Market Insights and Trends

 

ECIA March Survey: Component Sales Sentiment Hits 5-Year High, Yet Lead Times Stretch for 61%  

The ECIA's latest survey released on April 22, 2026 shows that the average electronic component sales sentiment index jumped to 149.2 in March 2026, the highest in nearly five years since April 2021. Seven of ten component subcategories posted confidence above 150, with two exceeding 160. The end-market index rose 16.1 points to 161.8. Only 2%-4% of respondents reported worsening market conditions, with most markets reporting zero deterioration, while positive industry trends have continued for 11 months.

 

However, supply chain pressures are mounting: the share of respondents reporting longer lead times surged from 34% in February to 61% in March, while those reporting shorter lead times fell to near zero. Among them, 78% said semiconductor lead times are extending, with severe supply-demand imbalance in advanced memory ICs being the biggest concern. Although sales confidence for April is expected to moderate slightly, the overall index remains high at 143.7.

 

 

 

 

Omdia Sees 62.7% Chip Revenue Jump in 2026, Compute and Storage to Top $700 Billion 

According to Omdia's significantly revised forecast, global semiconductor industry revenue will grow 62.7% in 2026, with DRAM market size expected to nearly double and NAND flash market size potentially quadrupling from 2025. Meaningful supply relief may not arrive until 2027. On the enterprise side, 2026 will see a major server replacement cycle, compounded by component shortages pushing up average selling prices. The computing and data storage segment is expected to grow 90% year-on-year, exceeding $700 billion.

 

 

Omdia notes that consumer electronics and wireless applications have positive prospects in 2026. While smartphone shipments are expected to remain flat, rising memory prices will significantly boost semiconductor revenue and bill-of-materials costs. Foldable phones and flagship models with integrated AI features will launch intensively, while smartwatches and wearables are also expected to contribute meaningful growth.

 

AI has greatly driven demand for memory and processing ICs, but suppliers' ability to expand capacity and the ROI of AI capital spending remain unknown. Beyond tariffs, energy costs and geopolitical pressures, the industry also faces risks from the sheer scale of AI infrastructure investment. Current semiconductor revenue growth is primarily driven by rising average selling prices rather than shipment volume growth—a dynamic unprecedented in both scale and scope.

 

 

TrendForce Lowers 2026 Server Growth Forecast to 13%; SemiAnalysis Warns Memory Spending Share to Hit 30% 

TrendForce's latest research shows that as suppliers prioritise higher-margin AI servers, lead times for many general-purpose server components have stretched significantly, cutting the overall server market's growth potential from nearly 20% to an estimated 13%. PCB and CPU lead times have extended to close to one year, PMIC lead times have stretched from 21–26 weeks to 35–40 weeks, and BMC IC lead times have lengthened from 11–16 weeks to 21–26 weeks. By contrast, AI server shipments are expected to grow roughly 28% year-on-year in 2026, fuelled by strong demand from cloud service providers, with ASIC AI servers outpacing GPU-based models, though TrendForce has slightly trimmed the ASIC share from nearly 28% to around 27%. Excess orders and procurement momentum are expected to carry into 2027.

 

Meanwhile, SemiAnalysis forecasts that memory spending will absorb roughly 30% of hyperscale data centre capital expenditure by 2026, nearly a fourfold jump from around 8% in 2023 and 2024, with a further climb expected in 2027. DRAM prices are projected to more than double in 2026, while LPDDR5 contract prices have already more than tripled since Q1 2025 and could pass US$10 per gigabyte on the open market this quarter. Memory already accounts for a significant portion of the roughly US$250 billion in new hyperscale data centre spending this year, and AI servers built on Nvidia's LPDDR platforms are seeing the steepest price increases, with B200 server prices expected to rise up to 20% by year-end. SemiAnalysis notes that Nvidia secures exceptionally favourable DRAM pricing well below market rates, compressing its own costs while suppressing the market pricing benchmark, whereas AMD, unable to access comparable discounts and shipping lower volumes, is far more exposed to rising memory costs. On the supply side, Samsung, SK Hynix and Micron have shifted capacity toward HBM and high-margin enterprise DRAM, with meaningful new capacity not due to come online until 2027 or 2028 at the earliest.

 

 

 

Rohm, Toshiba and Mitsubishi Electric to Merge Power Chip Units, Targeting Second Place; IDM Lead Times Extend to 30 Weeks 

Japan's power semiconductor sector is witnessing a heavyweight consolidation, with Rohm, Toshiba and Mitsubishi Electric signing a memorandum of understanding to merge their power semiconductor businesses and set their sights on becoming the world's second-largest supplier. According to Omdia data released in March 2026, Infineon led the global power semiconductor market in 2025 with a 24.4% share, followed by ON Semiconductor at 7.9% and STMicroelectronics at 5.4%. Among Japanese players, Mitsubishi Electric ranked fourth with roughly 3%, while Toshiba and Rohm each held around 3% or less. If the three-way integration is completed, the combined entity would command approximately 11.3% market share, vaulting to second place globally behind only Infineon. In the small-signal discrete component market, a merged Rohm and Toshiba would capture a combined 19.7% share, also good for second worldwide. The post-merger business mix would be well-balanced, split roughly into automotive at 36%, industrial at 35% and other segments at 29%, positioning the new company to capture demand across electric vehicles, industrial automation and AI data centres simultaneously.

 

At the same time, power component shortages are deepening. As AI server power consumption continues to climb, lead times for select power components from major IDMs have stretched to 30 weeks, with a wave of price hikes looming. Texas Instruments posted quarterly revenue of US$4.825 billion, up 19% year-on-year, and guided current-quarter revenue to US$5 billion to US$5.4 billion, both ahead of expectations. By sub-segment, the silicon carbide and gallium nitride power device market is forecast to exceed US$14 billion, with the power SiC market alone projected to surpass US$10 billion within five years and the power GaN market expected to top US$3 billion by 2031. On the supply chain front, TSMC's exit from GaN foundry services has triggered a reshuffle, with IDMs such as Infineon and Rohm pivoting toward more vertically integrated manufacturing strategies. With AI data centre demand continuing to ramp, power component shortages and upward pricing pressure are expected to persist for the long term.

