Market Insights and Trends

 

Global Chip Sales Hit $791.7B in 2025, Poised to Top $1T in 2026 

Global semiconductor sales reached a record $791.7 billion in 2025, up 25.6% from $630.5 billion in 2024, according to the Semiconductor Industry Association (SIA). Fourth-quarter sales hit $236.6 billion, a 37.1% year-on-year surge, while December sales rose 2.9% month-on-month to $78.9 billion. Fueled by AI and other transformative technologies, the market is on track to exceed $1 trillion in 2026.

 

By region, Asia-Pacific and other regions saw sales jump 45.0% in 2025, followed by the Americas at 30.5% and China at 17.3%. By product category, logic chips led with $301.9 billion in sales, up 39.9%, while memory chips followed closely at $223.1 billion, a 34.8% increase.

 

AI and data centers are now the primary growth engines. The four major North American cloud providers are set to invest between $600 billion and $670 billion in AI infrastructure in 2026, up over 60% year-on-year. Analysts forecast global AI chip demand from AI servers to hit 16 million units in 2026. The memory market is a key beneficiary, with storage requirements 8 to 10 times higher than traditional servers. Contract prices for conventional DRAM are expected to jump 55% to 60% quarter-on-quarter in early 2026. On the supply side, top foundries project a 60% CAGR for their AI semiconductor revenue from 2024 to 2029. 

 

 

ECIA: Semiconductor Confidence Soars 24 Points, Over 70% Expect Q1 Growth 

A recent ECIA survey reveals a strong rebound in electronic component sales confidence for early 2026. The December sales confidence index hit 134.0, nearly 33 points above expectations, with the overall index climbing over 4 points to 138.3. This reversed earlier market doubts. The second half of 2025 marked the most optimistic period in over three years, with a June-December average of 121.5.

 

The semiconductor sector led the surge, with its confidence index jumping over 24 points to 147.6. This aligns with WSTS’s bold projection of 26.3% growth for the global semiconductor market in 2026, outpacing the 22.5% forecast for 2025. Categories beyond memory and logic ICs, including analog, microcomponents, discretes, optics, and sensors, are expected to see mid-single-digit growth, broadening the recovery.

 

Electromechanical and passive components also showed steady gains. The overall end-market index stood strong at 118.8 in December, with all categories scoring above 100 for the past three months. Over 70% of respondents anticipate positive sales growth in Q1 2026.

 

Supply chain pressures are mounting. In December, 33% of respondents reported longer lead times, up from 21% in November. For DRAM and NAND, 80% noted extended lead times, while other semiconductor categories saw figures between 44% and 58%, underscoring tightening market conditions.

 

 

 

 

 

Top 8 CSPs to Surpass $710 Billion in Combined 2026 Capex 

Global cloud service providers are ramping up AI infrastructure investments at an unprecedented pace. According to the latest report from TrendForce, the combined capital expenditure of the world's eight largest CSPs is projected to exceed $710 billion in 2026, marking a staggering 61% YoY growth.

 

Looking at individual spending plans, Google's parent company Alphabet is expected to lead the pack with over $178.3 billion in 2026 capex, a 95% annual surge. Driven by strong demand for TPUs from AI platforms like Google Cloud and Gemini, TPUs are forecast to account for nearly 78% of Google's AI server shipments in 2026. Google remains the only major CSP where ASIC shipments outpace GPU-based servers.

 

Meta is set to invest more than $124.5 billion in 2026, up 77% YoY, with Nvidia and AMD solutions continuing to dominate its AI server fleet—GPUs are expected to maintain over 80% share. Microsoft continues to procure Nvidia rack-scale solutions for its AI server deployments. Amazon has recently raised its procurement of Nvidia's GB300 and V200 rack systems, with GPUs projected to make up nearly 60% of its in-house AI server mix by 2026. On the ASIC front, its next-gen Trainium 3 is expected to begin ramping in Q2 2026, succeeding Trainium 2/2.5.

 

Oracle is steadily expanding its GPU rack deployments in support of projects like Stargate and OpenAI. Among Chinese CSPs, ByteDance is allocating over half of its capex to AI chip procurement, with Nvidia's H200 emerging as a leading candidate alongside expanded adoption of domestic alternatives such as Cambricon. Tencent continues to rely on Nvidia GPU solutions for cloud services while collaborating with local partners on ASIC development. Alibaba's T-Head and Alibaba Cloud are providing AI infrastructure, while Baidu plans to roll out its new Kunlun solutions starting in late 2026.

 

 

TI Acquires Silicon Labs for $7.5B to Boost IoT Wireless Connectivity  

Texas Instruments (TI) announced on February 4 its agreement to acquire Silicon Labs, an IoT wireless connectivity chipmaker, for $231 per share in an all-cash deal valued at approximately $7.5 billion. This marks TI's largest acquisition since its $6.5 billion purchase of National Semiconductor in 2011. Expected to close in the first half of 2027, the deal is projected to generate about $450 million in annual manufacturing and operational synergies within three years.

 

The acquisition aims to strengthen TI's leadership in embedded wireless connectivity, adding roughly 1,200 products supporting various wireless standards and protocols. Silicon Labs, known for its expertise in Zigbee, Bluetooth, and Thread, has achieved a compound annual revenue growth rate of about 15% since 2014. Post-merger, TI plans to integrate Silicon Labs' manufacturing into its own U.S. 300mm fabs and packaging facilities, leveraging optimized 28nm processes for cost-effective, large-scale production. 

