Market Insights and Trends

 

IDC Lifts 2026 Semi Forecast to $1.29T, China Chip Exports Double 

IDC's latest forecast sharply raises total semiconductor revenue for 2026 to $1.29 trillion, a 52.8% year-on-year leap from $842.8 billion in 2025, significantly pulling forward the timeline for the industry to cross the trillion-dollar mark. Memory chips sit at the heart of this shift. DRAM revenue is projected to reach $418.6 billion in 2026, soaring 177% year-on-year, fueled by surging HBM and DDR demand from hyperscale data centers and AI infrastructure providers. NAND flash revenue is expected to hit $174.1 billion, up 138.5% year-on-year. Total memory revenue will jump from $226 billion in 2025 to $594.7 billion in 2026, before climbing further to $790.4 billion in 2027. Non-memory semiconductors are also on a solid growth trajectory, reaching $693.5 billion in 2026.

 

AI infrastructure has become the dominant structural force. In the third quarter of 2025, hyperscale data center capex surpassed $100 billion for the first time. In 2026, combined capital expenditure from the top four players is expected to grow 70% year-on-year to approximately $600 billion. Data center semiconductor revenue will hit $477.1 billion in 2026, with the "smart" data center segment—encompassing CPUs, AI accelerators, GPUs, custom ASICs, and networking chips—valued at $281 billion. By 2030, data center semiconductors will contribute $843.2 billion, accounting for nearly half of the entire semiconductor market. HBM remains the primary bottleneck in the AI accelerator supply chain, with most capacity already locked in through 2026 and forward allocations extending into 2027. Its cost per bit is several times that of standard DRAM, and meaningful new supply will not enter the market until late 2026 at the earliest. Enterprise SSD prices have already risen sharply, as hyperscale customers locking in supply tighten channel availability for the consumer segment.

 

China's chip exports are playing an increasingly significant role, growing at a triple-digit pace. Customs data shows integrated circuit exports reached $31.085 billion in April 2026, up 100.1% year-on-year, with 32.04 billion units shipped in a single month. Cumulative exports from January to April totaled approximately $103.5 billion, an 83.7% increase year-on-year, accounting for 12.2% of the nation's total electromechanical product exports. Monthly growth accelerated progressively, with the January-to-March cumulative figure of $72.47 billion marking a 77.5% rise from $40.83 billion a year earlier. The import side remained equally robust, with April imports hitting $53.87 billion, up 54.8% year-on-year and setting a new monthly record for the second consecutive month. Total electromechanical product trade for the first four months reached $1.24429 trillion, up 23% year-on-year, with exports at $849.04 billion—a 21.1% increase and the 14th straight month of record highs. The trade surplus in this category stood at $453.8 billion, exceeding the overall goods surplus by $106.1 billion. SIA data indicates global semiconductor sales hit $298.5 billion in the first quarter of 2026, up 25% quarter-on-quarter, with March alone reaching $99.5 billion, a 79.2% year-on-year jump. China's March sales rose 74.8% year-on-year. Currently, China holds nearly 30% of global production capacity for mature nodes at 28nm and above. With 70% to 75% of global chip application demand concentrated in mature processes, AI power supply and optical communication companies are purchasing Chinese MCU chips on a large scale. Overall prices for optical communication MCUs have risen roughly 40% year-to-date. The explosion in computing power investment is squeezing mature node capacity while simultaneously opening new incremental opportunities for MCUs.

 

 

Infineon Hikes Prices Twice in One Fiscal Year, TI Logs Fourth Hike in Two Years 

Infineon sent a confidential price adjustment notice to customers on May 27, announcing price increases on select products effective July 1, 2026. This marks the second hike within the same fiscal year, following the previous increase announced in February and implemented on April 1, with less than a single quarter separating the two moves. Unlike the earlier round, which focused on power switches and related chips, Infineon did not disclose specific product lines or the magnitude of the increases. The notice notably emphasized that the "scope" of demand growth has far exceeded earlier projections, signaling that shortages and price pressures may no longer be confined to premium AI-specific components and that general-purpose products are now facing supply constraints as well. Three converging pressures were explicitly cited as drivers: geopolitical tensions raising costs across the entire value chain, including energy, raw materials, logistics, and services; a sharp surge in demand across the product portfolio with a scope well beyond what was anticipated months ago; and the company's significantly accelerated investment in capacity expansion to maintain supply reliability, a cost burden it can no longer absorb internally.

 

AI data center business runs through Infineon's latest pricing actions as the central thread. During the fiscal Q2 2026 earnings call in early May, the company delivered an across-the-board upward revision, upgrading its full-year guidance from "moderate mild growth" to "significant growth," fueled by robust demand for AI data center power solutions and a recovery in automotive orders. CEO Jochen Hanebeck provided key figures: revenue from AI data center applications is set to reach approximately €1.5 billion in the current fiscal year, climbing to around €2.5 billion in the next, nearly doubling over two years with an average annual increment exceeding €1 billion.

 

Texas Instruments, meanwhile, has pushed the analog chip industry into a new "value-based pricing" cycle with four price increases in two years. TI issued a formal price hike letter to partners on May 7, announcing adjusted pricing across all products effective July 1, 2026, applicable to both new orders and shipments. This marks the fourth increase since June 2025 and the second within 2026. The April 1 hike covered select analog and embedded product lines, with distributors revealing increases ranging from 15% to 85%. The two 2025 rounds affected over 3,300 part numbers with average increases exceeding 10% and some reaching 40% to 70%, as well as 60,000 product models with hikes of 10% to 30%. In the first quarter of 2026, TI's data center business revenue grew approximately 90% year-on-year and over 25% sequentially. With power consumption per AI server now exceeding 10kW—five to ten times that of traditional servers—each rack requires dozens of TI power management chips, gate drivers, and signal chain products, as data center architectures transition from 12V to 48V and even 800VDC high-voltage systems. ADI also initiated an across-the-board price increase on February 1, 2026, followed by domestic players such as SG Micro and Novosense from late 2025, while Taiwan-based firms including Silergy, GMT, Anpec, and Fitipower stand to benefit from the pricing uplift.

 

From Infineon to TI to ADI, top-tier IDMs have made concentrated pricing moves within less than half a year, resonating with shortages in passive components and packaging substrates such as MLCCs and FC-BGA. The AI infrastructure buildout is now transmitting cost pressures beyond computing chips themselves to every link of the electronics supply chain. Analysts note that TI's four price hikes in two years mark the analog chip industry's shift from cutthroat volume-for-price competition toward "value normalization." Over the next one to two years, price wars will fully recede, with technological barriers becoming the pricing anchor. Manufacturers possessing automotive-grade certifications, high-reliability isolation technology, and high-precision signal chain capabilities will seize pricing initiative. For downstream customers, the dual pressure of locked capacity and cascading cost pass-through is unlikely to ease in the near term. 

 

 

Power Semi Market Eyes ¥7.33T, Lead Times Hit 30 Weeks, PMICs Surge 

The global power semiconductor market is experiencing a structural shortage driven by the twin engines of AI data centers and new energy vehicles, with price increases spreading from discrete power devices to power management ICs and propelling the broader semiconductor industry toward the trillion-dollar mark. Fuji Keizai forecasts the global power semiconductor market to reach ¥7,349.5 billion by 2035, nearly doubling from ¥3,755 billion in 2025. Silicon-based power semiconductors will account for ¥4,841.8 billion, while next-generation materials including silicon carbide, gallium nitride, and gallium oxide will collectively reach ¥2,507.7 billion. SiC power modules are projected to surge from ¥350.6 billion in 2025 to ¥1,874.9 billion by 2035, by which time roughly 70% of electric vehicles will be equipped with SiC traction inverters. GaN will rise from ¥58.5 billion to ¥316.9 billion, and gallium oxide will emerge from near zero to ¥14.9 billion. TrendForce expects the global SiC power device market to hit $5.33 billion in 2026, with EVs contributing $3.98 billion, while GaN power devices will reach $920 million, up 58% year-on-year.

 

On the supply side, lead times and pricing are tightening in tandem. Power semiconductor lead times at some major IDMs have stretched to 30 weeks, with MOSFETs and IGBTs broadly tightening. TSMC and Samsung cutting eight-inch capacity will keep global output in negative growth through the first half of 2027, while utilization rates at top-ten foundries for eight-inch lines have rebounded from roughly 80% to nearly 90%, with some foundry quotes rising 5% to 20%. On the packaging and testing front, PTI and ChipMOS are running at full capacity with price hikes approaching 30%. Texas Instruments implemented increases of 5% to 85% starting April 1, while Infineon raised prices by up to 25% on the same day with retroactive application to existing orders, as AI data center power switches and ICs are already in shortage. Nearly ten domestic firms have followed suit, with MCU price increases of 15% to 50% at Sino Wealth, a 10% hike on small-signal chips at Silan Microelectronics, and MOSFET increases of 10% to 20% at Jiejie Microelectronics, with IGBT prices expected to be adjusted upward on May 1. In power management ICs, TI, MPS, NXP, and Richtek are expected to adjust pricing in June to July, while Injoinic and Maxic have already led with increases, Anpec forecasts hikes of 0% to 15%, and most categories face adjustments of single-digit percentages or more.

