Semiconductor Market Dynamics
Executive Summary: The global semiconductor industry is defined by three core trends: AI-driven structural growth, broad-based price hikes across the supply chain, and widening supply-demand gaps. WSTS, Yole, IDC and other major research houses have all raised their industry forecasts, with global semiconductor output expected to surpass $2 trillion as early as 2027. The memory chip shortage inflection point has been pushed back to 2028, while competition for HBM capacity intensifies. Price increases are now spreading across power semiconductors, MCUs, analog chips and passive components, with cost pressures cascading down the supply chain. Meanwhile, FPGA lead times have extended beyond 52 weeks, server CPUs remain in short supply, and MLCCs have entered a super-cycle—component-level shortages are no longer confined to select categories but are now spreading across the entire industry.
I.Industry Overview: Structural Growth Drives Semi Market Toward $2T
Multiple Institutions Upgrade Industry Forecasts, AI-Driven Structural Expansion Underway
Yole Group's latest report indicates that the global semiconductor device industry is poised for a historic milestone—2026 industry revenue is projected at $1.6 trillion, with the $2 trillion threshold potentially crossed as early as 2027. Unlike previous cyclical recoveries, this expansion is driven by the convergence of four core catalysts: AI computing infrastructure, high-bandwidth memory (HBM), advanced packaging technology advancements, and large-scale global data center investments—collectively signaling that the semiconductor industry has formally entered a new phase of structural growth.
A technology executive from Zeiss Semiconductor Manufacturing Technology further projected that industry output could reach $1 trillion as early as 2026—four years ahead of prior estimates—and potentially climb to $2.5 trillion by 2030.
China IC Exports Jump 96% YoY in H1
Customs data shows China's integrated circuit exports reached $177.28 billion in H1 2026, up 96.1% year-on-year, with export volume of 179.44 billion units. Average unit price rose from $0.54 in the year-ago period to $0.99—an 83% increase. This growth is primarily driven by global memory price increases and spillover effects from the AI industrial chain, with the export structure shifting from "volume expansion" to "value enhancement." It should be noted that export figures include processing trade components—chips imported, packaged, tested, and re-exported—and do not entirely represent value-added from domestic chip production.
II. Memory: Shortages Extend to 2028; HBM Capacity Battle Intensifies
Inflection Point Pushed to 2028 as Supply Tightens
IDC has explicitly stated that supply constraints have pushed the memory market turning point to the second half of 2028 at the earliest, with DRAM supply shortage projected to widen to approximately 13% by Q4 2027. Even with DRAM capacity growing by nearly 10%, over 30% has been allocated to AI products like HBM and SOCAMM, with a significant portion locked in through multi-year long-term agreements. In terms of wafer output efficiency, HBM3E yields are approximately 35%, far below the roughly 1,750 good die per wafer for conventional DRAM; HBM4 yields are expected to drop to around 25%.
Bloomberg and industry stakeholders remain divided on the supply-demand inflection point. Bloomberg warns of potential oversupply by 2028, while industry players insist shortages will persist through 2030. Where both sides converge: significant new capacity will come online in 2027–2028, and the simultaneous shift on both supply and demand sides could trigger a cycle reversal.
It is important to distinguish between slowing price growth and easing supply shortages. Consumer-grade product price stabilization will likely require full capacity release in 2028.
DRAM: HBM Consumes New Capacity, Legacy Supply Stays Tight
Q3 server DRAM contract prices are expected to rise 13%–18% quarter-on-quarter. Micron holds over $100 billion in order backlog, with HBM capacity effectively sold out. Long-term agreement mechanisms constrain suppliers' pricing power with top-tier customers, with price increases increasingly falling on non-LTA clients and incremental purchases.
On the spot market, 64GB server DRAM quotes have surpassed $3,100—approximately 146% above late-June contract prices of $1,380. Historical spot premiums typically range from 10% to 20%. Sharp spot price increases often signal imminent adjustments to long-term contract pricing. Citrini Research projects a global DRAM supply-demand gap of 28.7EB by 2030, representing 18% of total demand, with conventional DRAM emerging as the primary bottleneck and the gap widening from 18% to approximately 25%.
