As SK hynix pushes ahead with HBM4 production, attention is turning to TC bonder suppliers. According to The Elec, citing sources, SK hynix has placed a new order for thermal compression (TC) bonders with Singapore-based ASMPT to support HBM4 manufacturing. The order reportedly comes amid an ongoing patent dispute between SK hynix’s other South Korean TC bonder suppliers, Hanmi Semiconductor and Hanwha Semitech.

Regarding the order size, the report says, citing sources, that SK hynix placed an order last month with ASMPT for seven TC bonder systems, each equipped with two bonding heads. With each unit priced at about 4 billion won, the total contract value is estimated at roughly 30 billion won.

The report says SK hynix is likely to place an order for roughly 100 TC bonder units in March next year for HBM4 production, as it moves to expand 1b DRAM capacity at its M15X facility in Cheongju. How those orders will be split among ASMPT, Hanmi Semiconductor, and Hanwha Semitech remains to be seen.

Currently, SK hynix is using about fifty sets of TC bonders for HBM4 production, with roughly half supplied by ASMPT, according to the report, citing sources. Hanwha Semitech’s bonders are not currently in use, the report adds.

ASMPT has already secured a separate order in October last year for dozens of TC bonder systems for HBM3E 12-high production, which SK hynix says helped enable the successful mass production of HBM3E, as the report indicates.

Patent Battles Complicate the TC Bonder Landscape

As the report explains, some sources believe ASMPT secured the recent order partly due to the ongoing dispute between rivals Hanmi Semiconductor and Hanwha Semitech, while others say the order had been expected earlier this year but was only finalized recently. In April, Hanmi recalled around 60 engineers dispatched to SK hynix as support staff—a move widely viewed as a protest against the chipmaker’s decision to source equipment from Hanwha.

The report notes that Hanmi Semiconductor filed a patent lawsuit against Hanwha Semitech in December last year. Hanwha responded in May by seeking to invalidate the patents and filed a countersuit for patent infringement two months ago.

Hanmi had been SK hynix’s exclusive TC bonder supplier for HBM production until last year, when ASMPT joined the supply chain toward year-end. Hanwha Semitech entered earlier this year, the report adds.

Meanwhile, according to The Korea Herald, Hanwha has recently established a strategic unit to explore mergers and acquisitions for businesses under Vice Chairman Kim, raising the prospect of M&A in collaborative robotics and semiconductor equipment. Hanwha Semitec competes with Hanmi Semiconductor for leadership in TC bonders, and analysts suggest M&A may be needed to support growth, the report notes.

Read more

• [News] HBM4E Seen Hitting 40% of 2027 Market; Samsung, SK hynix Reportedly Aim for 1H26 Completion
• [News] HBM Hybrid Bonder Demand Reportedly Rising in 2H25, as BESI and ASMPT Eye Growth

(Photo credit: ASMPT’s LinkedIn)

As a key beneficiary of Google’s TPU boom, U.S. chipmaker Broadcom’s ASIC business is drawing heightened attention. On its latest earnings call, despite Broadcom usually keeps its major clients under wraps, CEO Hock Tan revealed that the company’s $10 billion custom-chip megadeal came from AI lab Anthropic, according to CNBC.

Barron’s, citing Tan, adds that a fifth AI-chip customer has joined Broadcom’s roster, placing a $1 billion order slated for delivery in late 2026.

CNBC reports that in its September earnings call, Broadcom disclosed a $10 billion custom-chip order from an unnamed client—sparking widespread speculation that the buyer might be OpenAI. However, the report notes that Tan now clarified that it was actually AI lab Anthropic, which ordered the latest TPU Ironwood racks and added another $11 billion order in Broadcom’s most recent quarter.

CNBC explains that Broadcom produces custom ASIC chips, while it helps Google manufactures the search engine giant’s TPUs, and last month Google highlighted that it trained its advanced Gemini 3 model entirely on these TPUs.

Margin Risks Mount Amid AI Revenue Shift

However, Reuters reports that while Broadcom holds a $73 billion backlog set to ship over the next 18 months, CFO Kirsten Spears cautioned that margins could slide. She noted that Q1 consolidated gross margin is expected to drop about 100 basis points sequentially, mainly due to a larger share of AI revenue.