 

Capacity expansion and consolidation among giants are accelerating in parallel. Infineon's AI data center revenue is set to reach approximately €1.5 billion this fiscal year and €2.5 billion in the next, with a serviceable addressable market of €8 to €12 billion by the end of 2030; the company has invested €5 billion in a new Dresden fab scheduled to begin operations in summer 2026. ON Semiconductor is building a $2 billion SiC facility in the Czech Republic and acquiring Qorvo's SiC JFET business for $115 million. STMicroelectronics' €5 billion eight-inch SiC fab in Italy is expected to enter production in 2026. Following integration, Toshiba, Rohm, and Mitsubishi Electric are projected to hold a combined global share of 11.3%, forming the world's second-largest power semiconductor company. Domestically, Times Electric reported semiconductor revenue of ¥5.36 billion, up 30.43%, while Starpower Semiconductor posted record revenue of ¥4.012 billion, an 18.34% increase. Infineon retained the top spot in automotive semiconductors for the sixth consecutive year with a 12.8% share, while Silan Microelectronics and BYD Semiconductor entered the top ten at 3.3% and 3.1%, respectively.

 

 

MCU Prices Jump 50%, 500+ Units Per Rack, Auto Shortage Runs to 2027 

The surging power demands of AI computing chips are propelling MCUs to the hottest corner of the global semiconductor market. Take Nvidia's GB series servers as an example: a single rack contains 18 computing units, each requiring over 10 power modules, with each module integrating 3 MCUs. The power boards alone consume more than 500 MCUs per rack, valued at $1 to $3 each. Based on an estimated 100,000 racks from Nvidia alone, the addressable market for its power MCUs is approximately RMB 450 million, with the overall market easily exceeding RMB 1 billion when adding customers like Google and Amazon. Optical module demand is exploding in parallel: Goldman Sachs projects global 800G optical module shipments to double this year from 2025 levels, while other institutions estimate 1.6T module shipments will leap from the million-unit range to over 20 million units, with each module requiring 1 to 2 MCUs for laser control and signal management.

 

Price increases have rolled out broadly with significant magnitude. Sino Wealth raised MCU and other product prices by 15% to 50% in January, Nations Technologies announced on March 26 that select products would increase by 15% to 20% effective April 7, and Wuhan Xinyuan Semiconductor implemented new pricing across its entire product line starting May 6. In optical communications, increases generally range from 15% to 20%, with certain specialty specifications seeing hikes above 50%; overall prices for China's optical communication MCUs have risen roughly 40% year-to-date. Rising upstream wafer foundry costs, climbing packaging and testing expenses, sharply higher precious metal raw material prices, and AI chips aggressively cannibalizing mature node capacity are together intensifying the structural MCU supply-demand crunch, with major international players including TI, STMicroelectronics, Infineon, and NXP following suit with price increases. Nations Technologies has already achieved volume shipments to top-tier global power management manufacturers, with power monitoring chips priced at $1.5 to $2 each, and has launched dedicated optical module MCUs supporting 400G, 800G, and 1.6T data rates. GigaDevice's GD32E501 series optical communication-specific MCUs, in mass production since early 2020, are already widely deployed in data centers. MPS, the global leader in high-performance power management ICs, posted first-quarter net profit of $193.2 million, up 43.1% year-on-year, with a gross margin of 55.3%; its stock has surged 73.58% year-to-date, pushing its market capitalization to $77.277 billion. Analysts note that the DRAM supply crisis is expected to persist until at least 2027, as AI giants command capacity with high profit margins, continuously squeezing chip supply for automotive and industrial control sectors and raising the specter of a new chip crisis on the horizon. 

 

 

ADI Buys Empower for $1.5B to Crack AI Power Bottleneck 

Analog Devices announced on May 19, 2026, the acquisition of Silicon Valley-based power management startup Empower Semiconductor in an all-cash deal valued at $1.5 billion, with the transaction expected to close in the second half of the year. The move follows ADI's landmark acquisitions of Linear Technology for $14.8 billion in 2016 and Maxim Integrated for $21 billion in 2020, representing another critical chess piece positioned for the AI era, directly targeting the increasingly prominent bottleneck in AI infrastructure power delivery.

Founded in 2014 and headquartered in San Jose, California, Empower's core capabilities center on three areas: integrated voltage regulators, silicon capacitors, and vertical power delivery architecture. Its technology integrates voltage regulation chips directly beneath or within the AI chip package, replacing traditional lateral power routing with vertical delivery. This shortens the power transmission path and reduces system power consumption by approximately 20%. Leveraging 80MHz high-frequency switching technology, it enables ultra-fast dynamic voltage regulation that matches the instantaneous surge in computational load demands of AI chips, while shrinking power solution size by a factor of ten. ADI's CEO stated that Empower's technology will significantly enhance the company's capabilities in AI infrastructure power management, complementing ADI's power management platform, scale advantages, and customer reach. The acquisition reflects a structural shift in AI server power management requirements—higher power density, faster dynamic response, and lower power consumption have become rigid demands. Traditional solutions struggle to meet the hundreds or even thousands of amps of current and microsecond-level load changes in AI chips, making power management, as the heart of AI systems, the next core battleground of competition. 

 

 

 

IDC Sees DRAM Revenue Tripling in 2026, NAND More Than Doubling 

IDC reports that global DRAM revenue is poised to nearly triple to approximately $418.6 billion in 2026, while NAND revenue is set to more than double to $174.1 billion. Micron shares surged 11.06% to close at $640.20, pushing its market capitalization past the $700 billion mark for the first time. Sandisk jumped 11.98% to $1,406.32, with its market cap crossing $200 billion for the first time. Micron has already begun shipping 245TB enterprise SSDs. Omdia projects global semiconductor market revenue to rise 62.7% year-on-year to roughly $1.35 trillion this year, with memory semiconductor revenue revised upward to $670 billion, an increase of $250 billion from prior estimates, representing an annual surge of 200%. DRAM is expected to double from last year, while NAND is forecast to quadruple. On the spot market, DDR4 8Gb has climbed to $16.00, up 23.08% month-on-month and nearly tenfold over the past year. NAND 128Gb averages $24.16, a monthly gain of 36.29%, marking 16 consecutive months of increases. TrendForce predicts Q2 contract prices for commodity DRAM will rise 58-63%, with NAND climbing 70-75%. Combined capital expenditure from the four tech giants is expected to reach $725 billion, far exceeding last year's $410 billion. Samsung reported Q1 semiconductor revenue of KRW 81.7 trillion and operating profit of KRW 53.7 trillion, with the memory business contributing KRW 74.8 trillion at a margin of approximately 70-75%. SK Hynix posted revenue of KRW 52.58 trillion and operating profit of KRW 37.61 trillion, achieving a 72% margin, surpassing Nvidia (65%), TSMC (58%), and Micron (68%).

 

Doubling Memory Market Tops $137B in Q1, CXMT Share Hits 7.7%
In the first quarter of 2026, the global memory market entered a structural upcycle driven by AI. According to CFM Flash Memory data, the combined DRAM and NAND Flash market reached $137.14 billion, up 81.6% sequentially and 245% year-on-year, setting a new quarterly record. On the demand side, AI computing infrastructure buildout and accelerating deployment of Agentic AI applications are driving the phased execution of long-term data center supply agreements, with HBM, server DRAM, and enterprise SSDs claiming priority access to capacity, while consumer and mobile demand face mounting pressure. On the supply side, manufacturers continue shifting their product mix toward high-value AI memory, and meaningful new capacity remains on a long release cycle, widening the near-term supply-demand gap.

 

The DRAM market has been particularly explosive. Global DRAM revenue reached $94.325 billion in Q1, up 81.5% sequentially, while Counterpoint Research put the figure at $97 billion, a staggering 260% year-on-year surge. Massive capacity cannibalization by HBM has driven ASPs sharply higher for both DDR and LPDDR products, with commodity DDR margins now exceeding those of HBM. Counterpoint forecasts DRAM prices to rise another roughly 50% sequentially in Q2 2026, covering both HBM and standard DRAM, with the full-year market growth rate likely to top 300%. In terms of vendor landscape, Samsung retained the top spot with DRAM revenue of $38.214 billion, up 98.4% sequentially, commanding a 40.5% share and further widening its lead over rivals. SK Hynix, constrained by a higher HBM supply mix, posted revenue of $27.925 billion, up 62.1% sequentially, securing second place with a 29.6% share. Micron ranked third with $18.768 billion in revenue and a 19.9% share. The most striking variable this quarter came from CXMT, whose DRAM revenue surged 115.1% sequentially to $7.309 billion, doubling its market share from approximately 3% a year ago to 7.7%—a year-on-year increase exceeding 700%—firmly establishing it as the world's fourth-largest supplier. This growth was fueled by the simultaneous explosion of AI-related DRAM demand across all segments within China, from smartphones to servers. Nanya Technology and Winbond followed with $1.547 billion and $562 million, respectively.