OPPO and Vivo have formally rejected Samsung's Q3 ~20% price increase proposal—Samsung's third price hike in 2026. Cumulative cost increases approaching 300% have pushed downstream tolerance to its limit. Memory chips now account for over 20% of total BOM costs in handsets, up from 10%–15%, and reach as high as 30%–60% in low-tier models. Omdia projects global smartphone shipments will decline 12% in 2026, with average selling prices reaching $565—a record $98 increase over 2025.
The price negotiation between upstream and downstream has entered an intense phase. OPPO and Vivo's rejection signals that memory price increases have reached the tolerance limit of downstream device manufacturers.
NAND Flash: Shortage Expected to Ease in Second Half of 2027
TrendForce estimates the 2026 NAND Flash supply-demand bit gap will range from -4% to -5%, with tightness expected to improve in H2 2027. Server demand now accounts for over 40% of NAND bit demand, surpassing smartphones as the largest application market.
SLC NAND prices are projected to surge 120%–170% in H2 2026. Extreme MLC NAND supply tightness is pushing industrial, automotive, and networking customers to adopt SLC NAND, while AI edge computing and data center demand continue to climb, widening the supply-demand gap.
Samsung has locked approximately 60% of its V-NAND capacity to NVIDIA's CMX requirements. NVIDIA's CMX system—equipped with 576 SSDs per rack and total storage capacity of 9,600TB—is expected to drive CMX-related NAND demand from 35 million TB in 2026 to over 100 million TB in 2027, effectively adding a customer of Apple-like scale to the NAND market.
NAND is transitioning from a cyclical commodity to a critical component of computing infrastructure. NVIDIA alone could consume one-tenth of global NAND capacity, keeping consumer-grade NAND supply under sustained pressure.
HBM: NVIDIA Captures 54% of Global Supply
Global HBM demand in 2026 stands at 32.279 billion Gb, with NVIDIA accounting for approximately 54%—meaning 54 of every 100 HBM chips produced are destined for NVIDIA servers. SK Hynix has secured 60%–70% of NVIDIA's HBM4 supply, with Samsung taking 25%–30%. SK Hynix's entire 2026 capacity is fully sold out, and 2027 capacity has likewise been cleared in advance.
Morgan Stanley models show HBM market size surging from $3 billion in 2023 to $94 billion by 2027—a 128% four-year CAGR. Even assuming all three major suppliers achieve 70% HBM yields by 2027, HBM production of 5.38 trillion Gb will still fall short of demand at 5.61 trillion Gb. SK Hynix CEO has stated unequivocally that 2027 will be the tightest supply year in memory industry history, with the shortage extending beyond 2030.
NVIDIA is not merely "purchasing" HBM; it is effectively determining how global HBM capacity is allocated. AI customers locked in 2027 capacity as early as 2025, while consumer electronics companies only realized in 2026 that they had been pushed down the priority list—this timing gap is the fundamental logic behind price increases.
HBM capacity allocation is the key to understanding the entire memory price surge. Consumer electronics have not entirely lost their opportunity, but in the near term, stacking hardware remains the most direct and effective path to advancing AI capabilities.
Big Three Suppliers Unveil $1T+ Expansion Plans
Samsung, SK Hynix, and Micron have recently announced large-scale expansion plans with combined investment exceeding $1 trillion. Samsung is investing approximately 140 trillion KRW in the Chungcheong region to build an advanced industrial cluster encompassing semiconductors, displays, batteries, and AI components. SK Hynix is investing 100 trillion KRW in Cheongju—80 trillion for its M17 NAND flash fab and 20 trillion for advanced packaging facilities. Micron is investing 1.5 trillion JPY in Hiroshima, Japan, to expand advanced DRAM production lines, with commercial shipments expected around summer 2028.
Following the expansion announcements, SK Hynix shares fell 14.57% in a single day, erasing approximately $160 billion in market capitalization. Historical patterns suggest that major memory supplier expansion announcements often signal an impending peak in the upcycle. The global memory market now faces a supply-side capacity test.