 

 

SiTime Buys Renesas Timing Unit for $2.9B, Targets AI Data Centers 

Analog chipmaker SiTime has agreed to acquire Renesas Electronics' timing products division in a cash-and-stock deal valued at approximately $2.9 billion. Under the terms, SiTime will pay $1.5 billion in cash and 4.13 million of its own shares (valued at $347.96 per share). The cash portion will be funded from internal resources and a $900 million debt financing facility from Wells Fargo. The transaction, SiTime's largest to date, is expected to close by the end of 2026 and aims to bolster its timing solutions for key markets like AI data centers. 

 

 

Infineon to Hike Power Chip Prices in April as AI Demand Triggers Industry-Wide Rallies 

Infineon has issued a customer notice announcing price increases for its power switches and integrated circuits effective April 1, 2026. The move is driven by robust demand from AI data center deployments, leading to product shortages. To support growth, Infineon is making significant additional investments to expand wafer fab capacity, while grappling with rising raw material and infrastructure costs. The company forecasts moderate revenue growth for fiscal 2026 and plans to invest around €2.7 billion to accelerate capacity expansion, focusing on AI data center power solutions. Revenue from this segment is expected to grow from approximately €1.5 billion in the current fiscal year to about €2.5 billion by fiscal 2027.

 

This wave of price hikes, fueled by AI computing demand, is not isolated. Analog Devices (ADI) previously announced price increases across its entire product line starting February 1, 2026, with industry sources indicating an overall hike of around 15% and potential 30% increases for certain military-grade products. Texas Instruments had earlier raised prices on over 6,000 product models in August 2025, with increases ranging from 10% to 30%, over 40% of which exceeded 30%, primarily targeting industrial and automotive applications. The industrial automation sector is also affected, with Omron implementing price adjustments ranging from 5% to 50% on products like PLCs, sensors, and relays effective February 7. Since late 2025, several domestic Chinese chipmakers, including Goke Micro, Zhongwei Semiconductor, and Injoinic, have also issued price hike notices.

 

 

 

TrendForce: DRAM Contract Prices Surge Up to 95% in Q1, Up Nearly 6x in Six Months 

According to TrendForce's latest survey, AI-driven demand is spilling over into the broader memory market. CSPs are expanding their data center buildouts from AI servers to general-purpose servers, shifting procurement from HBM3e to RDIMMs of various capacities. This has triggered sharp increases in conventional DRAM contract prices. In Q4 2025, DRAM industry revenue grew 29.4% QoQ to $53.58 billion.

 

On ASPs, conventional DRAM contract prices rose 45%–50% QoQ in Q4 2025, while the blended average including HBM increased 50%–55%. Entering Q1 2026, despite the seasonal slowdown in consumer applications, CSPs are aggressively securing supply and remain flexible on pricing. As a result, conventional DRAM contract prices are expected to surge 90%–95% QoQ, with the blended average including HBM up 80%–85%—marking another acceleration.

 

By application, PC DRAM is set for a QoQ increase of over 100% in Q1 2026, the highest on record, driven by a seller's market. Server DRAM prices are forecast to climb around 90% QoQ, also a historic high, as buyers compete fiercely for supplier allocation. In the mobile segment, widening supply-demand gaps are expected to drive LPDDR4X and LPDDR5X contract prices up by approximately 90% QoQ in Q1.

 

On vendor performance, Samsung led in Q4 2025 with revenue of $19.30 billion, up 43.0% QoQ. Its ASP grew about 40% QoQ—the strongest among the three major players—boosting its market share back to 36.0% and reclaiming the top spot. SK Hynix posted $17.22 billion in revenue, up 25.2% QoQ, with relatively stable pricing due to its high HBM exposure. Micron reported $11.98 billion, up 12.4% QoQ. Among Taiwan-based players, Nanya Tech saw revenue surge 54.7% QoQ to $970 million, with ASP up ~30% and operating margin jumping from 6.0% to 39.1%. Winbond reported QoQ revenue growth of 33.7% to $297 million, with ASP up ~35% QoQ. Powerchip's DRAM revenue edged up 0.6% QoQ to $33 million.

 

 

DRAM and NAND Prices Hit Record Highs 

Market data shows the average fixed contract price for mainstream PC DRAM (DDR4 8Gb) hit $11.50 in January 2026, up 23.66% from the previous month—the highest since tracking began in 2016. Prices have climbed for 10 straight months since April 2025, with cumulative gains of 115% to 120%. Meanwhile, mainstream NAND flash (128Gb MLC) surged to $9.46, a staggering 64.83% monthly jump and the 13th consecutive monthly increase. The price gap between DDR5 and DDR4 modules has widened from 6% last quarter to 12%.

 

The core driver is a structural supply shortage fueled by AI demand. Memory suppliers are prioritizing high-value products like HBM and 3D NAND for AI servers, severely limiting supply for traditional products. Global cloud providers—Google, Meta, Amazon—and AI leaders like Nvidia have locked up most HBM capacity from SK Hynix, Micron, and Samsung, further straining conventional DRAM supply.

 

On the technology front, DRAM scaling has slowed dramatically, with density doubling over the past decade compared to 100x growth during its peak. The supply-demand imbalance is severe. Hyundai Motor Securities analysts note that current supply is about 10% below demand, and even a 5% gap could drive prices up 40% to 50%, with potential hikes of up to 2.5x before the cycle turns.

 

The shortage shows no quick fix. Micron executives expect global memory shortages to persist until 2028, while Synopsys' CEO sees tight supply extending through 2027. Citigroup analysts believe the current upcycle, starting in 2024, may outlast the 2001-2007 boom. Industry insiders suggest AI has kicked off a "quasi-supercycle" that could last over five years. Memory chip prices are expected to jump another 70% to 80% in Q1 2026, though gains may slow from Q2 onward. 