Analysts cited by Reuters highlight two key risks: the company’s heavy concentration among just five AI customers, and the lower margins on AI systems—which are expected to comprise a larger share of total revenue in coming quarters, particularly in the second half of fiscal 2026.

According to CNBC, Broadcom topped fourth-quarter estimates with adjusted EPS of $1.95 (beating $1.86 expected) and revenue of $18.02 billion (above $17.49 billion consensus), fueled by AI demand. The company projected Q1 revenue at approximately $19.1 billion—28% above last year and exceeding the $18.3 billion analyst forecast. The report, citing CEO Hock Tan, adds that the company’s AI chip sales are expected to double year-over-year to $8.2 billion this quarter across custom chips and networking semiconductors.

Read more

• [News] Meta Reportedly Weighs Google TPU Deployment in 2027, Boosting Broadcom, Taiwan’s GUC

(Photo credit: Broadcom)

Samsung Electronics has rapidly strengthened its partnership with Tesla after securing the automaker as a major foundry customer. According to South Korea’s JoongAng Ilbo, citing sources, Samsung Chairman Lee Jae-yong met with Tesla CEO Elon Musk while visiting Samsung’s Taylor, Texas fab during his U.S. trip.

As the report notes, Samsung’s Taylor fab—constructed with an investment of more than KRW 50 trillion—is close to beginning operations, with the meeting strengthening its partnership with Tesla, now one of its largest foundry customers. The two companies are said to have conducted in-depth discussions on production lines, focusing on yield stabilization and future technology collaboration.

Hands-On Collaboration at the Taylor Fab

Citing sources, the report notes that Musk was granted a dedicated workspace inside Samsung’s Taylor fab, effectively allowing Tesla to maintain an on-site presence to oversee chip manufacturing. The move highlights a collaboration that extends beyond a conventional customer–supplier arrangement.

Industry experts cited by the report say the development marks an experimental shift in the foundry industry, with customers taking part across the entire process, from chip design and fab construction to production-line layout and packaging, thereby accelerating feedback cycles. The report also notes that Tesla’s Austin headquarters is located just 45 minutes from Samsung’s Taylor facility.

Samsung Prepares to Close the Gap with TSMC

Notably, the report says industry observers expect Samsung’s foundry business to intensify its push against TSMC from mid-next year, once the Taylor plant is fully completed.

The report adds that Samsung’s foundry business could raise utilization at its U.S. manufacturing base through its partnership with Tesla, while also improving its standing in the U.S. market, potentially paving the way to win additional major customers such as NVIDIA and AMD.

Samsung has also stepped up hiring in the U.S. to strengthen foundry yield solutions. According to SEdaily, citing sources, Samsung Device Solutions’ Americas division (DSA) is recruiting experienced talent for Customer Engineering (CE) roles, which focus on resolving technical issues and optimizing processes just before chips move into mass production. The hiring suggests Tesla is entering the final stages of preparation for mass production of key products, including the AI5. Industry sources estimate Samsung’s 2nm process yield has currently risen to around 55–60%, the report adds.

Read more

• [News] Tesla AI5 Production Split Between Samsung, TSMC; Musk Cites Samsung’s “Advanced Equipment”
• [News] Samsung Reportedly Speeds Up U.S. Hiring as Tesla Prepares for AI5 Production

(Photo credit: Samsung)

One of China’s key Apple suppliers is making an unexpected push into the AI era. Consumer electronics giant Lens Technology plans to acquire 100% of PMG International, according to Commercial Times and Anue, giving it indirect control of Yuans Technology and access to server-rack and liquid-cooling capabilities—opening a pathway into NVIDIA’s supply chain.

The reports note that Lens Technology announced late on the 10th that it intends to purchase all shares of PMG International, thereby securing roughly 95.12% ownership of Yuans. Founded in 2003 and based in Jiaxing, Zhejiang, the company specializes in metal components, mold manufacturing, and related technical services, as per the reports.

Notably, Anue reports that Yuans supplies rack enclosures for NVIDIA’s GB200 and GB300 platforms and holds both AVL (Approved Vendor List) and RVL (Recommended Vendor List) certifications directly from NVIDIA.