Downstream Pushback: OPPO and Vivo Reject Samsung
Memory chips now account for over 20% of total BOM costs in handsets (up from 10%–15%), reaching 30%–60% in low-tier models. Samsung is negotiating Q3 general-purpose DRAM pricing with customers, targeting a 20% increase, but OPPO and Vivo have formally rejected the proposal. TrendForce forecasts Q3 contract price growth of 13%–18%, below Samsung's 20% target—indicating that downstream resistance is constraining upstream pricing power. Omdia projects global smartphone shipments will decline 12% in 2026, with ASPs reaching a record $565 due to memory price increases.
III. CPU & GPU: AI Reshapes Compute Landscape
Server CPU: Intel Backlog Mounts; China Prices Surge Over 10% Monthly
Intel posted Q2 revenue of $16.13 billion, up 25.4% year-on-year—the strongest quarterly growth since 2011. Data Center and AI segment revenue reached $6.3 billion, up 59%, with operating profit surging from $633 million to $2.474 billion. Management noted this marks the seventh consecutive quarter of results exceeding guidance, with all business segments experiencing demand that consistently outpaces growing supply—the company is effectively selling everything it can produce.
Intel has signed ten server CPU long-term supply agreements, some with fixed pricing and others focused on volume commitments. AMD is similarly active in signing long-term agreements with Chinese customers, typically locking in volume but not price. Server CPU prices in the China market continue to rise, with some products increasing over 10% month-on-month and cumulative increases exceeding 40% since the start of the year. Intel's lead times on certain products have stretched to six months, while AMD's lead times have extended to 8–10 weeks. Bank of America projects the global server CPU market will grow from $43 billion in 2026 to $125 billion by 2030.
Agentic AI applications are redefining CPU's strategic role in the data center. Unlike the early ChatGPT era's 1:8 CPU-to-GPU ratio, agentic AI requires CPUs to orchestrate AI agents and continuously feed data, with the ratio expected to narrow further to 1:2 or even 1:1. Bernstein projects the data center CPU market will grow from $37 billion in 2025 to $223 billion by 2030.
Key Takeaway: CPUs are regaining a central position at the AI table. The rise of agentic AI is elevating CPUs from supporting roles to critical components that determine GPU utilization efficiency—CPUs are transitioning from cyclical commodities to core components of AI compute infrastructure.
NVIDIA Vera Ships, Shaking Up CPU Market
NVIDIA's first fully self-developed server CPU, Vera, began deliveries to customers including OpenAI and Anthropic in June, with estimated revenue of approximately $20 billion for this fiscal year. Vera features NVIDIA's self-developed Olympus core, focusing on single-core speed and memory bandwidth, delivering 50% higher performance than x86 chips for agentic AI workloads. The next-generation Rosa CPU, featuring the Rigel core, is also in development.
NVIDIA's entry into the CPU market reflects the fundamental shift in AI compute architectures. Agentic AI systems require substantial CPU cores for task orchestration, scheduling, tool calling, and sandbox execution—low-latency sequential tasks that GPUs alone cannot handle. Financial markets have already begun pricing in this trend—AMD and Intel shares have surged 128% and 149%, respectively, far outpacing NVIDIA's 8% gain this year.
Key Takeaway: The next phase of AI infrastructure competition has expanded from "who has more GPUs" to "who can support more agentic parallel operations." CPUs are reclaiming center stage, and the competitive landscape is set for a major realignment.
Memory Costs Pass Through: AMD GPUs +10%, Intel CPUs Up Thousands
Memory chip price increases are accelerating their pass-through to CPUs and GPUs. AMD has notified key AIB partners of a ~10% price increase on GPU die and GDDR memory bundles, effective July—the second memory-driven price adjustment in six months. GDDR6 spot prices have surged from approximately $2.5 to $7.5 per GB—a threefold increase.
On the Intel side, Core Ultra 200 Series Plus desktop processors are up $30–$50, while data center Xeon 6 series are seeing per-chip increases of hundreds to thousands of dollars—the Xeon 6980P rising from $12,460 to $13,955, a $1,495 increase per chip. Consumer CPU prices have risen 5%–10% since March, with server CPUs up 10%–20%, and lead times have extended from 1–2 weeks to 8–12 weeks.