 

 

DDR4 Prices Plunge 20% in a Day, but BofA Lifts 2026 Outlook on AI Demand 

According to the latest Omdia data, the total wafer output in 2026 from the world's top three DRAM makers—Samsung Electronics, After a year-long rally, DDR4 prices saw a sudden sharp drop. Market data on February 12 showed 8GB DDR4 quotes falling from 260-270 yuan to 180-190 yuan, a nearly 20% single-day decline, with 16GB dropping from 800 to 650 yuan. This ended DDR4's "super bull run"—overall DRAM prices had surged 386% last year, with 16GB DDR4 doubling in just three months. However, Huaqiangbei traders noted limited actual pullbacks, with some models still priced at 1,800 yuan, attributing the dip to pre-Chinese New Year cash flow adjustments.

 

The root cause traces back to AI-driven demand. Major players like Samsung and SK Hynix slashed DDR4 capacity to chase high-margin HBM orders. Data shows DDR5 penetration in PCs and servers reached 70% to 80% in 2025, leaving DDR4 with under 30% market share. SMIC co-CEO Zhao Haijun warned that while AI boosts high-end products like HBM, it's shifting mid-to-low-end capacity to AI servers—a "demand transfer" rather than "total demand expansion." Foundry trends show sustained growth in AI and HPC-related chip orders, while consumer electronics and industrial control orders shrink.

 

According to a recent BofA report, DRAM spot prices saw their first pullback since September 2025 after 4-5 months of gains. Spot prices now significantly exceed contract levels, with Tier-1 OEMs still securing DRAM at $10-20 per GB. Despite short-term volatility, BofA has raised its mid-term outlook, boosting 2026 DRAM and NAND ASP forecasts by over 20%. Global DRAM sales are projected to jump from $134 billion in 2025 to $262 billion in 2026, a 95% year-on-year surge, with NAND climbing from $81 billion to $147 billion, up 82%, driven by 50-60% ASP increases. AI server demand remains the key driver, with the HBM market expanding rapidly and SK Hynix maintaining its lead through 2026-2027.

 

Industry experts believe DDR4's super cycle is likely over as DDR5 adoption widens, local supply expands, and speculative trading cools. Looking ahead, BofA sees a multi-year growth cycle for memory, with ASPs stabilizing from Q2 to late 2026 and a potential sub-10% correction in 2027. Short-term spot fluctuations haven't altered the long-term AI-driven upward trend. 

 

 

HBM4 Hits $700, Samsung and SK Hynix Enter Era of 30 Trillion Won Quarterly Profit 

Driven by surging AI demand and deepening shortages, high-bandwidth memory prices are skyrocketing. Industry sources reveal Samsung's recently mass-produced world-first HBM4 is priced at around $700—nearly 30% higher than SK Hynix's $550 Nvidia supply six months ago and 20-30% above the previous HBM3E generation. SK Hynix is expected to follow suit. With general DRAM profitability now rivaling HBM, memory makers have significantly strengthened pricing power.

 

Supply-demand dynamics continue to tighten. KB Securities data shows major clients' memory chip needs were only about 60% met as of February 2026. Citigroup forecasts 2026 DRAM and NAND demand growth at 20.1% and 21.4%, respectively, both outpacing supply growth of 17.5% and 16.5%. PC general DRAM prices are expected to jump 91% quarter-on-quarter in Q1, with server DDR5 up 99%. Smartphone memory prices have nearly tripled over the past year.

 

Soaring prices are driving record profits for Korea's two memory giants. Samsung posted its first-ever 20 trillion won quarterly operating profit in Q4 last year, with SK Hynix close behind at around 19 trillion won. Market consensus for Q1 2026 sees Samsung potentially reaching 32 trillion won and SK Hynix exceeding 28 trillion won, pushing their combined quarterly profit past 30 trillion won—ushering in Korea's "30 trillion won quarterly era." Full-year outlooks are equally bullish, with Morgan Stanley forecasting 2026 operating profits of 245.7 trillion won for Samsung (up 464% YoY) and 179.4 trillion won for SK Hynix (up 280% YoY).

 

Competitively, Samsung's early HBM4 mass production could secure about 30% of the global HBM market this year, with HBM sales tripling. AI accelerators featuring Samsung HBM4 are set to debut at Nvidia's GTC 2026 conference starting March 16. Despite massive capex announcements from the big three memory makers, new capacity can't fill the gap quickly due to construction timelines. Micron's $200 billion expansion plan won't deliver effective supply from new fabs until after 2028. Analysts broadly expect strong memory pricing to persist through 2027. 

 

 

Memory Capacity Race: Micron Commits $200B, SK Hynix and Samsung Ramp Up for AI-Driven Super Cycle 

Micron admits it can only meet 50% to two-thirds of key clients' demand. KB Securities data shows major customers' memory chip needs were only about 60% fulfilled as of February 2026, with server DRAM supply gaps below 50%. DDR5 chip prices have skyrocketed nearly 500% since September 2025, with shortages spreading from high-end HBM across the entire memory market. Citigroup predicts 2026 DRAM and NAND demand growth of 20.1% and 21.4%, outpacing supply growth of 17.5% and 16.5%. The supply crunch is widely expected to last through 2027.

 

Rising prices are boosting profit forecasts. Micron expects gross margins to hit 68% this quarter, nearing Nvidia's level. SK Hynix's chairman notes average analyst estimates for 2026 operating profit have climbed from $50 billion late last year to $70 billion, with some projections exceeding $100 billion. Bernstein Research sees Q4 2026 DRAM gross margins potentially reaching a record 77%.