 

 

China MCU Crunch Sparks 10%–20% Price Hikes, Led by Zhongwei and Nations Technologies 

China's domestic MCU industry is grappling with severe capacity tightness and mounting cost pressures. On April 9, 2026, Zhongwei Semiconductor announced an across-the-board price increase averaging over 10%, following an earlier round of 15%–50% hikes on MCU and NOR Flash products on January 27. The company is facing extended wafer lead times, steadily rising order volumes and a growing backlog of unfulfilled orders, all while inventory sits at historic lows. For full-year 2025, Zhongwei posted revenue of 1.122 billion yuan, up 23.09% year-on-year, with net profit surging 107.68% to 284 million yuan; shipments of high-end 32-bit MCUs jumped 43% year-on-year. Nations Technologies implemented new pricing on April 7, raising select products by 15%–20%. The company recorded 2025 revenue of 1.36 billion yuan, up 16.51%, while narrowing its net loss by 50.96% to 115 million yuan. Puya Semiconductor announced price increases on general-purpose MCUs effective April 15, having earlier signalled that NOR Flash prices would maintain a moderate upward trend, with large-capacity product hikes gradually filtering through to small and medium-capacity segments. The domestic MCU sector has now entered a wave of coordinated price adjustments: Chipsea Technologies raised prices 10%–20%, Goke Micro lifted co-packaged products by 40%–60%, Silan Micro added 10% on small-signal diode, transistor and MOS chips, and NCE Power hiked MOSFET products by a minimum of 10%. Industry expectations point to a continued semiconductor price upswing through the first half of 2026, with more MCU, power device and analogue chip makers likely to follow suit.

 

 

Semiconductor Materials Costs Surge Broadly: Shin-Etsu Leads with 10% Increase; Tungsten Powder Surges 6-Fold in One Year 

The global semiconductor materials market is facing a fierce wave of price increases. Shin-Etsu Chemical will raise all silicon product prices globally by more than 10% effective May 1, 2026, following an earlier announcement by the world's top silicon wafer supplier of another 5% to 8% hike in the second half, with 12-inch polished wafer average prices expected to reach US$80 in the third quarter. Copper-clad laminates are seeing even sharper increases, with Taiwan Union Technology raising some series by 20% to 40% and Elite Material lifting high-end materials by roughly 10%. Tungsten prices are undergoing a dramatic surge — driven by China's tightened export controls, tungsten powder has soared six- to sevenfold in a year, from roughly 300 yuan per kilogram in early 2025 to 2.32 million yuan per tonne in early April, while the international benchmark ammonium paratungstate price has jumped 557% since February 2025. Japan's Kanto Denka Kogyo and Central Glass have already issued supply disruption warnings to Samsung Electronics and others.

 

Key raw materials for photoresist are also facing a severe shortage crisis. Samsung Electronics and SK Hynix have received supply cut notices from Japanese suppliers, with the critical shortfall centred on PGME and PGMEA — essential components for photoresist and HBM temporary bonding adhesives. Japanese photoresist suppliers collectively control around 76% of the global market. The root cause lies in a naphtha supply breakdown — Japan relies on Middle Eastern imports for over 40% of its naphtha, and the blockade of the Strait of Hormuz has sent naphtha spot prices soaring 92% from roughly US$600 to US$1,190 per tonne. The ripple effects have pushed PGMEA market prices up 40% to 50%. South Korea holds only about two weeks of naphtha inventory and has imposed a provisional five-month ban on naphtha exports. Switching raw material suppliers requires customer requalification, a process that typically takes around one year, putting advanced-node chips at greatest risk; production of HBM and advanced DRAM at both Samsung Electronics and SK Hynix could face direct impact. In contrast, China has lower reliance on Middle Eastern naphtha, and domestic producers already possess mass production capabilities at electronic-grade levels, potentially accelerating import substitution.

 

 

 

Memory Spot Prices Rise 4-7 Times Since Last September, Supply Gap to Last Through 2027 

Current DRAM spot market prices have risen more than four times above September 2025 levels, while NAND flash spot prices are nearly seven times higher. Contract prices have also more than doubled from 2025 levels. Quarterly capital spending by the six major hyperscalers has surged from the usual $30-50 billion to $70 billion in Q3 2024, approaching $130 billion by Q4 2025.

 

 

 

As HBM yields only one-third the die capacity per wafer compared to standard DDR DRAM, supply is struggling to keep pace with demand even without considering the AI investment boom. Dell's CEO predicts that AI accelerator memory demand by 2028 will reach 625 times 2023 levels—driven by per-chip memory capacity expanding 25-fold (from 80GB to about 2TB), combined with a roughly 25-fold increase in AI accelerator shipments over five years.

 

Samsung, SK Hynix and Micron together control about 91% of the global DRAM market, yet global DRAM supply by the end of 2027 will only meet about 60% of market demand. Research firm Counterpoint notes that DRAM capacity needs 12% annual growth between 2026 and 2027 to ease shortages, but the three majors' actual expansion plans amount to only 7.5% annual growth. Memory prices surged 110% year-on-year in Q1 2026.

 

Samsung plans to start its Pyeongtaek P4 fab in 2026, but full mass production will not come until 2027 or later. SK Hynix's Cheongju M15X fab is the only meaningful capacity addition this year, targeting 50,000 wafer starts per month. Micron's new fabs are expected to contribute gradually from 2027 to 2028. SK Group's chairman has warned that AI memory supply constraints could persist until 2030. Memory costs already account for about 20% of low-end smartphone manufacturing costs, expected to approach 40% by mid-2026. Counterpoint forecasts supply-demand balance will not return to normal until 2028.