IV. Analog & Embedded: Lead Times Stretch, Price Hikes Intensify
Broad-Based Lead-Time Extensions; FPGAs Hit 52 Weeks
Chip delivery lead time extensions have spread from individual suppliers to the entire industry. Analog Devices has notified customers that certain product lead times have extended to six months. STMicroelectronics MCU lead times have stretched to 52 weeks, with distributors already confirming full-year 2027 order requirements. Texas Instruments implemented its third price increase for power management ICs and MOSFETs on July 1, with European suppliers including Infineon and NXP following suit.
FPGA lead times present the most acute challenge, with standard-series lead times now generally exceeding 40 weeks and industrial-grade, automotive-grade, and high-reliability models reaching 52 weeks. Certain high-end AI series models are seeing wait times exceeding one year. China's domestic high-end FPGA self-sufficiency rate remains below 5%. The 52-week lead time means overseas FPGAs can no longer meet domestic demand from industrial control, security, and communications equipment—domestic FPGAs are now filling "emergency orders" rather than simply "replacement orders."
On July 1, nearly 20 leading companies across analog chips, memory, connectors, and passive components simultaneously rolled out new pricing structures, with increases generally ranging from 10% to 30%—the largest wave of chip price increases this year. AI server and data center power management ICs and high-voltage signal chain analog ICs saw the most pronounced increases at 15%–25%, followed by industrial automation and energy storage isolation chips at 10%–15%, while low-end consumer-grade products saw more moderate adjustments.
Key Takeaway: Lead time extensions and price increases have become a systematic industry-wide trend. FPGA lead times exceeding 52 weeks signal that component supply chains have entered a state of comprehensive tightness, with domestic substitution transitioning from strategic choice to urgent necessity.
Suppliers Escalate Price Actions; Microchip, CR Micro Lead
A number of major semiconductor suppliers have issued new price-increase notices for Q3, with increases generally ranging from 10% to 30%. Microchip announced price adjustments on select products effective August 14, covering MCUs and analog chips across all categories. Sino-IC's Q3 product prices are up 15%–25%. Domestic crystal oscillator leader TKD Crystal has raised full-series prices by 10%–30% (65% black tungsten concentrate prices have risen 125% since the start of 2026). China Resources Microelectronics implemented a 15%+ price increase across all categories effective July 1. Jinyu Semiconductor raised prices on diodes, triodes, MOSFETs, and ICs by 10%–30%.
These price increases are driven by three converging factors: soaring raw material costs, tight wafer capacity, and surging AI/new energy demand. Power devices, crystal oscillators, and MCUs are all affected, with component price upward pressure unlikely to ease in the near term.
Japan's three major power semiconductor players—Mitsubishi Electric, Toshiba, and Rohm—are accelerating their business integration talks, targeting a joint venture agreement by September. Based on Omdia data, the combined entity would hold 11.3% of the global power semiconductor market, ranking second only to Infineon's 24.4%.
Key Takeaway: Price increases have evolved from isolated supplier actions to a systematic industry-wide movement. Microchip's price adjustment carries a bellwether effect that may trigger follow-on moves across the sector.
Power Semi Diverges: Low-End Glut, High-End Shortage
The power semiconductor market is experiencing a structural divide—excess low-end capacity coexisting with insufficient high-end effective capacity. AI data centers and grid infrastructure are emerging as new growth engines, complementing the EV market.
AI cluster expansion is dramatically raising requirements for power delivery efficiency and density. Single AI servers require substantially more high-voltage MOSFETs, with some suppliers already implementing price adjustments and AI power component lead times reaching 40 weeks. Infineon projects AI-related revenue of €1.5 billion for fiscal 2026 and €2.5 billion for fiscal 2027 (up from just €700 million in fiscal 2025). Global leaders are making major bets on structural growth: Infineon's €5 billion Dresden Smart Power Fab is already in production, STMicroelectronics' Catania 200mm SiC facility represents a ~€5 billion investment, and onsemi is planning up to $2 billion in Czech operations.
On the domestic front, Sino-IC is planning a ~¥20 billion 12-inch mixed-signal chip line in Shaoxing, while Silan's 8-inch SiC project—a ¥12 billion investment—began trial production in January 2026.