 

Facing surging demand, the three memory giants are launching unprecedented capacity expansions. Micron plans $200 billion in U.S. investments, including $5 billion to expand its Boise headquarters with two new fabs—the first expected to start DRAM production by mid-2027, with full operations by end-2028. It's also building a $100 billion fab complex in New York and finalized a $9.6 billion plant in Japan.

 

SK Hynix has accelerated its Yongin fab construction, moving trial runs from May 2027 to February-March 2026. The facility will be six times larger than its Cheongju M15X fab. Its over 20 trillion won Cheongju M15X fab is already in trial production, targeting 10,000 wafers monthly initially, expanding to 55,000-60,000 by end-2026. Samsung plans to boost HBM4-related 1c DRAM capacity by about 170% in 2026 with a new line at its Pyeongtaek P4 fab, aiming for 100,000-120,000 wafers monthly by Q1 2027. In NAND, Samsung will expand 9th-gen V9 production at its Xian fab in Q2 2026, adding 40,000-50,000 wafers monthly, while SK Hynix targets 30,000 monthly wafers of 9th-gen 321-layer NAND at its Cheongju M15 fab.

 

On pricing, latest HBM4 quotes around $700 are 20-30% above HBM3E and nearly 30% higher than August 2025. SK Hynix dominates with about 60% of the HBM market, powering Nvidia's AI accelerators with 2026 capacity fully sold out and DRAM/NAND inventory at just four weeks. Samsung's early HBM4 lead could secure roughly 30% global market share this year.

 

Despite massive expansion plans, new capacity can't fill the gap soon. A Micron VP notes "never seen such disruptive demand surge," with 2026 supply growth limited to existing tech upgrades and new fabs not delivering until after 2028. SK Hynix's chairman warns rapid tech shifts could "easily turn into $100 billion losses." Bernstein expects even with 2027 price normalization, DRAM gross margins could hold near 62%, with HBM bit shipments growing 100% in 2026 and another 35% in 2027. 

 

 

SK Hynix Hikes DDR5 by Another 40%, Q2 Prices Set to Double 

SK Hynix and Samsung Electronics have issued updated price notices to clients, announcing sharp DRAM price increases for the second quarter. SK Hynix is raising DDR5 prices by an additional 40%, while Samsung—aside from major clients like Apple—is hiking contract prices by more than 100% for most buyers. Some smaller and mid-tier customers may need to accept increases of over 200% to secure supply. As of February 27, multiple memory module makers have suspended quotations pending cost recalibration.

 

Spot prices are surging in tandem. A mainstream 32GB DDR5 module is expected to jump from around NT$10,000 to NT$20,000. TrendForce reports that PC DRAM prices for DDR4 and DDR5 surged 105% to 110% QoQ in Q1 2026. According to CFM Flash Market data, the price of DDR5 16Gb eTT chips skyrocketed from $4.10 in August 2025 to $24.00 as of February 24, 2026—a nearly sixfold increase in six months. SK Hynix also noted that its combined DRAM and NAND inventory has dropped to just four weeks of supply, with all HBM capacity for 2026 already sold out.

 

The price shock is now reaching end-user markets. HP indicated that memory and storage costs will account for 35% of total PC bill of materials this year, nearly double last year's level. In the smartphone sector, major brands including OPPO, vivo, Xiaomi, and Honor are planning a new round of price adjustments starting in March. IDC forecasts the average smartphone selling price will rise to $465 in 2026, with global market revenue reaching a record $578.9 billion. Gartner predicts overall memory prices—including DRAM and SSD—will climb 130% by year-end, with the upward trend continuing at least through the second half of 2026. 

 

 

 

Omdia and Goldman: Memory Supply-Demand Gap Hits 15-Year Peak in 2026 

Omdia and Goldman Sachs report that that DRAM, NAND, and HBM will all face supply constraints in 2026, signaling the start of a prolonged demand-supply imbalance in the memory industry. According to Goldman Sachs, 2026 is set to be the tightest year for DRAM supply in 15 years, with a global supply deficit of 4.9%, followed by a 2.5% gap in 2027. The NAND market is projected to see deficits of 4.2% in 2026 and 2.1% in 2027, while HBM is expected to face even tighter conditions with gaps of 5.1% and 4.0%, respectively.

 

On the pricing front, DDR5 chip prices have surged nearly 500% since September 2025. Omdia forecasts NAND ASP to approach $0.15/GB in Q1 2026 (up 70% QoQ), and reach $0.18/GB in Q2 2026 (up 25% QoQ). Enterprise and client SSD contract prices have climbed over 60% QoQ recently, with some deals seeing doubled prices. NAND wafer prices rose by mid-single to double digits MoM in January. The NOR market is also on solid ground, with Omdia projecting revenues to recover to $3.54 billion in 2026 and hit $4.03 billion in 2027.

 

In terms of demand structure, servers have become the primary growth engine. A single AI server consumes 3 to 5 times more DRAM than a traditional server. Goldman Sachs estimates server DRAM demand (ex-HBM) will grow 39% in 2026 and 22% in 2027. Including HBM, server-related DRAM will account for 53% of global DRAM demand in 2026, rising to 57% in 2027. Enterprise SSDs are expected to make up 36% of total NAND demand in 2026 and 39% in 2027, with corresponding demand growth of 58% and 23%.

 

Nvidia’s Vera Rubin AI supercomputing platform is driving massive hardware demand. Each server features 72 GPUs, each requiring 16TB of SSD storage—totaling 1.152PB of NAND per system. A conservative estimate suggests that if 30,000 units ship in 2026, global NAND demand would increase by 2.8%; if shipments reach 100,000 units in 2027, that share could jump to 9.3%. 