 

 

Micron Says 2027 Capacity Sold Out, Memory Turns Strategic Asset 

Micron CEO Sanjay Mehrotra stated that memory supply is exceptionally tight and will remain so through 2027, with AI-driven demand for DRAM and NAND set to exceed 50% of total industry output. Micron is currently meeting only about 50% to two-thirds of core customer requirements, and meaningful new capacity will not come online until 2028 at the earliest. Samsung further disclosed that customers have already begun placing orders for 2027, with the entire year's capacity now sold out, and expects the DRAM supply-demand gap to widen further in 2027, with relief unlikely before 2028 under the most optimistic scenario. SanDisk has signed five long-term agreements covering more than one-third of its NAND shipments for fiscal 2027, with three of these agreements carrying a minimum combined contract revenue of approximately $42 billion over terms of up to five years; customers have already put down billions of dollars in collateral to lock in supply.

 

The wave of price increases has fully detonated across financial results. Samsung reported Q1 revenue of KRW 133.87 trillion, up 69% year-on-year, and operating profit of KRW 57.23 trillion, surging 756%. Its memory business posted revenue of KRW 74.8 trillion, up 292%, with operating profit soaring nearly 4,800% to KRW 53.7 trillion, pushing margins past 66%. High-end DDR5 module prices have rocketed from roughly $149 in September 2025 to approximately $869 in April 2026, a more than fourfold increase. Goldman Sachs expects DRAM prices to rise 250% to 280% in 2026, with NAND prices gaining 200% to 250%.

 

In the face of this historic shortfall, supply expansion is nowhere near catching up with demand. AI servers consume several times more memory than traditional servers, and combined capital expenditure from the four tech giants is expected to reach $725 billion in 2026, a 77% surge from last year. Micron has launched a total $200 billion expansion plan in the United States, simultaneously advancing fab construction in Boise, Idaho; Syracuse, New York; and Manassas, Virginia. The Boise facility is expected to produce its first wafers by mid-next year, while the New York site is planned for four fabs, with the goal of raising domestic US production from roughly 10% currently to 40%. Samsung and SK Hynix are actively deploying as well, but fab infrastructure alone takes years to build, making it difficult to close the gap quickly in the near term. The industry has entered an AI-driven long-cycle upswing, with memory evolving from a cyclical commodity into the strategic bottleneck that will determine the pace of AI infrastructure buildout. 

 

 

Micron Pours $2B Into US DDR4, Output Quadruples But Shortage Stays 

Micron Technology has begun mass production of DDR4 and LPDDR4 on its 1α process at its Manassas, Virginia fab, with total investment exceeding $2 billion and $275 million in US CHIPS Act subsidies. DDR4 wafer supply from the facility will quadruple, with full-scale production expected by year-end. CEO Sanjay Mehrotra acknowledged that the company is currently meeting only about 50% to two-thirds of core customer demand, with the overall shortage extending beyond 2026 and truly meaningful new capacity not arriving until around 2028.

 

Yet the consumer spot market has not cooled in response. This week, spot prices for industry-standard DDR4 8GB and 16GB memory modules fell 12.50% and 8.82%, respectively, bringing cumulative declines since mid-to-late March to nearly 40%. The key lies in the nature of this expansion, which is not a net addition. TrendForce notes that some equipment for the new Manassas line was relocated from Micron's OMT facility, where 1α capacity will be phased down quarter by quarter and fully refocused on DDR5 and HBM. Manassas will exclusively serve long-lifecycle segments such as automotive, defense, aerospace, and industrial medical applications, meaning there is no net capacity increase planned for DDR4 and LPDDR4 overall. TrendForce estimates Micron's DDR4 will shrink to roughly 7% of its total DRAM output in 2026 and decline further each year, while industry-wide DDR4 will remain in shortage with prices on an upward trajectory, supported by networking and communications demand.

 

The three major memory makers are collectively exiting DDR4 supply, causing automotive and industrial buyers' buffer inventories to shrink dramatically from the normal 31-plus weeks to just 6 to 8 weeks. S&P Global Mobility forecasts global automotive DRAM contract prices will surge 70% to 100% in 2026 compared to 2025 and warns that supply of older-generation automotive DRAM will be depleted by the end of 2028. DDR4 die of the same density have at times commanded a premium of roughly 40% over DDR5, with gross margins temporarily exceeding those of certain high-end HBM products. It is precisely this profit signal, combined with US manufacturing reshoring policy, that prompted Micron to recommit over $2 billion to DDR4.

 

The consumer market has shown no reaction to the expansion, primarily because upstream supply of DDR4 from manufacturers to module makers has long been largely stagnant. Even when individual buyers secure very small quantities at unit prices above $2, the cost far exceeds the threshold that PC DRAM finished products can bear. Module makers are increasingly turning to sub-tier resource solutions, while since the second quarter, industry customers have been extremely cautious in building inventory and low-price competition among vendors has intensified, leaving the market mired in a chaotic mix of shipment difficulties and price undercutting. Sanjay Mehrotra stressed that fab infrastructure and civil engineering cycles are the longest and most challenging, with building structures alone taking several years. Analysts conclude that the memory chip supply-demand tightness will persist for one to two more years, and the price divergence across different tiers of the market will continue to widen throughout this super cycle. 

 

 

 

NAND Surges 63% in a Month, Enterprise-Consumer Split Widens 

The global NAND flash market in 2026 is witnessing an extreme divergence between enterprise and consumer segments. TrendForce data shows NAND Flash contract prices are expected to rise another 70% to 75% sequentially in the second quarter, with meaningful new capacity unlikely to arrive at scale until late 2027 or even 2028. South Korean customs data indicates that DRAM and NAND prices surged between 165% and 500% year-on-year from April to May 10, with NAND alone soaring 63.1% month-on-month. The cumulative Q2 increase is projected at 70% to 75%, while HBM rose 18.7% sequentially and 165.5% year-on-year, and bare DRAM chips posted sequential gains exceeding 20%. ADATA Chairman Chen Li-bai further expects NAND price increases to outpace DRAM over the next three months, with the bull market extending into 2027.

 

Manufacturer financials have exploded across the board, with both volume and pricing clearly on the rise. SanDisk posted Q3 revenue of $5.95 billion, surging 251% year-on-year. SK Hynix reported revenue of KRW 52.58 trillion, up 198% year-on-year, with NAND average selling prices rising 70% sequentially. Micron's NAND revenue hit a record $5 billion, up 82% year-on-year, with price increases ranging from 70% to 79%.

 

A stark contrast is unfolding between a red-hot enterprise market and a tepid consumer market. Global NAND revenue reached $44.447 billion in Q1, surging 82.9% sequentially, driven primarily by sharply higher unit prices. Robust AI and data center demand has kept enterprise SSD order books full, with leading cloud service providers continuously locking in capacity, making this segment the core growth engine. On the consumer side, however, better-than-expected shipments largely reflect early channel inventory building rather than a genuine recovery in end demand. Weakness in PC and mid-to-low-end smartphone demand is increasingly evident, with consumer TLC SSD spot prices already down 30% to 40%, as storage cost pressure gradually passes through to end devices. The premium smartphone segment has shown relative resilience, with stable demand for high-capacity, high-performance storage in flagship models effectively offsetting softness in the mid-to-low-end market.

 

On the pricing trajectory, NAND continues its upward trend in 2026 but with gradually narrowing gains, following a sharp-rise-then-gentle-climb pattern. Increases vary markedly across process nodes: traditional products such as SLC and MLC, facing contracted supply and rigid demand, are seeing month-on-month gains approaching 50%, while mainstream TLC and QLC products are posting more moderate increases in the 15% to 20% range. Mobile storage products including eMMC, eMCP, UFS, and uMCP are rising in tandem, covering the full range of handsets, tablets, and other end devices.