The report adds that as Yuans’ other shareholder is Taiwan-based Pintec Technology, with both companies sharing the same chairman, Lu, this cross-company connection suggests that Lens Technology’s acquisition would deliver a dual benefit: immediate access to Yuans’ NVIDIA-certified rack products, plus potential integration of Pintec’s liquid-cooling technologies.

Pintec Technology also boasts an impressive client lineup. According to Anue, the company serves major global server and system brands, currently supplying thermal modules to AMD, NVIDIA, and Foxconn for AI chips and servers, including NVIDIA’s GB200 and GB300 platforms.

Lens Technology’s Ambition

According to nbd.com, Lens Technology’s planned acquisition of an AI-server supply-chain player marks a new expansion beyond its core consumer-electronics business. As one of Apple’s key suppliers, the company known for its significant market share in touch sensor glass panels manufactures a wide range of products for the Cupertino giant, including the iPhone, Apple Watch, Mac, and Apple Vision Pro, the report adds.

However, in recent years, Lens has also been working to reduce its reliance on Apple by pursuing new growth drivers, extending into AI glasses, robotics, and smart vehicles. According to the report, in February, it became the full-device assembly partner for China’s domestic AR brand Rokid. In robotics, active since 2016, the company now reportedly covers liquid-metal materials, six-axis force sensors, head and joint modules, and dexterous hands.

The company is also making a strong push into the AI server space. According to Securities Times, Lens Technology said at a November investor briefing that it is building out its AI computing infrastructure, starting with server chassis components and expanding into liquid-cooling modules and SSD assemblies. Chassis parts are already shipping, with SSD assembly set to reach scale in 2026, the report notes.

Read more

• [News] NVIDIA H200 Export to China Approved, Unlocking Gains for Taiwan OEMs, OSATs, and Server Suppliers

(Photo credit: NVIDIA)

As major tech companies race to define the future of smart wearables, Apple’s long-rumored smart glasses are drawing increasing attention. According to 9to5Mac, citing sources, Apple Glasses are expected to be unveiled in 2026, though shipments may not begin until 2027, with rumored features including a built-in camera, Apple Intelligence–related functions, voice interaction via Siri, and more.

As 9to5Mac notes, Apple is now said to be developing an AI-focused “Apple Glasses” device—one that does not include AR features—driven in part by the broader AI trend and Meta’s momentum with its Ray-Ban lineup. The first model is expected to be without a built-in display, while Apple is reportedly working on a higher-end version that will feature a display, the report adds.

Further details suggest the glasses will use an S-class chip from the Apple Watch, designed for heightened power efficiency and to manage multiple cameras enabling visual-intelligence features similar to those on the iPhone, as 9to5Mac indicates.

The report also highlights that the revamped Siri will play a central role in the Apple Glasses experience. In addition, the report indicates that Apple is expected to offload some processing tasks to a paired iPhone, since the glasses will be positioned as an iPhone accessory, much like the Apple Watch.

The AI smart-glasses market is poised to intensify. Reuters notes that although Ray-Ban Meta glasses currently dominate the AI eyewear space, their first-mover advantage may diminish as competitors roll out more products. According to Reuters, Alibaba launched its new Quark AI-powered glasses in China in November—one of the few markets where Ray-Ban Meta is not available. Google is also developing its own model in partnership with Warby Parker and luxury group Kering, with its first product expected in 2026, the report notes.

Meta Reportedly Resets VR Pricing Strategy Amid Product Delay

Meanwhile, as another major player in smart wearables, Meta is reportedly planning to increase prices for its virtual-reality devices, according to an internal memo reviewed by Business Insider. The memo indicates that the company intends to position future devices at higher price points and may shift to a slower hardware release cycle. The company’s flagship VR headset, the Meta Quest 3, is priced at $499.99, while its entry-level model sells for $299.99, Business Insider adds.

The move comes as Meta has reportedly delayed the launch of its new mixed-reality glasses, codenamed “Phoenix,” shifting the release window from the second half of 2026 to the first half of 2027, as Business Insider highlights.