 

 

Samsung Lifts Q2 DRAM Contract Prices Another 30%, Two-Quarter Gain Hits 130% 

Samsung Electronics has confirmed Q2 price negotiations with major clients, raising prices approximately 30% on top of the 100% doubling in Q1 2026, covering HBM and all general DRAM products. Based on early 2025 levels, Q2 2026 supply prices have soared to 2.6 times the baseline. SK Hynix and Micron are following with similar increases, with the three majors moving in lockstep.

 

SemiAnalysis forecasts that global DRAM supply will remain about 7% below demand in 2026, with the HBM supply gap widening from about 5% in 2025 to about 6% in 2026, and further to about 9% in 2027. DRAM now accounts for 35% of total smartphone costs, with NAND flash at 19%, together reaching 54%—making memory chips the absolute core of hardware costs.

 

The latest server DRAM "DDR5 16Gb" has surged to 37 per chip, up from just 6.01 last June—a more than five-fold increase in under one year, with per-gram pricing now exceeding gold. TrendForce data shows Consumer DRAM price momentum in March 2026 concentrated in sub-4Gb capacities, with DDR4 4Gb average prices rising over 20% month-on-month, while DDR3 and DDR2 saw 20%-40% increases across most capacities. Consumer DRAM contract prices are expected to rise another 45%-50% quarter-on-quarter in Q2 2026.

 

Daiwa Securities analysis indicates that DRAM supply tightness will persist throughout 2026, with Samsung and SK Hynix gradually converting quarterly supply contracts into three-year or longer long-term agreements, and the upward trend expected to continue at least through 2027.

 

 

Samsung Exits LPDDR4/4X, Unlocking $15 Billion Opportunity for Chinese Memory Makers 

Samsung Electronics officially stopped taking new orders for LPDDR4 and LPDDR4X mobile DRAM on April 17, with full production cessation expected by Q1 2027 as lines convert entirely to LPDDR5/5X and HBM. The series has seen cumulative shipments exceed 50 billion chips, with prices soaring from 6 per chip in March 2025 to 28.50 in January 2026—a nearly four-fold increase in under one year. Samsung's Hwaseong Line 12 has ceased 2D NAND production and is being converted into a 1c DRAM back-end packaging facility with monthly capacity of approximately 80,000 to 100,000 12-inch wafers.

 

As a result, memory chips now account for 30%-40% of smartphone bill-of-materials costs, up sharply from 10%-15%, with OPPO, vivo and Xiaomi expected to raise mid-tier phone prices by 300-500 yuan. In Q1 2026, LPDDR5/5X captured 70% of the smartphone DRAM market, projected to exceed 90% by 2027. Samsung's exit creates approximately $15 billion in market replacement opportunity for Chinese memory makers. Changxin Storage and other domestic players have achieved mass production in the LPDDR4/4X segment, with Chinese makers' share of the LPDDR4X market rising from 5% in 2024 to 18% in Q1 2026. Changxin Storage has announced plans to increase capacity by 30% and launch next-generation 12nm products.

 

 

Samsung and SK Hynix Push 3-5 Year Long-Term Contracts; Apple Secures DRAM at Twice the Price 

The memory market is seeing rapid expansion of long-term agreements. Samsung and SK Hynix are actively negotiating multi-year contracts lasting three to five years, with Microsoft and Google proposing 10%-30% prepayments to secure DRAM supply. SK Hynix is in final negotiations with Microsoft on a multi-trillion won three-year DDR5 contract, while also discussing long-term supply of server DRAM and HBM with Google. According to DRAMeXchange, DDR4 fixed transaction prices have soared nearly ten-fold from $1.35 last March to $13. Long-term agreements are expected to expand beyond HBM to general product lines including DDR5 and enterprise SSDs, with some contracts possibly including price floor provisions to stabilize supply.

 

Apple is also aggressively stockpiling memory chips. Samsung recently signed a supply contract with Apple for 12GB LPDDR5X for the iPhone 18 series, with Apple accepting roughly double the previous price—chip prices have risen from around $30 early last year to about $70. With SK Hynix focused on HBM production, Samsung is nearly the only supplier capable of meeting Apple's volume requirements. Apple targets 240 million iPhones annually, with average memory capacity of about 10GB per iPhone based on Pro series shipment share—consuming approximately 2.4 exabytes (2.4 million TB) of LPDDR5 DRAM each year. This massive demand, combined with Apple's financial strength, gives the company significant ability to distort the memory market. Industry insiders believe long-term agreements and prepayment-driven liquidity will provide critical funding for manufacturers' capacity expansion, pushing the memory industry toward longer-term cycles.

 

 

SK Hynix Q1 Profit Surges 405%, Record 72% Margin Tops TSMC and Nvidia 

SK Hynix reported first-quarter 2026 revenue of 52.58 trillion Korean won (roughly 35.57 billion US dollars), a 198% jump from a year earlier and the first time quarterly revenue has surpassed the 50 trillion won mark. Operating profit surged 405.5% to 37.61 trillion won (about 25.44 billion dollars), delivering an operating margin of 72% and a net margin of 77% — comfortably outpacing TSMC's 58.1% in Q1 and Nvidia's 65% in the fourth quarter of last year. For context, the company's operating margin had slumped to minus 67% in the first quarter of 2023, completing a stunning turnaround in just three years. DRAM accounted for 78% of total revenue with average selling prices climbing roughly 65% quarter-on-quarter, while NAND contributed 21% with prices up around 75%. According to TrendForce, some DRAM contract prices rose nearly 83% sequentially, and select NAND products soared about 160%. SK Hynix's market capitalisation has surged 83.77% from 473.93 trillion won at the end of last year to 870.92 trillion won. The company expects second-quarter DRAM bit shipments to grow 7% to 9% sequentially and NAND bit shipments to increase 16% to 17%. The market widely anticipates Q2 operating profit will reach 51.28 trillion won, setting a fresh record for the fifth consecutive quarter.