 

The supply side is unfolding with a differentiated expansion landscape. Leading Chinese manufacturers continue to increase capital expenditure and push capacity expansion, focusing on mainstream process nodes to grow global market share. Among overseas players, Solidigm is concentrating on high-density NAND capacity and ramping advanced node investment to meet data center demand for high-capacity storage, with industry resources tilting toward high-density, high-margin products. The near-term supply-demand tight balance remains intact. Looking further out, full-year revenue will climb quarter by quarter, likely peaking in Q1 2027, after which gradually easing supply and moderating marginal demand will bring a mild price correction and a gentle revenue decline. Certain mobile storage categories will see modest price pullbacks in Q2 2027 but remain at elevated levels overall. The industry has entered a new phase of AI-driven structural growth, where manufacturers with high-density capacity and advanced technology advantages will command market leadership. 

 

 

Buyers Lock Three-Year Deals, Taiwan Memory Inventory Tops NT$100B 

AI momentum continues to build, and the memory industry is undergoing a structural supply-demand reversal with the shortage gap stretching all the way to 2028. Capital expenditure from the four major US cloud service providers has already reached $700 billion this year and could climb further to $800 billion to $1 trillion in 2027, with prepayments already made to manufacturers to lock in capacity for the next two to three years. Suppliers such as Samsung posted single-quarter price increases of as much as 40% in Q2, underscoring the steep upward trajectory.

 

The three major memory makers have completely sold out their DRAM capacity for this year, and negotiations for next year's allocation are already underway, with some customers even moving to secure three- to five-year long-term agreements. ADATA Chairman Chen Li-bai disclosed that in response to robust demand, the company has begun selective order-taking, allocating orders based on product mix, profitability, and long-term customer relationships. Nanya Technology President Lee Pei-ing also noted that the explosion of AI applications has created rigid demand, and the company is facing a situation where customers are frantically competing for capacity.

 

Silicon Motion Technology pointed out that the expansion cycle for the three major manufacturers spans two to three years, meaning meaningful new capacity is unlikely to materialize before 2028. With NAND bit growth at only around 20% while AI demand grows exponentially, the supply-demand gap will persist until at least 2028. Chen Li-bai analyzed that with no significant capacity expansion on the supply side, DRAM and NAND shortages will remain unchanged this year and next, making price increases a foregone conclusion, with only the magnitude and speed differing. Given NAND's lower price base and demand simultaneously driven by enterprise SSDs, AI servers, autonomous vehicles, and robotics, NAND price gains could outpace DRAM over the next three months. On the DRAM side, DDR4 and DDR5 remain in a rotational price rally, with recently resurgent DDR4 demand pulling prices higher amid persistently tight supply. ADATA expects quarter-over-quarter growth throughout the year, with Q2 revenue reaching NT$36 billion and the first-half total hitting NT$62 billion. If demand holds in the second half, full-year 2026 revenue could surpass NT$120 billion. The company currently holds approximately NT$40 billion in inventory and aims to raise that to NT$50 billion by the end of Q3, with elevated inventory providing support for revenue expansion in the second half.

 

At Nanya Technology, President Lee Pei-ing believes the memory supply shortage will extend through the end of 2027, with supply remaining tight into 2028. Customers are actively seeking long-term agreements to lock in pricing and supply, with the longest terms now reaching three years. Chairman TS Tsou noted that the memory industry is shifting from cyclical swings to structural growth. Pilot products on the 1C process are expected to complete verification in the second half of this year, with 16Gb DDR5 validation to be finished this year and LPDDR5 pilot production to advance. As the new 5A fab gradually ramps up, monthly capacity is projected to reach 20,000 wafers by the first half of 2028.

 

On the fundraising front, Taiwan's two memory heavyweights have raised over NT$100 billion from overseas markets this year, marking the largest capital-raising effort in the history of Taiwan's memory industry. Nanya Technology recently conducted a private placement of 350 million shares, bringing in funds from four giants spanning Japan, South Korea, and the US—Kioxia, SanDisk, SK Hynix, and Cisco—raising a combined NT$78.7 billion. Phison recently priced its first overseas unsecured convertible bond, raising $800 million. Phison CEO K.S. Pua stated the company is transforming from a traditional NAND controller IC and storage solution provider into an AI storage infrastructure and edge AI computing platform player, with the fundraising explicitly earmarked for foreign currency component procurement. With robust demand for AI SSDs, enterprise storage, and high-end NAND, the company is building a sufficient war chest in advance to meet subsequent procurement and shipment needs. 

 

 

NAND Makers' Q1 Revenue Soars 83.7% to Record $38.9B 

The global NAND Flash market experienced an explosive first quarter in 2026, driven by simultaneous volume and price increases. According to TrendForce's latest survey, combined revenue of the top five brands jumped 83.7% sequentially, shattering the $38.9 billion mark. The core driver was the accelerating buildout of AI server infrastructure, with enterprise SSD demand growing exponentially. This was compounded by a structural shortage of traditional HDDs, which forced a wave of orders to shift toward QLC enterprise SSDs. With demand surging and supply virtually frozen, average selling prices across manufacturers generally exceeded expectations, and bargaining power remained firmly in the hands of suppliers.

 

The rankings and growth rates of the five major players paint a picture of an across-the-board breakout. Samsung Electronics held the top spot with $13.51 billion in revenue, up 104.7% sequentially—the fastest growth among the top five—lifting its market share from 28% to 31.6%. SK Hynix Group ranked second with approximately $7.53 billion in revenue, up 44.6% sequentially, with its share adjusting to 17.6%. Its subsidiary Solidigm benefited from a continued surge in high-capacity QLC enterprise SSD orders, contributing a significant incremental boost. The battle for third place was nearly a dead heat: Kioxia posted revenue of $5.96 billion, up 80% sequentially, while Micron and SanDisk each recorded $5.95 billion, both surging 96.7% sequentially, with all three locking in a 13.9% market share. SanDisk's data center business saw quarterly revenue growth exceed 200%.

 

Heading into the second quarter, the supply-demand imbalance shows no signs of easing. Smartphone and PC demand is being dampened by rising memory prices and higher device costs, but server-side orders are more than sufficient to fill the gap. Manufacturers generally expect shipment volumes to continue growing in Q2, with pricing mechanisms sustaining ASPs at elevated levels. TrendForce explicitly notes that major NAND Flash manufacturers have virtually no new capacity coming online in 2026, and with AI demand remaining robust, the entire year will be characterized by supply shortages. Products with more than 200 layers will firmly establish mainstream status by year-end, with capacity resources continuing to tilt toward server applications.

 

Yet even after prices have soared more than twentyfold, manufacturers remain almost indifferent to large-scale capacity expansion. The head of SanDisk's flash division pointed out that the cost-per-gigabyte gap between NAND Flash and HDDs has widened to roughly 20 to 25 times, far exceeding the 2 to 3 times conversion threshold that major cloud service providers set based on total cost of ownership. Some data center operators are now re-evaluating their storage architectures, and a portion of demand may flow back from NAND Flash to HDDs. SanDisk, Seagate, and Western Digital have all secured three- to five-year long-term agreements. On the supply side, manufacturers remain extremely conservative after five to nine consecutive years of losses. The typical cycle from capital expenditure to capacity output takes roughly 15 months, meaning prices must continue rising and remain elevated for several quarters before suppliers gain sufficient confidence to expand aggressively. For now, manufacturers prefer migrating existing capacity from older nodes to advanced nodes rather than increasing total output. Supply elasticity will remain severely limited for the foreseeable future, even as the structural demand increment from AI servers continues to accelerate. 

 

 

Samsung, Kioxia, Micron Quit 2D NAND, MLC Prices Triple  

Samsung, Kioxia, and Micron have successively shut down their 2D NAND production lines. Samsung ceased production of 80,000 to 100,000 wafers per month at its Hwaseong fab in March, with final shipments of MLC products scheduled for June. Kioxia will accept last orders through the end of September and fully exit by 2029. TrendForce forecasts global MLC NAND capacity will drop 41.7% this year. MLC NAND prices rose 20-30% in April alone, with 64Gb products climbing from $6-7 at the end of 2025 to $19-20, a surge of 200-300%. SLC NAND 2Gb has risen to $4-4.5, up 120-150%. Macronix and Winbond together now command approximately 59% of the 2D NAND market. Macronix posted April revenue of NT$5.91 billion (approximately $188 million), up 33.7% month-on-month and soaring 153.7% year-on-year. Winbond expects its SLC NAND bit growth to exceed 80% next year, with its 24nm process contributing more starting from Q2. 

 

 

 

CPU Shortage Hits Dozens of Models, Lead Times at 4 Months, Q3 Hikes Loom 

The global semiconductor supply chain shortage has now spread fully from memory chips to CPUs, with the processor returning to the core of AI infrastructure in commanding fashion. Acer Chairman Jason Chen stated publicly that CPUs have become the most critical bottleneck in the current supply chain—memory can be downgraded to keep shipments moving, but once CPUs are in shortage, end devices simply cannot be delivered. In terms of specific shortages, roughly twenty to thirty models of Intel's fifth and sixth-generation commercial CPUs are out of stock, with lead times stretched to two to four months. AMD is seeing shortages across ten to fifteen models spanning its fourth and fifth-generation lines, with wait times of two to three months. Major PC and server manufacturers are urgently engaging with CPU suppliers to secure capacity.