Read more

• Global Tech Giants Accelerate Investment, AR Glasses Shipments Projected to Reach 32.1 Million Units by 2030, Says TrendForce
• Meta’s New AR Device to Lift LCoS Share to 13% by 2026, Intensifying Competition with LEDoS, Says TrendForce

(Photo credit: Apple)

Japan’s Kioxia is making a strategic push at a time when major memory makers remain cautious about expanding NAND capacity. According to Nikkei, as cited by MoneyDJ, the company plans to begin mass production of its 10th-generation NAND at its Kitakami plant in Iwate Prefecture in 2026—aiming squarely at surging demand from AI data centers.

Although the move may appear to be a technological leap, with Kioxia’s 10th-generation NAND adopting a 332-layer stack versus the 218 layers in its current 8th-generation lineup, the company is reportedly taking a more pragmatic approach on the manufacturing side. Instead of building a new fab, Kioxia is expected to repurpose its recently opened Kitakami K2 facility—which began operations in September—to produce the new generation, MoneyDJ reports.

In terms of details on Kioxia’s 10^th-gen NAND Flash, Tom’s Hardware reported in February that Kioxia’s 10th-generation NAND extends the company’s CMOS directly Bonded to Array (CBA) technology—first introduced in 8th-gen products—with substantial performance upgrades.

The new chips, according to the report, feature 332-layer stacking, up 38% from the previous 218 layers, enabling 59% higher storage density. Data transfer speeds reach 4.8 Gb/s, a 33% improvement over the prior generation.

TrendForce observes that the Kioxia/SanDisk consortium—lacking DRAM businesses—is leading with the most assertive strategies to enhance their market positions. Kioxia/SanDisk plans to increase investments by 41% year-on-year to $4.5 billion in 2026, focusing on expanding BiCS8 production and supporting BiCS9 R&D, according to TrendForce.

Read more

• [News] U.S., Japan Rumored Mulling Joint NAND Fab, with Kioxia, SanDisk Likely In
• [News] Kioxia Quarterly Profit Tumbles 62% YoY, But Signals Bottom with Record 3QFY25 Sales Forecast

(Photo credit: Kioxia)

TSMC is reportedly reassessing its strategy in Japan amid surging AI demand. According to Nikkei, the company is considering shifting its second Japan fab toward producing more advanced chips than originally planned. As the report highlights, TSMC is weighing a move to 4nm production at the second plant—a change that could require design adjustments and potentially delay the project. The plant had initially been slated to produce 6nm and 7nm chips.

Construction on the second Kumamoto facility has already been put on hold, the report adds. Located in the same prefecture as TSMC’s first Japan fab, the plant is scheduled to begin operations in 2027.

TSMC Rethinks Japan Strategy Amid Shifting Demand

As the report explains, demand for 6nm and 7nm chips has weakened since TSMC announced the second Japan plant in 2024. Utilization at the company’s main Taichung fabs has fallen below desirable levels as major customers—including NVIDIA, Apple, Google, and Amazon—migrate to more advanced process technologies.

Meanwhile, alongside its potential shift to 4nm, sources cited in the report say TSMC is also considering introducing advanced packaging technology in Japan, given its importance for AI processors. None of these plans have been finalized, the report notes. It further adds that NVIDIA’s latest Blackwell chip is built on TSMC’s 4nm process and paired with HBM using the company’s CoWoS packaging.

In addition to the possible changes for its second Kumamoto fab, the report, citing sources, also notes that TSMC is continuing to delay equipment additions at its existing Kumamoto plant, which manufactures mature-node chips for industrial, consumer electronics, and automotive markets.

Read more

• [News] TSMC Reportedly Delays Kumamoto 2nd Fab Launch to 2029 Amid U.S. Expansion Push
• [News] TSMC Speeds Advanced Packaging: AP7 Targets 2026 Output; Arizona P6 Eyed for U.S. Packaging Hub

(Photo credit: TSMC)

Japan is advancing its 2nm ambitions with Rapidus, and the investment required is enormous. According to Japanese outlet Kyodo News, sources say Canon is in the final stages of negotiations to invest in Rapidus, with the amount expected to be in the several-billion-yen range. As a semiconductor equipment maker, Canon aims to support Rapidus’s plan to begin mass production in the latter half of FY2027 by joining the supply chain as a capital partner. The report also notes that SoftBank intends to make an additional investment.

As the report notes, Canon has supplied a semiconductor lithography system to the prototype line at Rapidus’s Chitose plant in Hokkaido. Rapidus is operating the system alongside EUV lithography tools made by ASML, the Dutch semiconductor-equipment giant.