Key Takeaway: The next phase of power semiconductors is framed as capacity expansion, but in reality, it is a strategic positioning battle for critical infrastructure in the electrification wave. The industry faces no shortage of production lines, but rather a shortage of validated capacity that can enter main-drive and AI power systems with proven long-term delivery stability.
V. Passive Components: MLCC Super-Cycle Arrives; AI Reshapes the Landscape
MLCC in Historic Shortage; High-End Models Surge 10x
MLCC is experiencing its "longest shortage in history." Murata's 1206-476 (47μF) model rose from ~¥300 per 2,000 units in May to ¥700–755 in early July. Selected scarce models have seen 3–5x increases since the start of the year, with individual part numbers up over 10x. Mid-to-low-end MLCCs have followed suit, with standard 0402 and 0603 types seeing cumulative increases of 200%–300%.
June BB ratios for the three major suppliers—Murata, Samsung Electro-Mechanics, and Taiyo Yuden—reached 1.30, 1.31, and 1.25 respectively, with Murata's order backlog ratio surpassing the peak seen during the 2018 shortage. Japanese and Korean suppliers are prioritizing high-end specifications (X6S/X7R) for AI equipment, significantly constricting consumer-grade mid-to-high-capacitance MLCC supply, with inventory levels for mainstream part numbers generally below 30 days. High-end MLCC lead times have extended from the historical 8-week average to over 20 weeks, with certain tight models reaching 26–40 weeks.
Samsung Electro-Mechanics implemented a 30% price increase on MLCC products effective August 1. Taiyo Yuden is raising prices again from September 1, with the rare caveat that even with the increase, it cannot guarantee delivery schedules. Samsung Electro-Mechanics has accumulated KRW 750 billion (approximately $510 million) in AI server MLCC long-term agreements this year—exceeding KRW 2.25 trillion ($1.5 billion) when including a KRW 1.5 trillion silicon capacitor LTA signed in May.
AI server MLCC consumption per unit is more than 10 times that of traditional servers—over 20,000 MLCCs per GPU, and up to 600,000 per rack. The substitution trend of MLCCs replacing aluminum electrolytic and tantalum capacitors is accelerating, with 47μF specification usage surging from 1,440 to 10,544 per board—a 632% increase.
Goldman Sachs believes the AI-driven advanced MLCC demand cycle could extend through 2030, far exceeding market expectations of a 2028 peak. This may become the largest and longest prosperity cycle in MLCC industry history.
Tantalum, Aluminum, Inductors Join the Rally
The passive component price rally shows structural divergence. MLCC and tantalum capacitors are the core AI-demand beneficiaries, with Yageo's KEMET holding nearly half of the global tantalum capacitor market share, benefiting from product irreplaceability and strong pricing power.
The aluminum capacitor industry is experiencing collective price increases, with Japan's top three suppliers—Nippon Chemi-Con, Nichicon, and Rubycon—along with domestic leader Jianghai, all implementing increases. Rubycon has further warned that additional increases cannot be ruled out. Aluminum, copper, and tin base metal prices remain elevated, and combined with rising AI server demand, aluminum capacitors are being driven by both cost-push and demand structure upgrade dynamics.
Yageo implemented price increases across its full capacitor product line effective July 1, covering MLCC, aluminum electrolytic, tantalum, film, and supercapacitors—the broadest range in recent years. Morgan Stanley expects Yageo's average product pricing to increase over 30% in H2 2026.
Key Takeaway: Passive component price adjustments are becoming the new normal. MLCC and tantalum capacitors are the core beneficiaries, while resistors and inductors are more cost-driven with limited margin improvement. Industry leaders' revenue and profitability performance warrants continued attention.
Japanese and Korean Suppliers Expand; Local Substitution Accelerates
Kyocera is planning ¥100 billion in MLCC expansion investment for AI servers; Murata is adding ¥80 billion specifically for high-ASP AI products; Taiyo Yuden is raising its mid-term expansion rate from 10% to 15% annually; Samsung Electro-Mechanics is targeting the silicon capacitor segment, recently securing a KRW 1.5 trillion supply contract. Domestic suppliers are actively capturing capacity transfer benefits from Japanese and Korean suppliers: Fenghua Advanced holds monthly capacity of 65 billion units with high-end product revenue share of 35%–40%; Sanhuan Group has monthly capacity of 90 billion units, 70% of which are high-capacitance AI and automotive-grade models; Micron Technology is pursuing a ChiNext IPO with automotive MLCC revenue CAGR of 264.7%.