 

The wave of price increases has rolled out broadly and more is expected. Since February this year, Intel and AMD have successively raised prices across their entire server CPU lineups, with overall increases of 10% to 15%. Certain high-end AI server CPUs have seen even larger spot premiums, while consumer-grade CPUs have risen 5% to 10% in tandem. Supply chain sources indicate the two companies may introduce another round of price hikes in the third quarter. From Infineon to passive components to FC-BGA substrates, the addition of CPUs to the shortage list means AI infrastructure cost pressure has now rippled through every link of the electronics supply chain.

 

The essence of this shortage is a structural demand explosion triggered by a shift in AI technology paradigms. As AI inference and agents enter the stage of large-scale deployment, the center of computing gravity is moving from compute-intensive training toward autonomous planning and multi-task orchestration. The CPU-to-GPU ratio in data centers is rapidly advancing from approximately 1-to-4, seen between 2022 and 2025, toward 1-to-1. Industry data shows that training requires only one CPU for every seven GPUs, while inference narrows the ratio to three or four to one. With the rollout of agentic AI, the ratio could even reach 1-to-1, fully exposing the CPU bottleneck. Bank of America expects the data center CPU market to more than double from $27 billion in 2025 to $60 billion by 2030. AMD CEO Lisa Su forecasts the CPU market will grow at an annual rate exceeding 35% over the next five years, and the company has formally raised its long-term server CPU total addressable market CAGR to over 35%, projecting the market to surpass $120 billion by 2030, significantly higher than its previous estimate of 18%.

 

Capital markets are already repricing ahead of the curve. On April 23, 2026, Intel CEO Lip-Bu Tan explicitly positioned the CPU as the orchestration layer and critical control platform for AI systems. Year-to-date, AMD and Intel shares have each gained roughly 25%, Micron has surged over 37%, and Corning has risen about 18%, with all four companies having more than doubled their market capitalizations. Intel leads the pack with gains exceeding 200%. ARM posted fiscal Q4 2026 revenue of $1.49 billion, up 20.2% year-on-year, as it accelerates its pivot from the mobile market toward AI infrastructure. AI data center CPU orders doubled to $2 billion within just five weeks, with major customers including Meta, OpenAI, Cerebras, and SK Telecom. Its data center and networking business has grown revenue 63% over the past year and now accounts for roughly 15% of the total. ARM is confident of reaching $15 billion in revenue by fiscal 2031, more than triple the $4.92 billion recorded this March. As of the second quarter of 2025, x86 architecture held a 75% share of the data center CPU market, with ARM at roughly 25%. Analysts note that the core question this cycle is not "who sells more," but rather "who can most effectively convert CPU demand growth into profit growth of their own." The fact that three CPU suppliers with entirely different architectures have simultaneously delivered results above expectations confirms that behind the shortages and price increases lies a systemic outcome of the comprehensive upgrade of AI infrastructure. On the PC side, unit shipments may decline roughly 11% this year, but higher average selling prices mean total output value could still surpass last year. AI-driven data center CPU demand will play an increasingly critical role as the growth engine. 

 

Nvidia CPU Business Eyes $20B, Taps $200B New Market 

Nvidia is extending its influence from GPUs into the CPU domain. Colette Kress, the company's CFO, announced during the earnings call that CPUs represent a $200 billion addressable market the company has never previously participated in, and that it expects to generate nearly $20 billion in standalone CPU revenue this year, with all major hyperscale data center operators and system manufacturers already collaborating to deploy these chips. Jensen Huang confirmed in Taipei that this $200 billion market forecast includes China, noting that the region holds enormous long-term demand. The Vera CPU integrates 88 custom Olympus cores, uses LPDDR5X memory, supports up to 1.2TB/s of memory bandwidth, and delivers up to 50% higher performance and up to 2x better energy efficiency compared to comparable rack-level CPUs under agent sandbox workloads. A fully loaded rack of 256 CPUs can support over 22,500 concurrent agent instances. The company has already shipped initial samples to customers including SpaceX, OpenAI, Anthropic, and Oracle Cloud Infrastructure, with Oracle planning to deploy hundreds of thousands of Vera CPUs starting this year.

 

Based on pricing and demand estimates, Grace CPUs are priced at roughly $3,000 to $5,000 per unit, while Vera, as its successor optimized for AI agents, is expected to carry an average selling price of $5,000 to $8,000. At an $8,000 ASP, $20 billion in revenue would translate to 2.5 million CPU units shipped. Each Vera CPU is equipped with eight SoCAMM slots, and at 192GB per module, capacity per CPU reaches 1,536GB. This would drive SoCAMM demand for Vera CPUs to 38.4 billion GB in fiscal 2027. With the three major DRAM manufacturers' CY2026 SoCAMM supply estimated at roughly 30 billion Gb, and considering the previously estimated 44 billion Gb of SoCAMM demand from 100,000 VR N72 server shipments in CY2027, the total CY2027 SoCAMM addressable market could exceed 80 billion Gb—nearly equivalent to the entire annual LPDDR5 market for smartphones. The shortage of LPDDR5, and DRAM more broadly, is likely to intensify further over time.

 

Jensen Huang stated that the world is rebuilding computing capacity for agentic AI and robotic physical AI, with Nvidia at the center of this transformation. AI agents require CPUs to execute a wide range of operations, and Vera was designed from the ground up as an intelligent CPU. The CPU market boom has already propelled Intel shares up 222% year-to-date, though the company still faces resource shortages with demand far outstripping supply. AMD's latest earnings report also highlighted increased CPU demand, with EPYC CPUs and Instinct GPUs driving a 57% year-on-year revenue increase to $5.8 billion. 

 

 

 

Nvidia Revenue Hits $81.6B, Backlog Breaks $1 Trillion 

Nvidia reported revenue of $81.6 billion for the first quarter of fiscal 2027, up 85% year-on-year and exceeding the market consensus of $78.9 billion. Net profit surged 211% year-on-year to $58.3 billion, while free cash flow hit a record $48.6 billion, nearly doubling from $26.1 billion a year earlier. Gross margin held steady at 74.9%, and operating margin climbed to 65.6%. Even more striking, Jensen Huang disclosed that the company's undelivered order backlog has surpassed $1 trillion and continues to grow—a figure that exceeds the entire global semiconductor market of $795.6 billion in 2025. That a single company's backlog outstrips the annual sales of the entire industry is difficult to fathom by conventional measures, underscoring Nvidia's overwhelming dominance in AI chips.

 

The data center business remains the absolute growth engine. The segment contributed $75.2 billion in revenue for the quarter, up 92% year-on-year, accounting for over 92% of total revenue. Within this, compute hardware revenue reached $60.4 billion, up 77%, while networking revenue hit $14.8 billion, nearly tripling. Under the new disclosure format, hyperscale cloud service providers contributed $38 billion, roughly half of data center revenue, while AI cloud, enterprise, and industrial markets generated $37 billion, up 31% sequentially, with AI cloud revenue more than tripling year-on-year. The gaming business has been folded into the edge computing segment, which posted total revenue of $6.4 billion, up 29% year-on-year. Jensen Huang noted that AI factory buildout represents the largest infrastructure expansion in human history, with AI inference token generation surging tenfold in a single year. Microsoft processed over 100 trillion tokens in the first quarter alone, a fivefold year-on-year surge, and adoption of the B200 chip has boosted inference token generation rates by three times. Cloud H100 rental prices have already risen 20% year-to-date, with A100 up nearly 15%, as the compute supply-demand imbalance continues to intensify.

 

Capacity bottlenecks and product roadmap advancements are progressing in parallel. TSMC's CoWoS advanced packaging and HBM high-bandwidth memory remain the two core constraints. DRAM average unit prices saw significant increases in January and February 2026, driven primarily by the three major memory makers prioritizing HBM3E output, with HBM3E commanding a per-bit price five to ten times that of standard DRAM. The Blackwell series uses 192GB of HBM3E, and capacity is gradually being released as supply ramps up. The next-generation Vera Rubin is scheduled to begin shipping around July 2026, delivering a substantial improvement in inference capability over Blackwell. Equipped with HBM4 chips and featuring data I/O pins increasing from 1,024 to 2,048, it will present even greater advanced packaging challenges and remain in short supply throughout its entire lifecycle. Nvidia is simultaneously entering the server CPU market, with the Vera CPU purpose-built for agentic AI and opening a $200 billion new market, with total CPU revenue this year expected to reach nearly $20 billion.

