Samsung AI Memory Revenue Has Redrawn the Storage Market
Samsung AI memory revenue has entered a different league, after the company reported record second-quarter results driven by constrained AI infrastructure demand. The change is visible across six years of supplier earnings, but it is not a simple storage recovery. Memory manufacturers now sit far above the companies that sell arrays, drives, data platforms, and protection software.
A six-year revenue review published September 7 shows Samsung, SK hynix, and Micron separating sharply from traditional storage suppliers. Their acceleration began after generative AI infrastructure spending intensified in late 2023. Kioxia and Sandisk later joined the upswing as enterprise NAND supply tightened.
The comparison exposes a reversal in where storage economics accumulate. Dell still leads the lower-revenue group, while Western Digital, Seagate, NetApp, HPE, and other specialists are growing. Yet the largest gains now sit deeper in the supply chain, where scarce memory components serve expensive AI systems.
Samsung AI Memory Revenue Opens a New Gap
The defining change is not that storage revenue increased. It is that AI memory suppliers moved onto an entirely different growth curve.
Samsung reported consolidated revenue of KRW 171.5 trillion for the quarter ending June 30, 2026. That represented a 28 percent sequential increase and another company record. Its Device Solutions division generated KRW 127.5 trillion, rising 56 percent from the previous quarter.
The division includes memory, foundry, and system semiconductor operations, so its revenue is not a pure storage measure. Samsung nevertheless said its memory business set quarterly records for both revenue and operating profit. Server products took a record share of the sales mix.
That distinction matters when interpreting Samsung AI memory revenue. The headline number includes businesses beyond DRAM, NAND, and high-bandwidth memory. HBM is stacked DRAM designed to move data rapidly between memory and AI processors.
Samsung attributed the record quarter to AI demand, limited capacity, and rising industry prices. The company focused available supply on server products, where customers were paying for performance and capacity. Its quarterly results also described growing sales of HBM4, DDR5, enterprise SSDs, and specialized memory modules.
This was a dramatic change from the preceding year. Samsung’s Device Solutions division produced KRW 27.9 trillion in revenue during the second quarter of 2025. One year later, its reported division revenue was more than four times larger.
The comparison includes foundry and system-chip sales, which prevents a clean memory-only calculation. Even so, Samsung identified memory as the main driver of the division’s record. The direction is therefore clearer than the exact storage contribution.
SK hynix shows the same acceleration through a narrower business mix. The company reported KRW 79.3187 trillion in second-quarter revenue, up 257 percent from one year earlier. Operating profit reached KRW 60.5426 trillion.
Its financial release credited high-value DRAM, HBM, and NAND products for the performance. The company had generated KRW 22.232 trillion in revenue during the second quarter of 2025. It also said first-half revenue exceeded KRW 100 trillion for the first time.
Micron completed the pattern among the three largest memory suppliers. It reported fiscal third-quarter revenue of $41.46 billion, compared with $23.86 billion in the preceding quarter. Revenue had been $9.30 billion during the same period one year earlier.
The scale of that change explains why conventional storage suppliers appear compressed on a shared chart. Micron, SK hynix, and Samsung are receiving both higher unit demand and sharply higher selling prices. Traditional vendors generally depend on narrower product categories and slower enterprise purchasing cycles.
Samsung AI memory revenue therefore marks more than a strong quarter. It shows that AI has changed which layer of the storage stack captures the largest revenue gains. The companies controlling scarce memory fabrication now hold the strongest position.
AI Demand Moved the Bottleneck Into Memory
AI infrastructure shifted storage demand from a supporting requirement into a capacity constraint that affects entire systems.
Training a large model requires processors to read and update huge collections of parameters. HBM supplies the bandwidth needed beside accelerators, while DRAM supports servers and NAND stores datasets, checkpoints, and inference data. These products address different tasks, but the same infrastructure expansion pulls demand across all three.
Early AI investment concentrated attention on graphics processors. Storage was often treated as an adjacent category that would grow after compute deployments. By 2026, that sequence had tightened considerably.
Dell’Oro Group reported that revenue for data center semiconductors and components rose 116 percent year over year during the first quarter. The firm said DRAM contributed the largest share of growth in both percentage and absolute terms. Storage, HBM, and high-speed networking benefited alongside accelerators.
This creates a chain reaction. More accelerators require more nearby memory, while larger clusters generate additional data for storage systems. Organizations also need capacity for model checkpoints, retrieval datasets, logs, synthetic data, and inference results.