Beyond funding, The Sankei Shimbun notes that the restart of Hokkaido Electric Power’s Tomari Nuclear Power Plant is seen as a crucial factor for Rapidus’s success. According to the report, Hokkaido Electric Power reportedly aims to bring the Tomari plant back online in 2027 after completing its seawall—a schedule that closely aligns with Rapidus’s planned start of mass production. For Rapidus, easing concerns about tight electricity supply would create the conditions needed to focus fully on development and manufacturing, Sankei Shimbun adds.

Rapidus Outlines Massive Expansion and Funding Roadmap

Rapidus has been outlining bold plans to catch up with top global foundries. Reports from Nikkei and Yomiuri Shimbun say that while its first Hokkaido Chitose fab has yet to begin mass production, the company aims to break ground on a second facility as early as FY2027, targeting 1.4nm chip output in 2029.

In parallel with its expansion plans, Rapidus has also submitted a business plan to Japan’s Ministry of Economy, Trade and Industry (METI) outlining its goal of reaching operating profitability around FY2030 and pursuing an IPO in FY2031, according to Nikkei. With its planned production roadmap, the company’s total investment is projected to surpass ¥7 trillion. METI, for its part, will provide an additional ¥1 trillion in support across FY2026–27. Beyond subsidies for technology development, the plan also includes additional capital injections, lifting total government support to ¥2.9 trillion, Nikkei adds.

Read more

• [News] Rapidus Reportedly Eyes 2029 1.4nm Production with 2nd Hokkaido Fab to Catch Foundry Giants
• [News] Japan Reportedly Adding ¥1 Trillion in Rapidus Support as It Targets FY2030 Profitability, FY2031 IPO

(Photo credit: Rapidus)

2027 is shaping up to be a pivotal year for SK hynix’s next-generation NAND ambitions. The company is accelerating its next-generation AI NAND development by teaming up with NVIDIA, aiming to deliver storage performance up to 30 times faster than current enterprise SSDs by 2027, according to ZDNet.

According to the report, the HBM leader is collaborating with NVIDIA on advanced AI NAND, with initial samples expected by the end of 2026. The company is already preparing a second-generation product, aiming for mass production by the close of 2027, the report notes.

Meanwhile, for the highly anticipated HBF (High Bandwidth Flash), developed in collaboration with SanDisk and dubbed “AI-N B,” SK hynix plans to begin evaluations once 2027 samples become available, the report adds.

AI-NAND Roadmap in Detail: AI-N P, AI-N B in Focus

ZDNet reports that while the AI industry splits between data centers, which demand massive processing, and on-device AI, which emphasizes high-efficiency, low-power operation, SK hynix is focusing on high-value AI memory tailored to each segment.

The report further points out that SK hynix’s AI-N initiative targets next-gen NAND with three approaches: AI-N P (ultra high performance SSD), AI-N B (High Bandwidth Flash), and AI-N D (High Capacity/ Low Cost SSD).

Among these, AI-N P is designed to efficiently handle high-volume data I/O in large-scale AI inference environments, minimizing bottlenecks between AI computation and storage, the report explains. Notably, SK hynix is achieving this through a redesigned NAND and controller architecture, and is conducting joint proof-of-concept (PoC) testing with NVIDIA.

The report, citing Kim Cheon-sung, Vice President of SK hynix, reveals that the firm targets ambitious performance milestones: sample products supporting 25 million IOPS—a measure of how many read/write operations a storage device can handle per second—over PCIe Gen 6 are expected to debut by late 2026. By the end of 2027, the company aims to mass-produce versions capable of 100 million IOPS, ZDNet notes.

According to Hankyung, current enterprise SSDs on the market achieve roughly 2–3 million IOPS, which is significantly slower compared to DRAM modules, capable of several billion up to 10 billion IOPS. By comparison, the AI-N P expected next year is projected to deliver 8–10 times the performance of today’s SSDs, with second-generation products potentially exceeding 30 times, ZDNet adds.