 

Revenue guidance for the second fiscal quarter stands at $91 billion, ahead of the $86.8 billion market consensus, with gross margin expected to remain around 75%. The guidance, however, excludes data center compute revenue from China, a category that generated $4.6 billion in the same quarter a year earlier. Combined AI capital expenditure from Google, Microsoft, Amazon, and Meta is expected to exceed $700 billion this year, a sharp increase from roughly $400 billion in 2025. North America, the Middle East, and Europe all posted net growth, with the sales decline in China fully offset by gains in other regions.  

 

 

NA CSP Capex Set to Top $770B, AI Inference Compute to Leap 122% 

The five largest North American cloud service providers are dramatically scaling up AI deployment. According to TrendForce's latest research, combined capital expenditure from Google, Amazon, Microsoft, Meta, and Oracle will surpass $770 billion in 2026, a year-on-year increase of nearly 87%, and is expected to account for over 60% of global demand for Nvidia GB and VR systems. This will drive total AI training compute across the five providers up by more than 56% year-on-year, while total AI inference compute will surge by approximately 122%. Measured in FP16/BF16, the five CSPs' combined training compute already exceeded 9 ExaFLOPS in 2025 and is projected to grow over 56% in 2026. Measured in FP4/NVFP4, their total inference compute surpassed 37 ExaFLOPS in 2025 and is set to leap roughly 122% in 2026—a growth rate significantly outpacing that of training, reflecting Nvidia's particular emphasis on inference performance in its next-generation GB300 and VR200 rack-scale solutions.

 

AI server shipments are forecast to grow more than 28% year-on-year in 2026, with high-end AI training models accounting for roughly 55%, though inference models are expected to dominate over the medium to long term. CSPs are simultaneously advancing their own custom ASICs. Google's TPU demand is projected to grow nearly 80% year-on-year, with the v7 generation upgrading to v8 in the second half of the year. Amazon's Trainium series will account for over 40% of its in-house AI server deployments. Nvidia, AMD, and the new generation of custom ASIC racks have all integrated liquid cooling. With thermal design power per chip continuing to climb, AI server system power consumption is undergoing structural amplification. The combined annual incremental power consumption of servers from the five CSPs is set to leap from 2.8GW in 2023 to 18GW in 2026, with a year-on-year growth rate of 116% from 2025 to 2026. 

 

 

 

Murata Adds ¥80B for MLCC Expansion, High-Voltage Lead Times Hit 18 Weeks 

Murata Manufacturing has announced an additional investment of approximately ¥80 billion over the next two years to expand MLCC production, squarely targeting the explosive demand from data centers and AI servers. The plan will be rolled out in two phases: roughly ¥40 billion in fiscal 2026 and another ¥40 billion in fiscal 2027, with funds primarily directed toward its Izumo City plant in Shimane Prefecture and existing MLCC production bases. The expansion is expected to boost overall capacity by 10% to 15% from current levels. On April 3, 2026, a new production building at Murata's Izumo subsidiary was completed, providing the physical infrastructure for the ramp-up.

 

The MLCC pull from AI servers continues to surge dramatically. A standard server requires roughly 2,000 to 4,000 MLCCs, while an Nvidia A100-class AI server needs approximately 48,000 units—more than ten times the amount. The latest GB200 or GB300 platforms consume around 30,000 units per server, with a single AI rack requiring as many as 440,000 MLCCs. Murata, the undisputed global leader with over 40% market share, commands an even more dominant 70% share in AI server MLCC supply. This expansion focuses on miniaturized, high-capacitance products, with the company already rolling out large-capacitance models such as 47μF in 0402 size and 100μF in 1608M size, capable of withstanding high temperatures and designed for placement directly adjacent to chips.

Murata's capacitor business already accounts for over 51% of total company sales, making it the fastest-growing segment. For fiscal 2026, the company projects total revenue of ¥1,960 billion, up 7.1% year-on-year, with operating profit of ¥380 billion, a sharp 34.8% increase from the prior period. Currently, mainstream MLCC manufacturers including Murata and Samsung Electro-Mechanics are all running at full capacity, with lead times for high-end products stretching to 24 weeks.

 

The high-voltage MLCC market is tightening in tandem. Delta Electronics has already absorbed all high-voltage MLCC capacity from Taiwan-based manufacturers, and Walsin Technology's subsidiary Prosperity Dielectrics has signaled that lead times will extend to 16 to 18 weeks. As data centers adopt 800V HVDC power architectures and Power Shelves advance from 3kW to 12kW and now the latest 110kW, high-voltage MLCCs have vaulted from a niche segment to a core component. The industry expects demand for high-voltage MLCCs driven by AI server power conversion to far outstrip supply, and with lead times stretching, a wave of price increases may be brewing. Murata's expansion will partially ease the supply tightness, but with AI server demand projected to grow 87% and 88% in 2026 and 2027 respectively, whether the MLCC market can achieve supply-demand balance remains an open question. 

 

 

MLCC Per Server Tops 30K Units, High-End Prices Surge 65%, Shortages Spread 

AI servers are propelling the MLCC market into a prolonged upcycle driven by structural shortages. A single Nvidia GB300 platform AI server now requires roughly 30,000 MLCCs—30 times that of a typical smartphone and three times that of a traditional fuel-powered car—with a single full rack consuming as many as 440,000 units. When the next-generation Vera Rubin platform ramps up, the number of MLCCs installed per chip is expected to increase by another 50%, with installation value rising approximately 60%. The GB200 NVL72 rack already demands over 50,000 high-capacitance MLCCs, with the total value per rack leaping to five to ten times that of conventional servers. MLCC count per board has also doubled from 6,500 to 12,000 units. Nvidia's B200 chip features power consumption of 1kW and operating current of 1,200A, with GPU nanosecond-level transient currents potentially surging from 200A to 1,000A, making MLCCs irreplaceable as the core component for instantaneous power delivery. By 2030, global server MLCC shipments are projected to exceed 400 billion units, rising from 2% of total shipments in 2026 to 5-10%, with high-end products commanding unit prices three to five times or more those of standard products and accounting for 20-30% of total market value, emerging as the single largest growth engine for the MLCC market. AI server-related MLCC demand is expected to maintain a 30% compound annual growth rate from 2025 to 2030, with demand volume by 2030 reaching 3.3 times that of 2025.

 

Price increases are spreading from high-end specialty models to general-purpose products across the board. KEMET, holding over 40% market share, raised prices effective June 1. Panasonic will hike SP-Cap prices by 20-30% starting July 1, following earlier tantalum capacitor increases of 15-30%, with overall capacitor price hikes ranging from 5% to 65%. MLCC spot prices began climbing in late December 2025, and by February 2026, spot prices at mainland Chinese distributors had risen up to 20%, with Samsung Electro-Mechanics passive component spot prices moving up in tandem and its stock price surging 35.31% between February 19 and 23. Taiyo Yuden was the first to issue a formal price hike letter on April 14, and Murata initiated its first large-scale price adjustment in nearly three years in early May, with AI server and high-end automotive-grade MLCCs seeing increases of 15-35%. By late May, general-purpose 226 and 106 series were completely unavailable, and Murata's lead times stretched from 8 to 12 weeks. Some customers, worried about worsening supply, have begun placing excessive advance orders, and a trend of double booking is emerging, which may further intensify upward pressure on general-purpose MLCC pricing.

 

Supply tightness and capacity squeeze are intensifying in tandem. Murata is running at 90-95% utilization, with order volume double its existing capacity. Samsung Electro-Mechanics plans to cut standard product capacity by 30-35% in 2026 and redirect it toward high-end lines. The expansion cycle for high-end MLCCs stretches 18 to 24 months, and global supply in 2026 is expected to grow only 17%. AI-dedicated capacity expansion is crowding out general-purpose models, with long-term supply agreements now pivoting from price negotiation to priority volume assurance, compressing any room for price declines. On the raw materials front, silver and copper prices continue to climb, with raw material costs accounting for over 80% of MLCC production. London spot silver hit multi-year highs in February 2026, while full-year copper prices have risen over 27%. The high-end supply-demand tightness is expected to persist through 2026 to 2027. The global MLCC market is projected to reach RMB 105.043 billion in 2025 and rise to RMB 132.62 billion by 2029. FC-BGA substrates are also in critical shortage, with AI substrate supply next year expected to be even tighter than this year. KB Securities has set target prices of KRW 2.2 million for Samsung Electro-Mechanics and KRW 1.6 million for LG Innotek, with both companies poised to join the "trillion won operating profit club" together for the first time since 2022.