Manufacturer Updates
Spot Market Insights
eMMC / NAND Flash / NOR Flash
July eMMC and NAND Flash markets exhibited a "firm early, softer late" trajectory, while small-capacity NOR Flash entered consolidation after completing a bottoming-out rebound.
Early in the month, eMMC supply was generally ample with prices range-bound, though Sandisk availability was relatively limited. Following OEM list price increases and reduced factory output, Taiwanese brands gradually adjusted prices upward. Winbond NAND Flash saw robust trading activity, MIXC agents offered sparse quotations, Macronix arrivals were minimal with most orders pushed to next year. Small-capacity NOR Flash saw below-cost inventory largely cleared, Micron quotes remained chaotic, Winbond executed a round of increases, and GigaDevice also raised prices due to factory supply controls and wafer shortages.
Into the second half, eMMC demand weakened notably, with prices showing slight softness across brands amid a lack of large-lot buying momentum. By month-end, eMMC demand remained sluggish while NAND Flash prices broadly stabilized. In small-capacity NOR Flash, Winbond entered consolidation post-hike, Micron sustained high quotes but faced limited takers, Macronix distributor arrivals remained scarce with tight market availability, and GigaDevice supply stayed constrained.
On the industry front, SK Hynix is planning a new NAND flash wafer fab in Cheongju, South Korea—which would mark its first major NAND capacity expansion in the region since the M15 fab came online in 2018, further solidifying its position as the world's second-largest NAND supplier. NOR Flash prices are steadily firming with isolated tight spots; in small-capacity MLC eMMC, demand remains robust with 8GB prices showing no signs of easing, while 32GB/64GB eMMC (SLC NAND) offers compelling value as an alternative and merits evaluation for qualification.
Market Outlook: eMMC is likely to maintain a consolidative-to-soft bias in the near term; NOR Flash supply-side constraints persist, though elevated prices curb buying interest. August is expected to see largely stable pricing.
DRAM
DDR3 & DDR4: Active Early, Broad Consolidation by Month-End
July DDR3/DDR4 markets followed a three-phase trajectory—"active early, divergent mid-month, consolidative late," with prices retreating to stable levels after an early rally.
Early in the month, DDR3 demand softened with prices range-bound. DDR4 4G saw a sharp uptick driven by large-lot buying from Winbond, overtaking Samsung in pricing and pulling Samsung higher in turn, with sustained market turnover. DDR4 16G×8 saw active trading, with SK Hynix leading all brands in steady price gains. DDR5 16G maintained robust trading activity with prices holding at elevated levels.
Mid-month, DDR3 trading picked up with modest price increases. DDR4 4G remained hot, with Samsung playing catch-up to Winbond as prices continued climbing. DDR4 16G×8 and SK Hynix 16G saw increased volumes with steady upward price momentum. DDR5 16G sustained high turnover. Small-capacity SDRAM/DDR prices accelerated upward amid tight industrial-grade supply.
Late in the month through month-end, DDR3 trading flattened, with Samsung prices range-bound and Micron seeing inquiries but limited upside. DDR4 4G/8G prices stabilized in the absence of large-lot buying; 16G saw sporadic inquiries but lacked upward momentum, while industrial-grade pricing held firm. DDR5 16G, despite continued trading volume, stabilized without significant upward movement.
On the industry front, Samsung is leading a new DRAM price-increase cycle, targeting up to 20% for Q3 general-purpose DRAM and over 20% for LPDDR, adopting a notably assertive stance in negotiations. AData's chairman noted that AI applications are driving sustained memory demand growth, with OEM capacity allocation for general-purpose DRAM and consumer NAND expected to tighten further next year. Micron projects DRAM and NAND supply tightness to extend beyond 2027, with improvement not expected until 2028.
