater. However, at present, the threat of substitution appears remote.

  Risks

  WDC is rapidly closing the gap in HAMR, amid shifting purchasing patterns among cloud customers.

• SanDisk Corp (SNDK.US)

  Benefits: As a leading supplier of high-capacity enterprise‑grade NAND, the company stands to gain directly from the explosive growth in storage demand during the AI preprocessing phase.

  Strengths

  QLC technology leads the industry, backed by a strategic partnership with Kioxia and robust long-term agreement (LTA) lock‑in capabilities.

  Weaknesses

  Capital-intensive businesses face significant depreciation pressures.

  Comparison

  Compared with Samsung and Micron, SanDisk has a more prominent position in the enterprise‑grade high‑capacity NAND sector.

  Risks

  NAND prices are experiencing cyclical declines, technological iteration has failed, and there is a risk of asset impairment.

Key data

• NAND vs. HDD price spread multiple>20xCurrently, the price gap per GB is more than two to three times the level required at the TCO crossover point.
• The proportion of NAND in the AI-related business60-70%Primarily driven by high-capacity enterprise-grade SSDs.
• Latest node NAND capacity share30-35%AI demand is concentrated in this area, resulting in structural shortages.
• SanDisk LTA Financial Guarantee Ratio~25%Of the $42 billion in procurement obligations, $11 billion is backed by financial guarantees.
• Number of HDD manufacturers2.5 companiesWDC, Seagate, and the struggling Toshiba

Impact & implications

Impact on Relevant Companies: 1. **Western Digital (WDC) & Seagate (STX)**: As HDD market leaders, they benefit from supply discipline and sustained demand for nearline storage in AI data centers. Their technological leadership in HAMR, coupled with an improving cost‑down trajectory, strengthens their long-term competitiveness. The research report maintains an “Outperform” rating, noting that valuations remain attractive relative to high growth expectations. 2. **SanDisk (SNDK)**: The company is poised to capitalize on robust AI‑driven demand for high‑capacity QLC NAND. The signing of a two‑way LTA enhances revenue visibility and cash‑flow stability. Despite cyclical risks, SanDisk’s position in the premium NAND segment remains solid. 3. **Pure Storage**: Its Direct Flash Module (DFM) architecture, which sidesteps traditional form‑factor constraints and enables more efficient utilization of NAND, has won favor among hyperscale cloud providers, underscoring the viability of non‑conventional storage designs. 4. **The Industry as a Whole**: The storage sector is shifting from pure commoditization toward differentiated competition—through innovations such as HBF, QLC optimization, and custom architectures. Price volatility is expected to moderate, ushering in a “new normal,” barring extreme events like an AI bubble burst.

Risks

• The bursting of the AI bubble has triggered a sharp drop in demand, sending memory prices into free fall.
• NAND technology is advancing faster than expected, significantly narrowing the cost gap with HDDs and accelerating the substitution process.
• Capital expenditures by hyperscale cloud service providers have entered a digestion phase, leading to reduced procurement of new storage equipment.
• HDD manufacturers have encountered yield and reliability challenges during their transition to HAMR technology, which has weighed on their gross margins.
• Long-term contracts are being renegotiated during sharp market downturns, thereby weakening their price-supporting effect.

What to watch

• Whether the price gap between NAND and HDDs falls back below 3x will trigger a reassessment of the TCO crossover point.
• Key capital expenditure plans and capacity expansion progress of major NAND manufacturers, particularly regarding investments in the latest process nodes.
• The commercialization progress of HBF (High-Bandwidth Flash) technology and its adoption rate in AI inference clusters.
• The evolving trends in the storage architectures of hyperscale cloud service providers (such as Google and Meta), and whether there are signs of a shift back toward HDDs.
• The penetration rate and reliability performance of QLC NAND in the enterprise‑grade market.