 

 

Kioxia Inventory at 87 Days, SK Hynix at Four Weeks, WD and Seagate Sold Out for 2026 

The global memory market is facing an unprecedented crunch, from NAND to HDDs. Kioxia's latest report shows inventory days dropped from 100 to 87 by end-2025, under three months. Q3 revenue hit ¥543.6 billion ($3.55 billion), up 21.3% QoQ, with NAND ASP up ~10% and bit shipments in mid-single digits. SK Hynix's situation is even tighter, with DRAM and NAND inventory at just four weeks—a historic low—and expected to fall further. All customer demand can't be met, and 2026 HBM capacity is fully sold out.

 

HDDs are also in deep shortage. Western Digital says nearly all 2026 capacity is sold, mostly to top seven clients, with three hyperscalers signing deals through 2028. Clients are even asking about 2030 supply. Cloud revenue now accounts for 89% of total, with consumer at just 5%. Last quarter revenue hit $3.02 billion, net profit $1.84 billion (up 210% YoY), with a projected 20%+ CAGR over 3-5 years. Seagate's nearline capacity is fully booked for 2026, with Q2 FY2026 revenue at $2.83 billion (up 21.5% YoY) and gross margin at 42.2%. Clients are already discussing 2028 forecasts.

 

Kioxia expects Q1 2026 operating and net profit to at least triple QoQ. It sees 2026 NAND demand outstripping supply, with bit growth at 17-19%. AI servers remain key, using 3-5x more DRAM, with enterprise SSD share rising.

 

The AI-driven memory crunch is rippling across the industry. Consumer HDDs are expected to face severe shortages in 2026, with reports suggesting Sony may delay the next PlayStation to post-2027. Phison's CEO warns consumer electronics firms may struggle to survive without supply. Amid structural shortages, memory makers are prioritizing high-margin products, shifting the market from spot trading to "capacity pre-booking." 

 

 

AI Servers and Auto Electronics Ignite NOR Flash Shortage, Q2 Prices Seen Up 40-50% 

Driven by AI server and automotive demand, NOR Flash faces severe shortages and surging prices. Q2 2026 contract prices are expected to rise 40-50% QoQ, with lead times for high-density parts stretching to 12-14 weeks. Legacy SLC and MLC NAND are becoming scarce as major players exit, with Q2 contract prices seen up 400-500% YoY.

 

AI server upgrades are boosting NOR usage. A GPU rack can now hold up to 30 NOR devices, with Nvidia's GB200 NVL72 system costing $600 for NOR flash, potentially rising to $900. Per-server NOR demand has jumped from 125-256MB to 256-512MB, with Nvidia's B300 spec requiring 1Gb per chip, fueling demand for 512Mb and 1Gb parts.

 

Supply is constrained. Foundries are prioritizing higher-margin MCUs and PMICs over NOR. Samsung's June 2026 exit from legacy MLC NAND tightens supply further. Winbond and Macronix are key beneficiaries. Winbond aims to boost NOR capacity 30-40% YoY, with 2025-2026 fully booked. Macronix sees strong 2D NOR demand for AI servers, with its MLC eMMC prices doubling since December and NAND revenue share rising to 21%, more than doubling YoY.

 

Winbond is shifting its Taichung fab's ~58,000 wafers/month toward higher-margin NOR and SLC NAND, targeting ~50,000 wafers. Macronix's NAND expansion will delay its 3D NOR launch by two years. As AI servers evolve to HBM4 and auto electronics advance, structural NOR demand growth will continue. 

 

 

 

Intel and AMD Server CPU Supply Tightens: Lead Times Hit Six Months, Prices Up Over 10% 

Intel and AMD have recently informed customers in mainland China about severe shortages of server CPUs, leading to significantly extended delivery times. Wait times for some high-end Intel Xeon processors could stretch to six months, while delivery for certain AMD EPYC series has been pushed to 8-10 weeks. The tight supply has directly pushed up market prices. Due to allocation limits, Intel server product prices in China have generally risen by over 10%. Industry reports indicate both giants are considering a 10-15% hike in average server CPU selling prices for Q1 2026.

 

The shortage primarily affects high-performance lines. Intel's 4th and 5th Gen Xeon processors are particularly scarce in China. Despite implementing an allocation system, order backlogs remain heavy. Supply chain sources reveal a major server ODM that used to receive around 50,000 Xeon chips per quarter now sees its allocation drop to roughly 30,000. With China accounting for over 20% of Intel's revenue, the impact is significant.

 

The reasons behind the shortage are complex. On one hand, a surge in AI infrastructure investment is driving explosive demand for high-performance CPUs in traditional data centers. Hyperscalers like Meta, Google, and Microsoft are aggressively buying server CPUs to support agentic AI applications. On the other hand, Intel faces its own manufacturing yield challenges, with capacity ramping lagging behind demand. While AMD relies on TSMC for manufacturing, TSMC's advanced capacity is prioritized for higher-margin AI chips, limiting CPU production. Additionally, the memory price surge in late 2025 prompted customers to place early CPU orders to lock in costs, further tightening the market.

 

In terms of market share, Intel and AMD continue to dominate the global server CPU space. Intel's share has dropped from over 90% in 2019 to about 60% in 2025, while AMD's has climbed from roughly 5% to over 20% in the same period. Intel warned of the tight supply during its January earnings call, noting that Q1 inventory is expected to be at an all-time low, with improvement anticipated starting Q2 2026. AMD, meanwhile, reaffirmed confidence in its ability to meet global demand through its strong partnerships with TSMC and other suppliers. 