Market Outlook: The DDR market has completed a significant rally, with all capacities except industrial-grade now in consolidation. Without a release of end-demand in August, prices may face further downward pressure. Small-capacity SDRAM/DDR supply tightness is unlikely to ease in the near term.
LPDDR: Consolidating with Demand Awaited
The July LPDDR market remained largely range-bound with subdued trading as the market awaited end-demand catalysts.
Early in the month, selective factory allocations emerged; after low-cost inventory was absorbed, prices were expected to recover. SK Hynix saw active trading, while Micron—despite list price increases—recorded sparse spot transactions. Mid-month, LPDDR5/5X inquiries increased with prices holding firm to slightly higher; LPDDR4/4X saw rapid price appreciation as low-cost inventory cleared amid reduced factory output and limited market stocks, with distributors less inclined to build inventory. SK Hynix remained actively traded. Late in the month, LPDDR prices showed limited movement, with server and handset demand yet to materialize and traders adopting a cautious stance.
Samsung plans to raise LPDDR prices by over 20%, signaling a strong pricing stance—though end-customer acceptance remains uncertain.
Market Outlook: LPDDR remains in a "coiled spring" phase. Should Q3 end-demand materialize, prices are likely to see renewed upward momentum—particularly for LPDDR4/4X, where the supply contraction thesis is well-established and offers greater upside elasticity.
Server Memory (DIMM): DDR5 64G Cools, Market Broadly Stabilizes
July server DIMM markets cooled notably from June levels, with the previously hot DDR5 64G segment pulling back into consolidation.
Early in the month, DDR5 64G remained a focal point but with diminished trading activity. From mid-month, DDR5 64G availability increased, prompting price corrections, while 32G/96G/128G held steady with trading activity below June levels. The consolidative pattern extended through month-end amid broadly lackluster demand.
Leading cloud vendors are scaling server demand, driving steady growth in DDR5 RDIMM requirements—though spot-market traction remains limited.
Market Outlook: Server DIMMs are in a digestion phase following the prior rally. DDR5 64G has stabilized after corrections; August is expected to remain consolidative.
Storage
SSD: Subdued Demand with Enterprise Shortages and Consumer Weakness Coexisting
July SSD markets were broadly soft, with enterprise high-capacity shortages coexisting with weak consumer demand, resulting in limited trading activity.
Samsung SSD demand was steady early in the month, with PM893 7.68T seeing trades and PM9A3 1.92T drawing interest. List prices were expected to rise ~20% though not yet formally announced. Mid-month demand moderated, with overseas inquiries concentrated on PM9D3a 7.68T, PM893 series, and high-capacity PM9A3 15.36T. By month-end, inquiries contracted further with overall trading light.
For Solidigm, list prices saw substantial upward adjustments early in the month, with SATA series up 50% and other series rising 20%-30%. Mid-month demand softened, with interest concentrated on P5520 15.36T and S4520 SATA 7.68T. Late-month inquiries shifted to S4520/S4620 SATA series.
With OEM capacity increasingly tilting toward server applications, supply conflicts among NAND resources, enterprise-grade dies, and SSD are intensifying, perpetuating structural shortages. Q3 OEM NAND and DRAM price increases are expected to drive spot SSD prices higher.
Market Outlook: The SSD market is in a post-rally digestion phase. The divergence between tight enterprise high-capacity supply and weak consumer demand persists; Q3 procurement cadence warrants close monitoring.
HDD: High-Capacity Demand Strengthens, Low-Capacity Holds Steady
July HDD markets remained broadly stable, with enterprise high-capacity demand continuing to firm while low-capacity pricing showed modest fluctuations.
Early in the month, demand was soft with channel replenishment occurring in small lots. July list price adjustments of ~30% were phased in gradually according to demand realization. Mid-month, the market remained in a wait-and-see mode, with 16T and below seeing marginal price declines while 20T/24T held firm—though elevated quotations largely remained at the offered level rather than translating into actual transactions. Late in the month, 20T/24T demand increased, with Seagate 24T seeing a wide trading price range, while 4T prices edged up and 8T remained stable.