 

 

Samsung Union Rally Slashes Output 18.4%; Full Strike Could Cost Trillions of Won a Day 

Approximately 40,000 Samsung Electronics union members held a general rally on April 23, significantly impacting domestic semiconductor production. During the rally, overall semiconductor output fell over 18% in a single day, including an 18.4% drop in memory chip night shift production, a 33.1% drop at Hwaseong Line 15, a 23.1% drop at Pyeongtaek P1D (DRAM line), and a 24.6% drop at P2D. Foundry operations fell 58.1% overall, including a 74.3% drop at Giheung S1 Line and a 67.8% drop at Hwaseong S3 Line.

 

The union demands that 15% of the company's 2026 operating profit (approximately 45 trillion won out of expected 300 trillion won) be distributed as performance bonuses, along with the removal of the 50% bonus cap on annual salary. Management has proposed using 10% of profit for performance bonuses and a 6.2% base salary increase—described as "unprecedented" terms—but no agreement has been reached.

 

The union warns of an 18-day total strike from May 21 to June 7 if demands are not met, with estimated production losses potentially exceeding 30 trillion won. An economics professor at Seoul National University projects daily losses of about 1 trillion won, with semiconductor division operating profit alone possibly reduced by as much as 10 trillion won. If the strike proceeds as planned, supply tightness in the global memory chip and foundry markets will further intensify.

 

 

 

Phison Fulfills Only 30% of Orders, Silicon Motion Sees 2027 NAND Gap Deepening 

Two major storage controller makers have sounded severe shortage alarms almost simultaneously. Phison CEO K.S. Pua revealed the company can meet only 30% of customer demand, leaving a massive 70% gap unfilled, and warned this supply crunch may persist for a decade — AI data generation is advancing at a hundredfold pace, while memory fabs can add at most 50% more capacity. To prepare for a possible "cash can't buy supply" situation by Q4, Phison launched its first syndicated loan of NT$12 billion (about US$380 million) in March alongside an US$800 million overseas convertible bond, raising roughly US$1.18 billion for stockpiling. By end-March, Phison's NAND flash inventory on book had topped NT$50 billion, with plans to add over NT$40 billion more in Q2. Financially, Phison grew full-year 2025 revenue 23.3% to NT$72.66 billion, with net profit up 9.9% to NT$8.74 billion; Q4 revenue jumped 81.3% to a quarterly record of NT$22.80 billion, while net profit surged 93.6% to NT$4.63 billion.

 

Meanwhile, Silicon Motion General Manager Wallace Kou warned that NAND flash shortages in 2027 will be even worse than 2026, with the output-demand gap widening further. On pricing, NAND products across different capacities have already surged 4 to 10 times between August 2025 and March 2026, with some low-capacity parts spiking 15-fold. Even as major NAND producers expand output, capacity gains mostly range between 15% and 25%, yet a single Nvidia NVL72 rack requires 1.16 petabytes of storage. Most DRAM and NAND makers see 2027 as the worst shortage year, with potential relief only after new fabs ramp in late 2027 or 2028. In the near term, ADATA has stated that Q2 DRAM and NAND flash contract prices will rise at least 40%. The AI industry is now shifting from training to inference, and Nvidia CEO Jensen Huang has pointed to over a trillion dollars in Blackwell and Rubin-related demand through 2027. On the fundraising front, Phison's board approved an overseas unsecured convertible bond of up to US$800 million, its largest-ever capital raise; this year the company has already completed a NT$6.06 billion convertible bond, the NT$12 billion syndicated loan, and an 18-million-share private placement. Separately, Nanya Technology brought in SanDisk, Kioxia, SK Hynix and Cisco through a private placement that raised NT$78.7 billion, while ADATA sealed a NT$12 billion syndicated credit facility in late March. Pua also flagged that memory makers' gross margins have soared above 80%, calling a profit model where a one-dollar cost inflates to a ninety-dollar end price "unhealthy for the industry." Nanya expects the supply-demand imbalance to begin easing only in the second half of 2028 at the earliest.

 

 

NAND Joins Long-Term Deal Wave as Kioxia Sells Out and Some Prices Jump Eightfold 

The NAND flash market is undergoing a structural shift as long-term agreements, once confined to HBM and DRAM, now extend into this segment. Japanese NAND maker Kioxia is negotiating a three-year deal with a major cloud service provider running through 2029, having already sold out its entire 2026 production capacity and warning that supply tightness will persist into next year. According to DRAMeXchange, 128Gb MLC NAND flash prices have rocketed from US$2.18 in January last year to US$17.73 this March, an eightfold surge, with Q1 prices expected to jump more than 40% sequentially.

 

This demand explosion stems from the AI industry shifting from training to inference. Enterprise SSDs, which read data over three times faster than HDDs, are in high demand, while the arrival of the four-to-six-year server replacement cycle in data centres and a shortage of nearline HDDs are further fuelling appetite for high-capacity SSDs. To lock in stable supply, SanDisk finalised a long-term deal with prepayment terms on January 29 and extended its joint venture agreement with Kioxia by five years to 2034, agreeing to pay US$1.165 billion for manufacturing services and supply guarantees. SanDisk's CEO noted that the growth forecast for data centre storage capacity demand has been sharply revised upward from around 20% at the start of the year to over 60%. Market leaders Samsung Electronics and Solidigm are also scaling up large-scale long-term contract signings in the enterprise SSD space.