 

 

 

Nvidia and Meta Strike Multibillion-Dollar Deal, Locking In Millions of AI Chips 

Nvidia and Meta recently announced a multi-year strategic partnership focused on deep collaboration around AI infrastructure. Under the agreement, Meta will deploy millions of Nvidia Blackwell GPUs and next-generation Rubin architecture GPUs across its hyperscale data centers. The deal is estimated to be worth hundreds of billions of dollars, with analysts pegging the transaction value at up to $50 billion. A key highlight is the first large-scale standalone deployment of Nvidia's Grace CPU, making Meta the first company to deploy Grace independently in its data centers. The two are also collaborating on Nvidia's next-gen Vera CPU, targeting broad commercial availability by 2027.

 

On the networking front, Nvidia's Spectrum-X Ethernet platform will be fully integrated into Meta's Facebook Open Switching System, delivering predictable, low-latency performance for distributed AI clusters. The partnership builds on Meta's massive capital expenditure plans, with the company previously forecasting 2026 full-year capex between $115 billion and $135 billion, a significant portion of which will go toward expanding Nvidia-powered data centers. Nvidia's latest earnings show its top four customers account for 61% of its revenue, with Meta holding firm as the second-largest buyer.

 

Meta isn't putting all its eggs in one basket. The company is also pushing forward with its own AI chip development and considering adopting Google TPUs in 2027, while continuing to use AMD products—a reflection of the "second supplier" strategy major cloud providers are adopting to navigate tight supply. From a competitive standpoint, Meta's large-scale adoption of Nvidia's CPU solutions puts pressure on traditional server CPU suppliers like Intel and AMD. Analysts note that Grace CPUs offer clear energy efficiency advantages in certain back-end workloads, with next-gen Vera expected to deliver even greater gains. 

 

 

 

Global Passive Component Prices Surge as AI and EVs Drive Structural High-End Shortages 

The global passive component market is seeing another wave of sharp price hikes. Industry leader Yageo has raised prices on select resistor products by 15% to 20% effective February 1—its third increase in a short period—while its KEMET subsidiary hiked prices on high-end tantalum capacitors by 20% to 30%. Walsin followed with a roughly 20% increase across its full resistor line, and Panasonic raised prices on 30 to 40 tantalum capacitor models by 15% to 30%. In China, Fenghua Advanced Technology has increased ceramic capacitor prices by 10% to 20% and thick-film circuit prices by 15% to 30% since December, with Sunlord and others following suit. Smaller manufacturers are generally raising prices by 5% to 20%. Spot market supply is tight, with lead times for some high-end parts exceeding 20 weeks.

 

The core drivers behind this round of price increases are structural supply-demand imbalances and cost pressures fueled by AI and new energy vehicles. On the demand side, a single AI server uses 3,000 to 4,000 MLCCs—more than double that of a traditional server—with one AI server rack requiring up to 440,000 units. Murata forecasts that AI server MLCC demand in 2030 will be 3.3 times higher than in 2025. For new energy vehicles, MLCC usage per vehicle has surged from around 3,000 in conventional cars to between 18,000 and 30,000, while resistor demand is more than triple that of traditional vehicles. On the supply side, surging raw material costs are adding pressure: NYMEX silver futures rose over 140% in 2025, and silver paste accounts for more than 50% of the cost of multilayer inductors and beads. Rising prices for metals like copper and nickel have pushed overall production costs up by 20% to 30%.

 

The market is showing a clear divergence between strong high-end demand and weak consumer demand. Benefiting from AI and automotive orders, Japanese and Korean manufacturers such as Murata and Samsung Electro-Mechanics are maintaining utilization rates above 80% for high-end MLCCs. Murata expects Q1 2026 orders for high-end MLCCs to grow 20% to 25% quarter-on-quarter. In contrast, the consumer electronics sector saw some ODM orders for MLCCs drop by an average of 5% to 6% month-on-month in January, with related manufacturers keeping utilization rates between 60% and 70%.

 

This trend presents an opportunity for Chinese manufacturers to capture market share through localization. China is the world's largest passive component market, with a 2024 size of 142.3 billion RMB, accounting for approximately 44% of the global share. Fenghua is investing 5 billion RMB to expand its high-end capacitor production base, which will add around 45 billion units of monthly high-end MLCC capacity upon completion. Sunlord's 01005-size inductors have entered Nvidia's supply chain, with related revenue up 73% year-on-year. Analysts note that this round of price increases is driven by structural growth led by AI and automotive electrification, and expect the tight supply-demand balance and upward price trend for high-end passive components to continue through 2026. 

 

 

MLCC Becomes New AI Hardware Bottleneck as Top Japanese and Korean Makers Run Full, Spot Prices Jump 20% 

The global MLCC market is entering a major price hike cycle, emerging as the latest constraint in the AI hardware supply chain following memory chips. Distributors in China have already raised spot prices, with mid-to-high capacitance, automotive, and industrial-grade MLCCs climbing 10% to 20%. Market leader Murata has initiated internal reviews to raise prices on its most advanced products, with its president revealing that current orders for high-end MLCCs have reached twice the available capacity. Samsung Electro-Mechanics, the second-largest player, plans to implement double-digit price increases starting in April. Its Tianjin facility, with a massive monthly capacity of 120 billion units, is already running at full utilization.