The three major HDD manufacturers continued price hikes in July: Seagate enterprise up 20%-35%, WD up 25%-32%, and Toshiba up 22%-30%. However, end-demand failed to materialize, prompting modest price pullbacks. Q2 overall demand remained weak with volumes down notably from Q1, and July showed no significant pickup. While AI-related demand has locked in 2026 capacity for the three majors—providing underlying support for enterprise-grade pricing—weak consumer demand and OEM resistance to repeated hikes have kept the market in a high-level consolidation pattern with limited upside.
Market Outlook: Enterprise high-capacity HDD demand continues to strengthen, with OEM price increases gradually transmitting to the market, though OEM resistance to elevated pricing persists. August is expected to see high-capacity prices trending upward while low-capacity remains stable.
CPU
PC CPU: Structural Divergence with Small-Core Models Seeing Activity
July PC CPU markets showed pronounced structural divergence, with low-end small-core and select models seeing active trading, while desktop demand showed some improvement but limited buying commitment.
Early in the month, Intel notebook platform demand continued to soften, with X-series embedded lead times extending beyond 45 weeks. Mid-month, desktop 12th/14th-gen (12500, 14500, etc.) saw increased large-lot demand; 13th-14th-gen HX supply tightened; small-core CPU (N95/N97/N100/N150) buyers remained cautious. Late in the month, N100 and N305 saw active trading, while desktop 14th-gen (14700, 14700K, 14700KF) saw rising interest but limited buying commitment; N95/N97/1135G7 registered K-level transaction volumes.
Market Outlook: Small-core and select models continue to see support, though overall demand remains subdued. August is likely to sustain the structural divergence pattern.
Server CPU: Continued Weakness with Gen 6 Pulled-Down Impacting Pricing
July server CPU markets remained broadly weak with prices trending downward, with Intel and AMD showing divergent trajectories.
On the Intel front, 4th/5th/6th-gen supply was relatively ample early in the month, with a July price increase telegraphed. Through mid-to-late month, weak system demand and component price inflation drove 4th-6th-gen prices lower, with 5th-gen facing intensified price competition due to inventory overhang. By month-end, increased availability of Xeon 6-series pulled-down units drove notable price declines, with further downside anticipated.
On the AMD side, early-month demand improved on value proposition. Following Q3 price adjustments mid-month, most models rose—though select models saw price fragmentation due to pulled-down unit inflows, with an overall stable-to-soft trend. By month-end, 5th-gen pricing held steady while 4th-gen continued to drift lower on weak demand, though factory list prices carry upward expectations, creating upstream-downstream price inversion that compresses channel margins.
Market Outlook: Near-term server CPU price weakness is likely to persist. A Q3 demand recovery would be required for stabilization. Continued inflows of 6-series pulled-down units suggest further downside remains.
Network Interface Cards
Network Interface Cards & RAID Controllers: Light Demand, Scattered Transactions
July Broadcom RAID controller and NIC markets saw generally light demand with scattered transactions and stable pricing.
Early in the month, 9460-8i and 9600-24i recorded small-lot trades, with MCX75310AAS-NEAT (400G CX7) NICs also seeing limited activity. Mid-month, overseas buyers focused on 9560-8i, with limited action elsewhere. Late in the month through month-end, RAID controller demand softened further.
Market Outlook: The NIC/RAID controller market lacks near-term catalysts; August is expected to remain subdued.
GPU
GPU: High-End Tightness Persists, Prices Firm with Upside Bias
July GPU markets saw limited overall demand, though high-end tightness continued to support firm-to-strong pricing.
The RTX Pro 6000 series showed divergent pricing—workstation variants continued their uptrend, while server variants also trended higher on tight supply, with a brief mid-month pullback reversing into renewed strength late in the month. RTX Pro 6000 and other high-end professional graphics cards remain tightly supplied, driven by structural demand from AI model training and visual effects rendering, with spot prices up over 50% and new orders facing 12-18 month lead times. Mid- and low-tier RTX Pro 4000/5000 models enjoy ample channel inventory and stable supply, with prices rising only 8%-12%.
Market Outlook: High-end GPU tightness is unlikely to ease in the near term, supporting further upside; mid- and low-tier segments are expected to remain stable.