 

Japanese and Korean manufacturers dominate roughly 80% of the global high-end MLCC market, while domestic Chinese players hold a combined share of about 12%, with their incremental growth still concentrated in consumer electronics. The supply shortage is accelerating the window for domestic substitution: Sanhuan Group now possesses the capability for AI server-grade high-end MLCCs, Fenghua Advanced Technology has achieved design-ins with leading domestic AI server manufacturers, and Lihexing stands to benefit directly as a dedicated testing equipment supplier. 

 

 

 

Manufacturer Updates

STM

Price hikes ignite a surge in MCU and sensor demand

STMicroelectronics' official price increase notice is pushing the MCU and sensor market into renewed volatility. Effective April 26, the company implemented across-the-board price adjustments, with MCU series rising uniformly by 5% and sensor series seeing notably steeper increases ranging from 15% to as much as 70%, depending on category. All unshipped orders are now subject to the new pricing. Compounding this, the strategic reallocation of general-purpose MCU production lines to automotive-grade products has lifted spot market MCU prices across the board. The STM32F and STM8S series are attracting the strongest inquiry activity.

 

ADI

Demand strengthens across multiple segments, AI and industrial orders continue to recover

Analog Devices is seeing a notable return of demand this month, with orders steadily recovering, shipments accelerating markedly, and spot prices fluctuating rapidly in response. In industrial automation, the order recovery extended into May, with the book-to-bill ratio, excluding price increases, climbing back above 1.0. Demand in smart manufacturing, energy, and automotive electronics remains particularly robust. The data center and communications segment is experiencing explosive growth. Demand for power management, signal chain, and automated test equipment essential for AI servers has surged, driving a 63% year-on-year revenue increase in the first quarter for this segment and marking a third consecutive quarter of double-digit growth. Within this, the ATE business accelerated further in the first quarter of fiscal 2026 on top of a 40% full-year increase in fiscal 2025, fueled primarily by the surge in testing demand for AI chips such as GPUs and ASICs. Meanwhile, automotive electronics serves as a second growth curve, with demand for onboard analog chips holding stable and rising EV penetration driving a 15% year-on-year increase in related revenue for the first quarter.

 

 

TI

New pricing set for July, with steeper increases on older part numbers

Texas Instruments plans to implement its new pricing structure on July 1. Older part numbers will see an increase of approximately 22%, while newer part numbers will rise by only 5%. This significant disparity is widely interpreted as the manufacturer's strategic intent to accelerate customer migration to newer part numbers. In anticipation, spot market prices are expected to adjust accordingly in the near term, and lead times for PPV orders may also see some extension.

 

 

Broadcom

AI-driven volatility rocks PCI-E components, SX series shortage remains intractable

Broadcom continues to see firm demand for AI-related products. Price swings for PCI-E components remain extreme. SS26, for instance, once breached the $1,000 mark before rapidly slipping below $900, only to recently climb back above $900. In contrast, SS24 has shown less fluctuation, largely stabilizing above the $600 level. Both components remain highly active and warrant close monitoring. Within the expansion card segment, the SX series remains mired in a prolonged shortage with no improvement in lead times. Key part numbers such as SX05, SX06, and SX03 continue to see strong interest, and spot availability is extremely limited. Customers with genuine demand are advised to secure inventory and place orders as early as possible.

 

 

Microchip

Demand remains subdued, deals sparse, lead times stable for now

Microchip's overall demand remains lackluster this month, with shortage-driven urgent requirements particularly scarce. Market activity is predominantly centered on PPV orders. Shortage demand appears only sporadically, limited to select PMC series networking components, Microsemi FPGA products, and some PIC16 series MCUs. However, customer acceptance of current pricing remains generally low, resulting in limited actual transactions. Lead times across product lines are largely unchanged, with general-purpose 8-bit and 16-bit MCUs at 4 to 12 weeks, Ethernet and USB products at 8 to 16 weeks, and FPGA products maintaining a range of 12 to 20 weeks.

 

 

Infineon

AI power chips become core earnings driver, revenue guidance raised sharply

Infineon continues to experience explosive demand for its AI data center power management chips. Following the official price increase announced in April, pent-up spot market demand has been unleashed, with particularly active trading in lower-priced, high-volume diode products. AI-related chips have now become a core profit line for Infineon. Popular models such as TDA22590 and TDA21470 face tight supply conditions, and spot prices continue to climb. Reflecting strong confidence, the company has officially raised its revenue outlook for the AI business this fiscal year from €1.0 billion to €1.5 billion. The automotive semiconductor business, in contrast, is exhibiting structural divergence within an overall stable trend. Demand for automotive MCUs and ADAS-related chips remains robust with steady downstream orders, while demand for high-voltage devices used in EVs has softened. Nevertheless, Infineon's supply chain remains well-ordered, keeping overall supply stable and sufficient to meet regular order requirements.

 

 

NXP

Second price hike of the year imminent, spot market reacts with upward momentum

Following its first round of price increases on April 1, NXP is set to implement its second price adjustment of 2026 starting June 1, with the market anticipating multiple rounds of hikes through the year. This latest wave is driven by across-the-board increases in raw material, energy, labor, and logistics costs. Automotive chips, industrial-grade MCUs, power ICs, and power discrete devices are bearing the brunt of these increases. Price adjustment magnitudes vary by customer and part number, with core automotive-grade components generally rising in a range of 5% to 15%. Driven by pre-buying activity ahead of the June price hike, NXP's spot market in May has entered a broad upward channel. Spot price volatility is most pronounced for automotive and industrial-grade products. Factory clients are advised to plan proactively, locking in spot inventory and placing orders as early as possible in the latter half of May.

 

 

Renesas

Industrial and automotive MCU supply under severe strain, spot scarcity drives prices higher

Renesas is grappling with rising demand, extended lead times, and persistent supply tightness across its industrial and automotive MCU product lines, pushing market prices steadily upward. Over the past two months, demand for models such as ISL99390 and ISL99360 has been exceptionally strong, resulting in rapidly escalating prices. Standard transistor lead times now exceed 32 weeks, while power management ICs like ISL9931 and RAA270 are forecast at 24 to 38 weeks. Driven by strong AI-related demand, Renesas has diverted some automotive-grade wafer capacity to AI server power management ICs, directly causing excessively long lead times and a severe spot market shortage for automotive-grade components.

 

 

onsemi

Q1 revenue grows steadily, AI data center emerges as a new growth pillar

onsemi reported first quarter 2026 results, posting revenue of $1.513 billion, a year-on-year increase of 4.68%. GAAP net loss attributable to the company narrowed sharply by 93.13% to $33.4 million. AI data center-related revenue grew over 30% sequentially, emerging as a critical growth engine. Demand in the automotive and industrial sectors also strengthened concurrently. In the spot market, the broader wave of price increases has spurred a rise in demand for MOSFETs, though lead times for discrete components and MOSFETs remain extended at 35 to 52 weeks.

 

 

Xilinx (AMD)

Capacity tilted toward AI creates year-long tight balance, sky-high premiums become the norm

Xilinx faces a market landscape expected to remain in a state of tight balance throughout the year, with only potential marginal relief in the fourth quarter. On the pricing front, official prices rose 10% to 20% in March, while spot market premiums range from 20% to 60%, exceeding 80% for automotive and AI-qualified models. Lead times stand at 24 to 40 weeks for general-purpose parts, stretching to a maximum of 52 weeks for industrial and automotive grades. Orders placed now are not expected to ship until the second quarter of 2027. The structural imbalance stems from a clear logic: the AI infrastructure boom has triggered a surge in global FPGA demand, prompting the manufacturer to allocate 70% to 80% of advanced capacity toward high-margin AI models. This has severely compressed the supply of general-purpose FPGAs, while robust demand from the industrial, automotive, and communications sectors offers no viable alternatives, ultimately resulting in continuously lengthening lead times and amplified spot premiums. To manage this risk, the market broadly recommends customers lock in orders 6 to 12 months in advance.

 

 

Lattice (Intel)

$1.65 billion firmware acquisition targets data center ambitions

Lattice Semiconductor recently announced the acquisition of AMI, a major American firmware company, for a total consideration of $1.65 billion. AMI's products are deployed in over 70% of server platforms globally. This acquisition will effectively close Lattice's gap in low-level firmware and system management software, enabling the deep integration of FPGA hardware with customized firmware to build a secure management and control platform purpose-built for cloud and AI data centers. The transaction is expected to close in the third quarter of 2026. Currently, Lattice's overall product lead times remain stable at 12 to 30 weeks as of May, with pricing trends similarly steady.
 