Agentic AI adds another layer of demand. An agent can perform multiple steps, call tools, retrieve information, and retain intermediate results during a task. Production deployments can therefore create persistent data beyond the original model and its training corpus.
Enterprise SSD demand reflects that expansion. TrendForce estimated that combined revenue for the five largest enterprise SSD brands reached approximately $37.59 billion in the second quarter. That represented a 103.6 percent sequential increase.
Samsung led the group with approximately $14.35 billion in enterprise SSD revenue. SK hynix Group followed with more than $8.63 billion, while Micron reached approximately $6.98 billion. Kioxia and Sandisk generated approximately $4.64 billion and $2.98 billion, respectively.
The enterprise SSD data connected those increases to higher shipments and contract prices. It also highlighted a transition among North American cloud providers from PCIe 4.0 to PCIe 5.0. PCIe is the interface that connects storage devices and other components to a server.
Faster interfaces do not create demand by themselves. They make high-performance SSDs more valuable when data pipelines would otherwise leave costly processors waiting. A storage delay becomes harder to tolerate when it reduces the utilization of an entire AI cluster.
Suppliers have responded by changing their product mix. Samsung increased shipments of 176-layer QLC products and expanded its higher-end PCIe 5.0 mix. QLC NAND stores four bits in each memory cell, increasing density while introducing endurance and performance tradeoffs.
SK hynix expanded shipments of 321-layer TLC products. Its Solidigm subsidiary used QLC technology for very high-capacity enterprise SSDs. Micron directed more capacity toward enterprise SSDs and HBM, including products based on 232-layer QLC.
These choices show why Samsung AI memory revenue is part of a wider mechanism. Manufacturers are not simply producing more of every memory product. They are steering constrained capacity toward server components with the greatest demand and revenue potential.
That prioritization also shifts pressure downstream. Storage-system vendors must secure drives and memory while managing higher component costs. Enterprise buyers face longer planning horizons, tighter supply agreements, and a stronger incentive to reserve capacity early.
The bottleneck has moved, but it has not disappeared. AI infrastructure can only expand as quickly as the supply chain delivers processors, memory, storage, networking, power, and cooling. Memory suppliers benefit because their products now constrain more of that system.
Component Makers Are Outrunning Storage Systems
The central contest is between scarce component suppliers and the system vendors that package those components for enterprise customers.
Dell illustrates the difference. The company reported record fiscal second-quarter revenue of $47.0 billion, supported by exceptional AI server demand. Its storage revenue reached $4.9 billion, up 26 percent year over year.
That is strong growth for an established systems vendor. However, Dell’s AI-optimized server revenue reached $16.4 billion during the same quarter. The company also reported $60.9 billion in AI server orders and a $95 billion backlog.
Dell’s quarterly disclosure captures the tension. AI demand is lifting the entire infrastructure portfolio, including storage. Yet servers containing accelerators and scarce memory are expanding much faster than the storage category itself.
Western Digital and Seagate form another important comparison. Both hard-drive suppliers have risen from the cyclical weakness that followed their 2021 and 2022 peaks. Their revenue trajectories strengthened during the second half of 2024 and continued upward through 2026.
Hard disk drives remain important because AI does not place every byte on flash. Large datasets, backups, archives, and lower-access storage tiers still favor HDD capacity economics. AI training and inference can therefore increase demand for both flash performance and disk capacity.
Western Digital’s corporate structure also changed during this period. The company completed the separation of its flash business into Sandisk in February 2025. Comparisons across the separation require care because later Western Digital results exclude Sandisk’s flash operations.
The remaining hard-drive company reported fiscal fourth-quarter revenue of $3.747 billion, up 44 percent year over year. Its full-year revenue reached $12.919 billion. Those gains confirm that AI-driven data creation is benefiting storage beyond HBM and enterprise SSDs.
Yet the magnitude still differs from the memory leaders. Western Digital and Seagate sell essential capacity, but they do not capture the same scarcity premium as HBM and DRAM suppliers. Their manufacturing constraints, customer contracts, and replacement cycles also follow different economics.
NetApp represents the enterprise array and data-management layer. It reported quarterly revenue of $2.03 billion for its fiscal first quarter of 2027, up 30 percent year over year. Its all-flash array revenue reached $1.3 billion, increasing 47 percent.
Those numbers indicate meaningful demand for systems that organize, protect, and serve enterprise data. They also demonstrate how the market rewards software integration, data services, and installed customer relationships. Still, NetApp remains far below the quarterly revenue scale reached by the largest memory manufacturers.