 

The core drivers behind this price surge are structural demand growth and cost pressures. On the demand side, a mainstream GB300 rack requires at least 450,000 MLCCs—nearly 300 times the roughly 1,500 units used in an iPhone. AI power rack systems have seen MLCC usage per unit jump from 2,200 to 30,000. Murata expects the AI capex boom to last at least three years; its computing segment posted 26.5% year-on-year revenue growth last quarter, with MLCC utilization rates hitting 90% to 95%. New energy vehicles use 10,000 to 15,000 MLCCs per unit, three to six times more than traditional combustion engine vehicles. Tantalum capacitors, already in short supply due to heavy adoption in AI servers, have seen early price hikes of 20% to 30%, pushing designers toward "tantalum + MLCC" hybrid solutions and further amplifying MLCC demand. On the cost side, silver accounts for 42% to 58% of MLCC production costs. International silver prices have risen sharply since Q3 2025, while copper and nickel prices climbed 27% and 19% respectively in 2025, compounding manufacturing cost pressures.

 

The market is showing a clear divergence between strong high-end demand and weak consumer demand. Japanese and Korean manufacturers are maintaining utilization rates above 80% for high-end MLCCs, with Murata expecting Q1 2026 orders to grow 20% to 25% quarter-on-quarter. In contrast, the consumer electronics sector saw some ODM orders drop by 5% to 6% month-on-month in January, with related manufacturers keeping utilization between 60% and 70%. Industry-wide inventories are at historic lows, with lead times for high-end components exceeding 20 weeks and significant short-term supply gaps evident. Industry insiders expect spot MLCC prices to rise further by 15% to 25% over the next one to three months, with the price hike cycle likely extending through the first half of 2026. Murata's shares jumped 6.9% in a single day after bringing price adjustments into the open. As Japanese and Korean manufacturers run at full capacity, spillover effects are expected to gradually impact the mid-to-low-end market. Driven by AI and automotive electrification trends, the tight supply-demand balance for high-end MLCCs is likely to persist. 

 

 

 

Manufacturer Updates

STM

General-part lead times Ease to 12-18 Weeks, Auto Lines Hold at 36-40 Weeks

Supply availability for standard STM32F/L series and select analog components has notably improved. This increased inventory is intensifying price competition in the market. In contrast, the automotive sector stays tight, sustaining strong demand from industrial and automotive applications.

 

 

Renesas

Industrial and automotive MCUs remain tight, supporting firm spot pricing

Supply constraints for Renesas' industrial and automotive microcontrollers show no signs of easing, with lead times staying extended. Although no broad price hikes have been announced, cost pressures across the supply chain are keeping spot prices firm, signaling potential near-term upside. Meanwhile, SiC supply stability and the company’s long-term power strategy are drawing increased market attention.

 

 

TI

Mainstream model prices up 5-15%; lead times extend to 12-20 weeks

TI's spot market was active in February. Driven by January price adjustments, tight 8-inch fab capacity, robust demand from AI servers and automotive, and pre-Chinese New Year inventory builds, prices for mainstream models rose 5-15% month-on-month, with some scarce parts seeing larger gains. Channel inventory is low, with lead times generally extended to 12-20 weeks. The market anticipates a restocking wave post-holiday, sustaining strong demand in automotive, industrial, and AI sectors, reinforcing a seller's market.

 

 

ADI

Sees Post-Price Hike Demand Growth; Select Vendors Offer Flexibility

After recent price adjustments, ADI has seen growth in both demand and orders this month, with market prices trending upward. However, feedback suggests some suppliers remain willing to offer competitive pricing support.

 

 

Broadcom

Secures $21B AI chip deal; server expansion cards remain scarce

Broadcom has signed a $21 billion hardware procurement agreement with AI firm Anthropic to supply approximately 400,000 Google TPU v7 "Ironwood" chips. This move marks its first major foray into AI accelerator sales. Concurrently, expansion cards for server products, such as the SX03, SX06, and SX09 series, remain critically scarce in the spot market, commanding high premiums with no immediate relief expected.

 

 

Microchip

Demand remains sluggish; lead times lengthen for automotive devices

Demand remained weak this month, with a quiet spot market seeing only scattered inquiries for ATSAM series MCUs. Lead times for general-purpose 8/16-bit MCUs are 6-12 weeks, but automotive-grade PIC devices and related series face longer waits. Ethernet and USB product lead times are 8-16 weeks, while FPGAs have stretched to 12-40 weeks. Overall, automotive-grade components are significantly tighter than general-purpose parts.

 

 

onsemi

GF fab acquisition closed; GaN samples expected in H1

onsemi finalized the acquisition and integration of the former GlobalFoundries 300mm wafer fab in New York in January. The facility will focus on automotive and industrial power semiconductors and image sensors. Simultaneously, onsemi is deepening its partnership with GF to co-develop 650V GaN power devices, with first engineering samples expected in H1 2026. These moves are key to its strategy to quickly scale GaN capacity and strengthen its position in EV charging and AI data center power markets.

 

 

Infineon

Power Devices See Spot Prices Surge Up to 50% Ahead of April Hike

Driven by AI demand and a structural shortage in power discretes, combined with Infineon's upcoming April 1 price adjustment, spot market supply is tightening and prices are rising. The increase covers power switches, power ICs, and PMICs. Mainstream products are up 5-15%, while automotive IGBTs, SiC modules, and high-end MOSFETs see increases of 10-20%, impacting new orders and existing backlogs.

By February, spot prices showed significant volatility: automotive IGBT modules (e.g., FF450R12ME4) rose 20-35%, SiC modules (e.g., IMZ120R045M1H) jumped 30-50%, with some high-end parts unobtainable. Spot prices for industrial and AI server high-voltage MOSFETs and high-density PMICs increased 15-40%, while low/medium-voltage general-purpose MOSFETs tightened, rising 10-25%.