 

The HDD market is booming in parallel: Western Digital has secured confirmed orders from seven hyperscalers, including three long-term agreements guaranteeing high-volume, fixed-price supply through 2027 to 2028. Seagate has sold out all nearline HDD orders this year, with long-term deals with key cloud customers now effective through 2027; it posted Q4 revenue of US$2.83 billion, up 21.5% year-on-year, with gross margins hitting a record high. NAND flash and HDD are now locked in a self-reinforcing cycle — soaring NAND prices highlight the cost advantage of HDDs, which cost roughly one-tenth of an SSD for the same capacity, driving up HDD demand, while HDD shortages in turn push demand toward high-capacity SSDs, sending the entire storage market into a boom cycle. On the capacity side, Kioxia plans to invest 400 billion yen this year, a 40% increase from last year but still below the 510 billion yen spent in 2023. The memory industry expects three-to-five-year long-term agreements to become the dominant contract model, though consumers and small-to-medium clients unable to secure large orders risk facing supply shortfalls, deepening market polarisation.

 

 

Kioxia Exits 2D NAND, Ending Four Decades of Floating-Gate Production

Following its mid-March 2026 announcement to discontinue small-capacity TSOP-packaged products, Kioxia issued another end-of-life notice on March 31, revealing plans to phase out its legacy floating-gate 2D NAND flash business entirely. This marks the end of the planar NAND architecture first commercialised by Kioxia's predecessor Toshiba in 1987, drawing a close to roughly 41 years of production. The discontinuation covers floating-gate 2D NAND built on 32nm, 24nm and 15nm process nodes, along with third-generation BiCS 64-layer 3D NAND, spanning SLC, MLC and TLC types in forms including bare wafers, BGA, TSOP, eMMC, UFS and SD cards. Customers must submit final purchase forecasts by September 30, 2026, with last shipments continuing through December 31, 2028, before Kioxia halts 2D NAND supply entirely in 2029. Industry analysts note the exit will leave a structural gap in the 2D NAND market, with Taiwanese players Winbond — focusing on 1Gb to 8Gb SLC NAND for automotive, industrial and IoT applications — and Macronix, which covers the 2Gb to 32Gb segment on its 19nm process, emerging as the primary beneficiaries.

 

Samsung and Kingston Lift SSD Prices Over 10% in Tandem 

Samsung Electronics has officially notified domestic distributors that its SSD suggested retail prices will rise by more than 10%. Kingston followed on April 22, announcing a uniform price increase of no less than 10% across its entire SSD lineup. Take the Samsung 990 PRO 1TB as an example — currently selling at US$300 to US$330, the new price will climb to a record US$330 to US$360, whereas the same product cost less than US$100 last year, meaning prices have already surged three to four times from their prior levels.

 

Earlier in April, both Samsung and Kingston had already nearly tripled prices on select high-end M.2 SSDs, and this latest round of across-the-board hikes confirms both the breadth and persistence of the uptrend. With global NAND flash supply remaining tight and AI infrastructure demand exploding, manufacturers are funnelling more capacity into enterprise SSDs and server applications. Coupled with soaring wafer prices that can no longer be absorbed internally, cost pressures have become inescapable. Against a backdrop where some DRAM contract prices have jumped nearly 83% sequentially and NAND products have spiked around 160%, the synchronised move by the two biggest names to raise prices over 10% likely signals the formal start of a new SSD price cycle, with further upward pressure expected in the months ahead.

 

 

 

CPU Prices Climb 20% While Retail Sales Sink to Decade Low; Intel Q1 Profit Up 156% 

The global CPU market is grappling with a severe supply-demand mismatch: server and consumer CPU prices keep climbing while retail sales have slumped to their lowest level in nearly a decade. Since March 2026, consumer CPU prices have risen 5% to 10%, with server CPU prices jumping 10% to 20%. Intel expects room for another 8% to 10% hike in the second half, while AMD's cumulative server CPU price increases are set to reach 16% to 17% over the second and third quarters. Meanwhile, average CPU lead times have stretched dramatically from one to two weeks to eight to twelve weeks. SemiAnalysis' founder noted that CPU resources across the entire cloud market have been essentially drained over the past six months — even with AWS tripling CPU output, shortages persist; Amazon's CEO revealed two major customers have inquired about buying all Graviton instance capacity for 2026, a demand that simply cannot be met. A January report from KeyBanc analysts also indicated that server CPUs from both Intel and AMD are virtually sold out.

 

The root cause is not a shortfall in production capacity but its reallocation toward AI demand. In the first quarter of 2026, TSMC's 3nm process accounted for 25% of wafer sales revenue, 5nm for 36%, and 7nm for 13%, with advanced nodes making up 74% in total; 2nm has entered mass production, and full-year capex is expected to approach US$52 billion to US$56 billion. The PC segment is feeling the ripple effects, with manufacturers like Dell and HP facing CPU supply shortages that have stretched lead times from two weeks to six months, and in some cases even paying a premium fails to secure supply. Yet this shortage coincides with a sharp contraction in end demand: total system upgrade costs have exploded by over 170% in the past year alone, with memory, SSDs, CPUs and GPUs all rising in tandem while motherboard costs remain elevated, pushing consumers to delay purchase decisions. In the fab allocation model, high-margin AI chips get priority access, while PC CPUs are pushed back due to lower profitability. Intel recently bought back a 49% stake in its Fab 34 in Ireland for US$14.2 billion; the facility is a critical site for Intel 4 and Intel 3 process nodes. On the financial front, Intel posted first-quarter 2026 revenue of US$13.6 billion, up 7% year-on-year and marking its sixth consecutive quarterly beat, while net profit soared 156% to US$1.5 billion. Shares jumped nearly 20% in after-hours trading following the report and are now up over 80% year-to-date. The data centre and AI segment led the way with revenue up 22% to US$5.1 billion. The company guided second-quarter revenue to US$13.8 billion to US$14.8 billion, noting that demand still outstrips supply across all businesses, particularly for Xeon server CPUs, with both the industry and the company expecting double-digit year-on-year server CPU shipment growth extending into 2027. Over the past twelve months, Intel shares have surged 211% and AMD shares 169%, with AMD EPYC CPUs now commanding 41% of the server market. On the partnership front, Google has committed to running AI workloads on multiple generations of Intel CPUs, and Xeon 6 has been selected as the host CPU for Nvidia's DGX Rubin NVL8 system; Elon Musk announced Intel as the manufacturing partner for his Terafab project, targeting a monthly output of one million wafers. Analysts expect the CPU market to remain highly tight through 2026 to 2027, with capacity — not demand — as the decisive factor, suggesting the uptrend in CPU pricing will persist.