 

 

 

Spot Market Insights

Memory

eMMC / NAND Flash / NOR Flash

• Overall Trend & Supply-Demand Dynamics: NAND Flash supply remains tight, with low-cost inventory largely depleted — the market has decisively shifted into seller territory. Customer acceptance of price hikes has improved notably, and procurement strategies have pivoted from cost reduction to supply assurance. By late May, eMMC prices were edging higher, low-cost spot supply had been swept clean, and incoming shipments from vendors remained minimal. Industry warnings indicate the NAND and eMMC supply-demand gap is still wide, with a clear upward pricing trajectory expected heading into H2.
• MLC/SLC NAND — Structural Shortage: Vendors continue phasing out MLC and 2D NAND capacity, creating an irreversible supply contraction. Low-density applications are migrating to SLC NAND, driving a sharp demand surge. Kioxia and SanDisk are maintaining minimal channel shipments; recent bulk buying has pushed SanDisk 8Gb and Kioxia 4Gb prices notably higher. Samsung led the upward adjustment of eMMC quotes, setting the tone for the broader market. Q2 official pricing for SLC/MLC NAND is expected to sustain upward momentum, with low-density eMMC and industrial-grade products commanding scarcity premiums. Price increases for these segments will continue to outpace high-density models, demonstrating stronger pricing resilience than consumer-grade counterparts.
• Taiwan-Based Brands: Winbond and MXIC have raised official pricing across the board. MXIC announced a 40% price hike effective next month, triggering a halt in quoting from agents and major channels, with quote validity windows shortening significantly. MXIC shipments from the vendor remain extremely low, shortages are acute, and order intake is currently suspended. ISSI quoting is similarly sparse, with agents largely withholding quotes and lead times continuing to stretch.
• Low-Density NOR Flash: The channel market is marked by a sharp bifurcation. Winbond NOR Flash faces competitive pressure from domestic replacement brands, with severe price inversions on select part numbers. MXIC, by contrast, is grappling with severe shortages and minimal vendor shipments; quoting is scarce, with most agents either withholding entirely or offering only orders with increasingly extended lead times.

 

 

DDR3 & DDR4

• DDR4: The DDR4 market in late May exhibited structural divergence and rotational dynamics. Samsung DDR4 8Gb pricing stabilized, but Micron and Nanya saw exceptionally active trading. Aggressive bulk buying by U.S.-based capital pushed prices to new all-time highs, with supply extremely tight. Samsung DDR4 4Gb entered a catch-up rally, supported by clear market sweeping activity, while DDR4 16Gb remained relatively stable. Winbond also surged in tandem. Underpinning the broader move: consumer and niche DRAM demand has surged as vendors redirected roughly 80% of capacity to HBM and DDR5. Large-scale production line shutdowns have triggered a supply cliff — the primary structural driver of the current cycle.
• DDR3: DDR3 pricing continues to break higher. Supply remains constrained with minimal vendor capacity allocation. Bulk buying orders persist in the market, driving all brands to new all-time highs. Trading is active, with a steady-to-firm pricing tone.
• Demand Spillover & Brand Highlights: Sustained price hikes from vendors have intensified cost-down pressure among end customers, driving a sharp uptick in orders for Taiwan-based brands as alternative sources. Nanya faces limited market availability due to stringent shipment controls, with quote increases outpacing those of Samsung. Industrial and automotive-grade DRAM remains in acute shortage.
  

 

 

LPDDR

• Market Overview:  LPDDR pricing has broadly stabilized, but brand-level divergence is pronounced. Micron LPDDR supply from the vendor is relatively steady; however, high premiums in the market are suppressing trading activity, with some inventory holders showing increased willingness to offload. Samsung and SK Hynix LPDDR supply remains tight, with relatively active trading. Automotive-grade products are especially scarce, with the supply-demand imbalance particularly acute.
• Product Segmentation: Vendors have announced discontinuation of LPDDR4/4X and are tightly controlling shipments, leaving channel inventory nearly depleted. Once remaining low-cost inventory is absorbed, the supply-demand gap is expected to trigger a rapid price rebound. LPDDR5/5X momentum is cooling, weighed by the crowding-out effect of AI servers and weaker smartphone shipment expectations. Trading activity is thin, with select part numbers showing signs of price softening.
  

 

 

Server Memory (DIMM)

• DIMM Market Spotlight & Pricing: Contract pricing remains on an uptrend, though momentum is moderating. The current market focus is concentrated on 64GB modules, where inquiries are increasing and prices are being bid higher. AI server capex continues to squeeze capacity, keeping supply for conventional server memory tight. Vendor quotes remain firm, with limited room for negotiation.

 

 

 

Storage

Hard Disk Drives (HDD)

• High-Level Consolidation Tone: Overall HDD demand remained weak in May. Select inventory holders offloaded stock, causing a modest market price dip. The broader picture, however, remains one of high-level consolidation without a directional breakout. Seagate and Western Digital have raised official pricing by approximately 30%.
• Enterprise High-Capacity: Demand for 20TB/24TB enterprise drives remains concentrated and continues to be the primary demand driver. Pricing is relatively firm, with channel prices steadily pushed higher. Trader-to-trader pricing is now on par with official distribution pricing. However, OEM customers with bulk procurement needs are showing clear resistance to a second round of price hikes, creating upward resistance. Under these competing forces, the market remains in a phase of high-level consolidation with limited upside momentum.
• Consumer Low-Capacity: Demand is sluggish. Pricing is largely flat, characterized by pronounced sideways movement and minimal channel price fluctuation. Weakness in PC and consumer electronics caps upside potential for low-capacity products — sticky upward resistance remains the core challenge. Genuine demand from select customers continues to provide a floor, sustaining baseline transaction volumes.

 

 

Solid-State Drives (SSD)

• Two-Tier Pricing Structure : The SSD market exhibits rigid contract price increases alongside divergent spot pricing. The contract market remains firmly under vendor control, while the spot market has bifurcated due to uneven end-demand.

• Samsung Enterprise: Vendors have initiated price hikes, with spot inquiry activity picking up notably. High-capacity NVMe drives are under multi-pronged pressure from official price increases, growing demand, and internal labor disputes. The PM9A3 15.36TB channel quote has surged 15%, with strike risks threatening further supply disruption.

• Samsung Consumer: PC and notebook demand remains persistently weak, with spot channels focused on inventory digestion, resulting in gradual price erosion. Overall increases are capped at around 10%. From a longer-term perspective, this spot softness is merely short-term channel destocking and carries no structural trend signal. Once low-cost inventory clears, consumer product pricing will be pushed upward as OEM input costs rise.
• Solidigm Enterprise: Channel activity reflects a pronounced cyclical pattern: low-demand conditions in April prompted inventory accumulation at lower price points in anticipation of Q3 price hikes. In May, a surge in overseas demand triggered aggressive buying of the P5520 7.68TB, driving a sharp price spike. Additionally, SATA interface models are scheduled for discontinuation by end-September, with the portfolio pivoting to PCIe products — a product transition window worth monitoring.
  

 

 

 

CPU

Mobile CPU

• Pricing Environment: Overall pricing remained stable in late May, with demand at normal levels. Inquiries were concentrated on models such as N97, N100, and N150.
• Demand Dynamics: Supply of the previously severely constrained Intel small-core lineup (N305, N97, 135H) remains tight, though prices did not spike further in the near term, suggesting the market has entered a wait-and-see digestion phase. 

 

 

PC CPU

• Trend Assessment: The overall pricing trend is stable, with non-suffix desktop models performing relatively steadily.
  
• Demand Dynamics: 12th-gen legacy models saw increased demand recently, with tight market supply unchanged. Prices remain on an upward trajectory, underscoring resilient demand for legacy platform life extension.

 

 

Server CPU

• Supply Conditions: The overall market is oversupplied, with pricing remaining chaotic. Some peers are engaging in distressed selling due to inventory pressure.
• Demand Focus: Demand is primarily driven by overseas orders, with key inquiries concentrated on Intel 5th/6th Gen and AMD 3rd/4th Gen processors. Intel 4th Gen and AMD 5th Gen pricing is showing early signs of softening, though spot availability persists.
• Segment Shifts: Pricing on Intel 4th and 5th Gen decommissioned units has declined, pulling down new unit pricing. However, the 4510 model edged higher, reflecting supply chain volatility and growing customer acceptance of decommissioned prod

 

 

 

Network Adapters

 

• Market Conditions: Overall demand for RAID cards remains relatively soft, with pricing stable and limited market volatility.
• Broadcom: Inquiry activity for Broadcom's high-end series remains active, with the 9500/9600 series drawing sustained attention.
• Mellanox: Mellanox network adapter demand continues on a moderate growth trajectory.

 

 

 

GPU

 

• Market Conditions: Overall demand is relatively muted, but structural tensions are pronounced.
• Supply Dynamics: PRO 6000 series pricing continues to climb across all three variants, exacerbated by hoarding activity and customs clearance hurdles, further intensifying the supply-demand imbalance. The RTX 4000 ADA remains in acute shortage, with market supply extremely scarce.