HPE’s storage business followed a less direct path. Its revenue declined modestly from 2020 through 2024 before recovering. Alletra adoption and renewed infrastructure demand have supported improvement, although the company has not matched the component suppliers’ acceleration.
Snowflake, Everpure, and Nutanix occupy different positions despite appearing near one another in revenue comparisons. Snowflake sells a cloud data platform, while Everpure combines storage hardware with subscription services. Nutanix focuses on hybrid cloud infrastructure and virtualization.
These companies should not be treated as interchangeable storage suppliers. Their revenue recognition, gross margins, hardware exposure, and customer contracts differ. A shared chart remains useful for scale, but it cannot determine operating quality or market share.
Rubrik and Commvault add cyber resilience to the picture. Their products help customers protect and recover data rather than manufacture storage media. Growing security concerns can lift their revenue even when underlying component conditions move differently.
The outcome is a layered market, not a single leaderboard. Memory manufacturers benefit first from scarcity and price increases. Drive makers benefit from growing capacity requirements, while systems and software vendors must translate data growth into differentiated services.
What the Revenue Curves Do Not Prove
The charts reveal a historic separation, but they do not prove that every dollar represents durable storage demand.
The first limitation is scope. Samsung’s consolidated revenue includes smartphones, displays, consumer electronics, and other operations. Its Device Solutions figure includes foundry and system-chip revenue alongside memory.
SK hynix has a more concentrated memory business, although its results still include different DRAM and NAND categories. Micron also sells products across data center, mobile, automotive, embedded, and consumer markets. Comparing total company revenue can exaggerate direct competition.
Fiscal calendars create another problem. Companies close quarters on different dates and label fiscal years differently. Normalizing results to HPE’s quarter sequence makes a chart easier to read, but it does not make reporting periods identical.
Currency movements also affect the comparison. Samsung and SK hynix report in Korean won, while several other companies report in US dollars. A converted revenue line can move because of exchange rates even when local-currency sales remain unchanged.
Corporate separations introduce further discontinuities. Western Digital’s flash separation means its current results represent a different business from its earlier consolidated results. Sandisk’s independent history begins with assets that previously sat inside Western Digital.
Product prices present the largest uncertainty. Rising revenue can come from additional unit shipments, higher prices, or a better product mix. During a constrained cycle, price inflation can make supplier revenue grow much faster than the underlying quantity of memory delivered.
That pattern is especially relevant in 2026. Gartner forecast that annual DRAM prices would increase 125 percent, while NAND prices would rise 234 percent. It projected worldwide memory revenue of $633.3 billion, nearly three times the 2025 total.
The same semiconductor forecast warned that memory inflation was severe but temporary. It expected meaningful pricing relief no earlier than late 2027. Gartner also said non-AI demand would face delays as expensive components redirected supply and budgets.
That warning complicates the bullish interpretation of Samsung AI memory revenue. Record supplier sales demonstrate pricing power and urgent demand. They do not establish that current growth rates will persist after new capacity arrives.
High prices can also reduce demand elsewhere. PC and smartphone manufacturers may lower memory specifications, delay launches, or accept weaker margins. Enterprise buyers may postpone general-purpose upgrades while reserving spending for AI projects.
Supply expansion presents its own risk. Memory markets have historically moved through sharp cycles because new fabrication capacity arrives slowly and in large increments. Tight conditions can become oversupply when producers expand together and demand growth moderates.
Chinese suppliers represent another variable, particularly in NAND. TrendForce expects rising Chinese output and domestic cloud demand to influence the competitive landscape. Additional supply could pressure prices, though qualification requirements limit how quickly enterprise customers change vendors.
Technology transitions add execution risk. Suppliers must qualify new HBM generations and higher-layer NAND with demanding customers. A product can enter sampling without reaching high-volume production on schedule.
HBM also consumes manufacturing resources differently from conventional DRAM. Its stacked design requires advanced packaging, thermal management, and strict quality control. Expanding wafer capacity alone does not guarantee an equal increase in finished HBM output.
Customer concentration deserves attention as well. A limited group of hyperscalers and AI infrastructure companies accounts for much of the current investment. Their purchasing decisions can move supplier demand more quickly than a broad consumer market.
The current curves therefore support a narrower conclusion. AI infrastructure has changed the revenue hierarchy, and memory suppliers hold the strongest position today. The charts cannot guarantee how long that position lasts.
The Memory Supercycle Has Uneven Winners
AI is raising revenue across storage, but the gains depend on each supplier’s proximity to the constrained components.