 

 

 

GPU Rental Rates Jump 40%, Rubin Output Target Cut 25% to 1.5 Million Units 

Explosive growth in global AI computing demand has plunged the GPU market into severe supply shortages. Anthropic's annualised recurring revenue skyrocketed from US$9 billion to over US$25 billion in just one quarter, nearly tripling, while surging adoption of open-source models such as Claude 4.6 Opus, Claude Code, GLM and Kimi K2.5, coupled with intense fundraising by Anthropic and OpenAI, have together created enormous demand for GPU resources. GPU rental prices have spiked in response: one-year H100 lease contracts have jumped from US$1.70 per GPU hour in October 2025 to US$2.35 in March 2026, a near 40% increase. On-demand rental capacity across all GPU types has sold out entirely, with customers scrambling for AWS p6-b200 spot instances at US$14 per GPU hour and some contracts being renewed through 2028. Even locating an 8-node (64-GPU) H100 or H200 instance has become extremely difficult — half of the suppliers surveyed are sold out, and all capacity scheduled to come online by August to September 2026 has been fully booked.

 

On the supply side, both new and old Nvidia chips are commanding premium prices: a two-year-old used H100 still sells at 85% of its original price, retaining 84% of its value into the third year. However, the next-generation Rubin architecture GPU has seen its 2026 output target cut due to HBM4 validation delays at SK Hynix and Micron. KeyBanc analysts estimate Nvidia has lowered its 2026 Rubin GPU production target by 25%, from the original 2 million units to 1.5 million, while the projected number of Vera Rubin server racks — pairing Vera CPUs with Rubin GPUs — has been halved from 12,000 to 14,000 units down to roughly 6,000. Even so, Nvidia has secured roughly 650,000 CoWoS interposer wafers for 2026, enough to support shipments of 5.5 to 6 million Blackwell series GPUs, 1.5 million Rubin GPUs and 1 million Hopper GPUs. On the demand side, the emergence of multi-agent workloads is driving parabolic growth in compute consumption, with Claude Code adoption soaring and expected to account for over 20% of all daily code commits by end-2026. Given that AI tool return on investment ranges from five to ten times, analysts see ample room for GPU rental prices to rise further.

 

 

Nvidia's China AI Chip Share Halved to 55%, Huawei Leads Local Market with 812,000 Units 

According to the latest IDC data, domestic GPU and AI chip vendors together captured nearly 41% of China's cloud AI accelerator market in 2025, delivering a combined 1.65 million AI accelerator cards. Of the roughly 4 million accelerator cards shipped in China overall, around 2.2 million were Nvidia-based, giving the company a market share of about 55% — nearly halved from the 95% share it once claimed before export controls took hold. Huawei emerged as the biggest winner in China's cloud AI chip market in 2025, shipping approximately 812,000 AI chips and seizing close to 20% of the domestic market. Alibaba's T-Head ranked third nationally and second among local players with roughly 256,000 units shipped, having cumulatively delivered 470,000 self-developed GPU chips at scale as of February 2026. AMD placed fourth with around 160,000 units and a 4% share, while Baidu's Kunlun and Cambricon each shipped 116,000 units, tying for fifth. Hygon, Muxi and Iluvatar Corex held 5%, 4% and 3% respectively of the domestic AI chip segment, translating to 2%, 1.7% and 1.3% of the broader China market.

 

US export controls have profoundly reshaped this landscape. The Trump administration's blanket ban on all AI GPU exports in April 2025 forced Chinese firms to turn to domestic chipmakers. Driven by this policy pivot, local GPU companies posted standout results: Biren Technology achieved gross profit of 557 million yuan in 2025, up 210.8% year-on-year with a gross margin of 53.8%, while Moore Threads grew full-year revenue 2.4 times to 1.505 billion yuan and narrowed its net loss from 1.618 billion yuan to 1.024 billion yuan. China's AI investment is projected to more than double to nearly US$27 billion by 2026, with over half directed at the hardware market; AI-related IT spending grew at roughly 27% annually from 2021 to 2026. The global server market is expected to sustain a 24.1% compound annual growth rate through 2029. Despite significant strides by domestic chipmakers, they still trail Nvidia and AMD by five to ten years in the AI data centre chip space. Even if the US were to allow Nvidia to export H200 chips to China, domestic efforts to steer demand toward homegrown semiconductors mean the company is unlikely to ever reclaim its pre-sanction 95% market share.

 

 

 

Murata Flags MLCC Price Increases as AI Server and Auto Demand Tighten Supply 

Industry sources suggest Murata Manufacturing is likely to outline its pricing strategy for multilayer ceramic capacitors when it reports earnings on April 30, with any decision on hikes and their magnitude expected after assessing first-quarter demand trends. Signs of tightening supply are already emerging in the MLCC market, particularly in high-value segments: lead times for high-capacitance and high-voltage MLCCs used in AI servers and high-reliability automotive components are stretching, with shortages appearing for select part numbers. As semiconductor performance continues to advance, demand for MLCCs in AI servers and automotive applications keeps climbing. That said, the demand recovery has yet to translate into broad-based price increases — while other passive components such as inductors and ferrites have seen adjustments, MLCC pricing remains relatively stable, with competitors watching closely for Murata's next move.