On lead times: Standard automotive IGBT/SiC modules are at 16-20 weeks, with high-end versions at 24-30 weeks; some critical parts are no longer accepting orders. AI server power device lead times are 20-50 weeks, with high-density power management ICs exceeding 52 weeks. Industrial and general-purpose device lead times are 12-18 weeks, but supply is tightening. Inventory among authorized distributors for key tight models is critically low at 1-2 weeks. Premiums for tight spot parts are high, while general-purpose supply is adequate but prices are nudged up by expectations.

 

 

NXP

Demand stabilizes, inventory tightens; S32N launch and MEMS sale finalized

NXP's overall demand remains stable, but decreasing inventory is causing structural shortages in some areas. The company launched its S32N series processors for software-defined vehicles and new EV battery monitoring ICs, strengthening its automotive portfolio. The sale of its MEMS sensor business line was completed in early February.

 

 

Xilinx

High-end FPGA lead times top 40 weeks; XC7 series spot prices rise

The ongoing supply challenges for high-end FPGAs show no signs of easing. Xilinx has extended lead times for its automotive-grade and Versal series to more than 40-52 weeks. In parallel, spot prices for the widely used XC7A and XC7Z series continue to climb, driven by lengthening lead times and sustained demand. The trend underscores persistent and severe supply limitations across the high-end FPGA segment.

 

 

Passive Components

High-end MLCCs tight; Yageo leads resistor price hike of 15-20%

In the passive component space, surging AI server demand, coupled with top-tier manufacturers shifting capacity towards high-end products, has led to sustained tightness for high-temperature, high-capacitance MLCCs. This strain is impacting other series, generally extending lead times. Under cost pressure, a wave of price increases has arrived. Yageo announced a 15-20% price hike on select resistor products effective February 1st, with major players like Samsung Electro-Mechanics, TDK, and Murata also following suit.

 

 

 

Spot Market Insights

Memory

eMMC / NAND Flash

• Before and after the Lunar New Year, some spot sellers did a small amount of selling to ease cash pressure. But after the holiday, customer demand stayed strong, and the supply-demand gap hasn't improved. The market is still seeing price hikes and shortages. 
• Prices for all brands and capacities are going up, and spot goods remain limited.
• Distributors are holding onto stock and are not willing to sell at low prices. Kioxia and Sandisk eMMC parts are very scarce, with quotes staying high. Some models are even priced above Samsung, and only a few offers are available. There's a rumor Kioxia plans to cut production by half in 2026, adding to supply worries.
• Winbond has increased official prices across the board and sometimes stops taking orders or releasing goods. 
• At the same time, small-capacity Flash from brands like MIXC, Micron, Winbond, and Kioxia are all seeing price hikes. Market quotes are messy and price swings are getting worse.

 

 

DRAM

• DDR4:  Consumer DDR4 is mostly in a holding pattern, with prices for some densities (like 4Gb–16Gb) a bit soft.
• DDR3: Due to high prices, some cost-sensitive buyers and applications are shifting demand back to the lower-priced DDR3, leading to a quick rise in inquiries.
• LPDDR: LPDDR is still in short supply, especially LPDDR5 and LPDDR5X. This tight supply situation is expected to continue through 2026. Major brands are controlling shipments tightly, so very little stock reaches the channel. Distributors are hoarding stock, and the price increases are bigger for higher capacities. Currently, only Micron has a small amount of stock available. Other brands have almost no new supply. The market is in a state where there are prices, but hardly any goods to buy.

 

 

Server DDR4 & DDR5

• RDIMM prices are stable and moving sideways. PC DRAM spot goods are very scarce, with quotes up slightly. DDR5 demand remains hot, and server-side prices haven't dropped.
• The PC market is held back by slower end-product shipments and higher memory prices, which is limiting the increase in memory per device. Demand growth has basically stopped.

 

 

 

Storage

HDD

• The top three HDD makers (Seagate, WD, Toshiba) have announced their 2026 production capacity is sold out. Only 3%-5% is for the consumer market, signaling a serious shortage for consumer HDDs this year. Retail supply is under clear pressure.
• Enterprise market prices (8TB/16TB/20TB/24TB) keep climbing, and consumer market prices (1TB/2TB) are following suit.
• Although overall demand is just so-so, prices are still trending up, and lead times keep getting longer. 8T/16T/20T/24T remain hot items. Seagate 24T prices edged up slightly last week, and supply is tight.

 

 

SSD

• Samsung SSD prices are mostly stable, but the 960GB model is seeing price increases due to shortages. 

• Price hikes for other models are slowing down. Market inquiries are active, but actual deals are few, with many waiting to see how things develop.

 

 

CPU

Mobile Platform

• Demand is steady but prices keep rising. The N-series is the most active, with the N97 becoming the hottest model, seeing significant price hikes. 

 

 

PC Platform

• Tray prices are mostly flat. Demand is weak, and trading activity is low.

 

 

Server Platform

• Demand is at normal levels, mainly for 5th and 6th Gen products. There are small-volume deals happening.

 

 

GPU

GPU

• Overall prices are moving up strongly. High-end models and those with large memory are seeing big premiums. 
• Nvidia hasn't changed official prices, but AIC partners and end-user prices have generally gone up by about 15%. 

 

 

Network & Storage Controllers

Network Cards

• After the Lunar New Year holiday, factory demand is slowly releasing. Broadcom RAID and HBA card prices keep rising. Demand for MCX series has dropped due to more stock arriving.
• Mainstream models like the 9560-8i and 9500-16i have seen price increases of over 10% after the holiday, and the trend looks bullish. There is also some small-volume trading for Broadcom RAID cards like the 9580-8i8e.

 

 

Passive Components

MLCC