Samsung, SK hynix, and Micron occupy the most favorable position because they manufacture memory at enormous scale. They can redirect capacity toward HBM, server DRAM, and enterprise NAND when those products offer better returns.
Kioxia and Sandisk benefit from the NAND shortage, particularly through enterprise SSDs. Their second-quarter acceleration shows that the AI effect now extends beyond the memory placed beside accelerators. Data must also move into high-capacity storage during training and inference.
Western Digital and Seagate benefit through another route. AI systems generate and retain large data collections that do not need flash-level performance at every moment. High-capacity hard drives can support colder tiers, archives, and backup environments.
Dell combines storage with servers, networking, services, and PCs. Its scale helps it capture broad infrastructure spending. However, its fastest growth comes from AI servers rather than storage alone.
NetApp, HPE, Everpure, and Nutanix sell enterprise platforms that sit above the media layer. Their opportunity depends on making data accessible, protected, and manageable across on-premises and cloud environments. They are less exposed to raw memory prices, but they must absorb or pass through component costs.
Snowflake represents the shift toward data platforms rather than physical storage ownership. Its customers pay to analyze and manage information in cloud environments. Growth can reflect higher data consumption even when Snowflake does not manufacture the underlying equipment.
Rubrik and Commvault benefit when organizations decide that more valuable data requires stronger recovery controls. Ransomware exposure, compliance, and operational resilience can expand their markets independently of HBM prices. Their business models also produce recurring revenue patterns that differ from hardware cycles.
Backblaze has pursued long-duration storage opportunities serving neocloud providers. A neocloud is an infrastructure operator focused heavily on GPU and AI workloads. These providers need lower-cost capacity for datasets and outputs that do not remain on premium flash.
The common theme is data multiplication. AI systems consume existing information, generate new artifacts, and often retain extensive operational records. That cycle supports more suppliers than the headline memory winners alone.
Still, the value distribution remains uneven. A component shortage lets manufacturers charge more for each available unit. A storage software vendor must instead win workloads, retain subscriptions, and differentiate its management layer.
System vendors face both opportunity and exposure. They can sell larger configurations, but rising component costs complicate pricing and delivery. Buyers may also redirect budgets from conventional infrastructure toward a smaller number of AI projects.
Enterprise customers therefore need to separate three questions. They should assess how much capacity their workloads require, which performance tier needs premium media, and how long current supplier pricing will remain constrained.
The answers differ for model training, retrieval systems, backup, analytics, and everyday business applications. Putting every workload on the fastest storage wastes money. Keeping active AI data on slower tiers can leave expensive processors idle.
This balancing act supports hybrid architectures. HBM serves the processor, DRAM holds active server data, flash handles performance-sensitive storage, and hard drives support capacity-oriented tiers. Software coordinates placement, access, protection, and recovery.
The largest suppliers do not win every layer. They win the layer where scarcity is currently strongest. That is why the revenue divergence is real without making smaller vendors irrelevant.
Three Signals Will Test the New Storage Order
The next phase depends on whether supply remains tight, enterprise storage catches up, and buyers continue funding AI infrastructure at today’s pace.
The first signal is third-quarter contract pricing for DRAM and enterprise NAND. Continued increases would strengthen the view that memory remains the central bottleneck. Price stabilization, especially alongside higher output, would weaken the revenue trajectory without necessarily reducing shipped capacity.
The second signal is the relationship between component revenue and storage-system sales. Dell, NetApp, HPE, Everpure, Western Digital, and Seagate are already reporting stronger demand. Their next results will show whether AI spending is spreading from servers and memory into broader data infrastructure.
A wider expansion would make this more than a component cycle. It would indicate that enterprises are building the storage, protection, and management layers required for production AI. Slower system growth would suggest hyperscaler purchasing still dominates the market.
The third signal is HBM4 and PCIe 6.0 qualification. Samsung and SK hynix have started shipping or qualifying newer HBM products, while enterprise SSD suppliers are preparing faster interfaces. Volume adoption would reinforce their product-mix advantage.
Qualification delays would create openings for competitors and constrain AI system deliveries. They would also show that fabrication capacity is only part of the supply problem. Packaging, controller design, firmware, and customer validation remain critical.
Samsung AI memory revenue has already established the central fact of 2026: memory suppliers capture a larger share of the AI infrastructure expansion than traditional storage vendors. The unresolved issue is duration.
Readers should watch supplier results as connected indicators, not isolated earnings stories. Compare price changes with shipment growth, system revenue, and qualification milestones. That approach will reveal whether the new hierarchy is becoming structural or approaching another cyclical turn.