 

Looking back at the market cycle, MLCC prices rose by an average of roughly 23% during the pandemic-driven IT demand surge of 2020 to 2021, only to fall sharply in 2022 as demand weakened and inventory piled up, with some general-purpose parts retreating to pre-hike levels and prompting pushback from equipment makers that forced suppliers to lower utilisation rates. The MLCC market today is dominated by Japanese, Korean and Taiwanese manufacturers, and customers can flexibly switch orders between suppliers. Moreover, the heavy use of commodity MLCCs in highly price-sensitive applications such as smartphones and home appliances caps the scope for broad price increases. External factors including tariffs, geopolitical tensions and major elections in key economies could still weigh on downstream demand. Industry observers believe that while upward price momentum is building, a repeat of the sharp spikes seen previously is unlikely in the near term. A more realistic scenario is gradual average selling price improvement led by high-value segments such as AI servers and automotive applications — price hikes are a question of timing, but the pace is expected to remain moderate.

 

 

Yageo Tantalum Capacitors Run Tight, Walsin Resistor Prices Climb 30% 

Passive component giant Yageo said at its April 15 earnings call that second-quarter results are expected to outperform the first, with AI-driven demand continuing to expand across servers, data centres and high-performance computing, and order visibility now stretching into the second half. AI-related revenue already accounted for roughly 13% to 15% of total sales in the first quarter and is expected to climb above 15% for the full year. On the product front, tantalum capacitors, particularly polymer types, have become indispensable in AI applications due to their high capacitance density and stability, with book-to-bill ratios running well above the company average. In early March, Yageo subsidiary Kemet announced price increases on its T523 series polymer tantalum capacitors, effective April 1. Yageo noted that costs for raw materials such as tantalum and silver, as well as energy and logistics, have been rising steadily since 2025, and the company has been adjusting prices since the fourth quarter of last year on high-end AI server products, general distribution items and SSD-related applications.

 

Walsin, a resistor maker under Yageo affiliate Kaimei, issued a price hike notice to distributors last week, raising prices on high-precision metal plate resistors by 20% to 30% — Kaimei's second increase this year — citing rising raw material costs and robust demand from high-end power applications in AI servers, industrial controls, energy storage and renewables. The adjustments cover the LRx and LM series in sizes 1206, 2010, 2512, 2725 and 2728 below specified resistance values, along with the entire LRH range. Metal plate resistors are critical components in AI server power management systems, and demand has surged as server power levels rise and data centres adopt new 800V HVDC power architectures. Internal Kaimei data show Walsin's metal plate resistor lines have been running at full capacity since the second half of last year. These products now account for 40% to 45% of Kaimei's total resistor revenue, with the specific items being repriced representing roughly 15% of overall resistor sales.

 

Yageo group companies have raised prices multiple times since the second half of 2025: the parent company lifted high-end tantalum capacitors for AI servers and automotive electronics by 20% to 30% from late October 2025, raised RC0402 and RC0603 resistor prices by 15% to 20% from February 1 this year, and implemented its third tantalum capacitor hike on April 1. On the subsidiary side, Kemet raised tantalum capacitor prices in June and November 2025 before pushing through another 20% to 30% increase in early March 2026, while distribution arm Yageo International recently notified customers of a 10% to 20% resistor price increase. On MLCCs, Yageo noted that market supply and demand have gradually returned to healthy levels, with overall inventory down compared to the same period last year and no signs of double ordering or panic buying.

 

 

 

Manufacturer Updates

ST

Mixed demand, lead times vary

ST sees moderate overall demand but more need for discrete components. Spot demand focuses on power management ICs and EEPROM. For general MCUs, only STM32F series has light demand. Auto-grade demand is up. Lead times for most general parts improved to 12-18 weeks. Low-voltage MOSFETs extended to 13-26 weeks. 32-bit MCU lead times also grew to 15-28 weeks.

 

ADI

Steady demand, better price acceptance

ADI demand is mild. Customers are gradually accepting price hikes. Service and automotive models remain in strong demand.

 

 

TI

New price hikes, high-end analog shortage

TI started a new round of price hikes this month. Popular devices keep rising, while general parts are stable or even cheaper in some cases. Power devices in short supply include TPS, LM, INA, and Q1 (auto grade). Lead times are 20 to 40 weeks.

 

 

Broadcom

More AI deals, focus on networking & data centers

Broadcom announced major AI deals with Meta, Google, and Anthropic. Hot demand centers on networking and AI data centers. Key parts: BCM57508 (smart NIC), BCM56870A0KFSBG (high-performance Ethernet switch), SS24-0B00-02 and SS26-0B00-02 (switch chips).

 

 

Microchip

Long lead times for new products, general MCUs out 4-6 weeks longer

Microchip launched the dsPIC33AK256MPS306 digital signal controller. Lead time is 20-24 weeks. General 8/16-bit MCU lead times are 4-6 weeks longer than last year. AI power parts like dsPIC33AK and industrial gateway PIC32MK have lead times over 20 weeks.

 

 

Infineon

Full series price hike, 5% to 30%

Infineon raised prices across all IC series starting April 1. Mainstream parts up 5-15%, high-end up 15-30%. Strong demand from AI servers, solar storage, and auto electronics. AI server power and low-voltage MOSFETs see a surge in demand. Capacity is squeezed. Seller's market.

 

 

NXP

10-20% price hike, focused on auto and AI

NXP raised prices on some products by about 10-20% starting April 1. On April 3, it launched the TEF8388 3rd-gen radar transceiver. With the S32R4 processor, it supports 576-antenna imaging radar. Mass production has started.

 

 

Renesas

MCU and power device price hikes, long auto-grade lead times

Starting April 1, 2026, Renesas raised industrial and auto MCU prices. Overall up 3-8%, but some auto-grade power chips and high-end MCUs are up 10-20%. Dual power MOSFETs (e.g., UPA2756GR-E1-AT) and 100V mid-voltage MOSFETs (e.g., 2SK3483-Z-AZ) are tight, with estimated hikes of 10-15% or more. Auto MCUs (e.g., RH850, R-Car) have lead times of 16-24 weeks, some over 30 weeks. Transistors and thyristors exceed 32 weeks. Power management ICs (e.g., ISL9931, RAA270) are 24-38 weeks. Some auto-grade capacity shifted to AI, causing spot shortages. Auto chip market expected to grow 8% YoY in Q1 2026. Renesas sees strong demand in body control, battery management, motor drive, and ADAS.

 

 

onsemi

Over 10% price hike, lead times extend to 24-52 weeks

onsemi raised prices on many products including discretes, power management ICs, and image sensors. Hikes are generally over 10%. Demand is rising for power management ICs, low-voltage MOSFETs, and CMOS sensors. Lead times for diodes, transistors, and MOSFETs extended sharply to 24-52 weeks. Logic ICs are 26-32 weeks. Supply is expected to stay tight.

 

 

Xilinx (AMD)

High-end FPGA lead times 40-50 weeks, shortages into 2026

AMD high-end FPGAs like Versal, UltraScale+, and RFSoC remain in short supply. Lead times are 40-50 weeks or more. Mid-range supply is mixed. Low-end Spartan supply has improved. Strong demand from AI, 5G, and auto keeps high-end prices firm. Shortages are expected to last into 2026.
 

 

 

Spot Market Insights

Memory

eMMC / NAND Flash

• Pricing: Slight decline; bargain hunting persists, strong demand for 8G drives active trading. Limited downside expected in the near term.
• Supplier Dynamics: Kioxia/SanDisk supply extremely tight, prices sideways with minor softening but remain above Samsung; only sporadic offers seen.
• NAND Flash: Winbond MIXC official prices raised across the board; MXIC, Micron, Winbond, Kioxia all trending upward.
• NOR Flash: Winbond prices down, pressured by domestic competition.
• Supply: MXIC shipments low, severe shortages; ISSI offers scarce, agents near total booking halt, lead times significantly extended. Supply-side pressure remains.

 

 

DDR3 & DDR4

• DDR4 16G: Sideways, prices stable.
• DDR4 8G: Volume surged, low-cost supply gone; strong demand for Nanya and Micron, pricing firm. Samsung's ultra-cheap inventory depleted.
• Brand Spread: Nanya supply tight (limited shipment), pricing well above Samsung; DDR4 4G nearly double Samsung's level. Industrial/automotive grades especially short.
• DDR4 4G: Samsung showing early uptick, visible market sweeping.
• DDR3 4G: Grinding higher, nearly on par with Samsung DDR4; Nanya DDR3 4G now priced above other brands.
• Taiwan Manufacturers: DDR3/DDR4 prices climbing across the board, narrow inversion risk, active trading.

 

 

LPDDR

• Spot Market: Slight pullback, inventory holders more willing to sell.
• Micron: Prior overbought rally pushed prices well above official levels; holders now actively offloading, near-term correction pressure building.
• Samsung/SK hynix: Supply still scarce, trading relatively active, but automotive-grade shortage remains acute.
  

 

 

Server Memory (DIMM)

• Market: Overall demand soft, pricing steady to slightly weaker with minor declines on select models.
• Supply-Demand: Supply ample, sellers eager to move stock, but buyers firmly pressing for low prices. Spread widening, deals stalled.

 

 

 

Storage

HDD

• Price Trend: April upward cycle initiated; industry expects 10%–20% hikes. Upstream signal clear.
• Supplier Adjustments: Seagate and WD large capacity (≥16TB) up 10%–25% first; WD enterprise +10% this month. Seagate distribution now at new T1 pricing, +10%–30%. Toshiba/WD expected to raise 30%–35% from May; substantial cost pressure ahead.
• Demand Structure: Small capacity largely flat. ≥16TB spot up 10–10–20; recent demand centered on 16T/20T/24T, moderate overall turnover.

 

 

SSD

• Samsung : Largely sideways; some models ticking down. Traders still actively seeking S4520 2.5" 480G, PM9D3A 3.84T E1.S — pockets of strength remain.

• Solidigm: April new official pricing live, +15%–25%. Channel reports end-demand weak, shipments and deals low. Hikes met with thin demand, limited market absorption.

 

 

 

CPU

Mobile Platform

• Pricing: Intel across-the-board hike in April, with an additional 20% increase announced for May.
• Demand: Steady; prices firming through continuous deals. Buying interest focused on N97, N100, N150, C741 — supply tight on specific SKUs.

 

 

PC Platform

• Trend: Uptrend intact. Intel full-line April increase, May 20% hike planned. Cost pressure continues to flow into spot market.
  

 

 

Server Platform

• Official Pricing: Intel and AMD both +15%–20% in April; Intel reveals further 20% in May.
• Supply: Shortages remain severe; Intel server CPU gap persistent.
• Demand Focus: Inquiries centered on Intel 5th/6th Gen (4510, 6767P, 9474F, 6530) and AMD 3rd Gen.
• Market Spotlight: AMD 3rd/4th/5th Gen server CPUs dominate active trading; desktop non-suffix SKUs stable.
• Micro Moves: Intel 5th/6th Gen slightly firmer; AMD 9000 series high-end easing somewhat but overall supply tight — structural shortage persists
  

 

 

 

NICs & HBAs

NICs

• Overview: Demand stable; SI, distribution and other channels all carrying inventory. Average trading, prices steady.
• Broadcom HBAs: Demand normal, inquiries around 9500-8i, 9500-16i. 9560-8i, 9560-16i prices softening — likely seasonal Q2 slowdown and weaker end-demand pull.
• Mellanox: Demand soft, thin trading, low market attention.