On the other hand, the report also highlights AI-N B (HBF), which significantly expands memory bandwidth compared to conventional SSDs by stacking NAND flash instead of DRAM. According to Kim, SK hynix is working on standardization of AI-NB with SanDisk, with the alpha version expected to be released around late January 2026. The company also plans to proceed with evaluation and testing once samples become available in 2027.

Read more

• [News] SK hynix, Samsung, and SanDisk Bet on HBF — The Next Battleground in Memory Sector
• [News] Memory Giants SanDisk, SK hynix Unite for HBF Standard, with Samples Expected in 2H26

(Photo credit: SK hynix)

China’s drive for tech autonomy through deeper integration has encountered an unexpected pause. According to South China Morning Post, supercomputer maker Sugon and chip designer Hygon have scrapped a long-anticipated mega-merger that had been under discussion for months.

As the report notes, the two Shanghai-listed companies said they ended merger talks because the market environment has shifted substantially since the deal was first conceived, and that conditions for executing such a major restructuring are not yet mature.

The cancelled merger has dampened expectations that Sugon and Hygon would form a major domestic ecosystem spanning advanced processors and high-performance servers, as the report notes. The report also states that the proposed transaction—first announced in late May—was expected to be valued at 116 billion yuan (US$16.4 billion), according to corporate filings in June.

Although the merger will not proceed, the report notes that both companies sought to reassure investors that their partnership remains intact. Sugon already holds the largest stake in Hygon, and the firms said they will continue to maintain close supply-chain collaboration, the report adds. As noted by STAR Market Daily, Hygon General Manager and Director Sha Chaoqun said the two companies maintain independent, market-driven operations and can still achieve coordinated, end-to-end development—from chip design to computing-power services—through industrial collaboration.

Hygon on H200’s China Entry

As for whether the recent U.S. approval of NVIDIA’s H200 sales to China will affect China’s chip market, STAR Market Daily notes that Sha said the H200’s entry could intensify competition in the domestic high-end chip segment. However, he added that the accompanying revenue-sharing requirement will push up procurement costs, meaning the chip’s market penetration will continue to face challenges.

As STAR Market Daily highlights, Hygon possesses R&D capabilities in both high-end CPUs and DCUs, and its “CPU + DCU” portfolio enables an integrated computing-power solution. In contexts such as large-scale AI model training and data-center development, its system-level compatibility is said to be significantly higher than solutions that combine third-party CPUs with domestic GPUs.

Read more

• [News] China’s Chip Industry Surges in 1H25 — 38 A-Share Listed Firms Reportedly Post Growth
• [News] China’s Chip Merger Push Reportedly Stalls as 8 Deals Collapse in 2025
• [News] Chipmaker Hygon to Acquire Sugon as China Pushes Chip Sector Consolidation

(Photo credit: Hygon)

Amid a sharp surge in memory prices, even Samsung is feeling the strain despite its in-house capacity. DealSite reveals that Roh Tae-moon, CEO and Head of Samsung Electronics’ DX Division, will meet Micron CEO Sanjay Mehrotra on the opening day of CES 2026 in Las Vegas, running January 6–9.

As the report highlights, such meetings are rare during the event, but sources say Samsung requested it to address tightening mobile DRAM supply. The discussion, according to DealSite, is expected to focus on LPDDR5X for the upcoming Galaxy S26, as soaring prices have left contracts with Samsung’s DS division and Micron unresolved.

As Sedaily reported previously, LPDDR5X 12GB, widely used in the Galaxy lineup, was trading at around USD 70 as of late November, more than double its price of about USD 33 earlier in 2025.

If confirmed, the move aligns with Samsung’s latest memory strategy. Sedaily reports that while the tech giant’s semiconductor division (DS) will supply memory chips to the Mobile eXperience (MX) unit, the deal will follow quarterly negotiations rather than the previously considered long-term, year-plus arrangement for mobile DRAM.

As Sedaily explains, Samsung’s DS division is reshaping its production lines to focus on higher-margin products, such as HBM for AI accelerators, aiming to maximize profitability. For now, both divisions have reached a broad agreement on minimum supply volumes for next year, helping the MX unit avoid a worst-case scenario of running short on mobile DRAM, the report adds.

Micron Remains Key Mobile DRAM Supplier for Samsung

Though rivals in the HBM market, Micron and Samsung share a closely intertwined relationship in mobile DRAM, blending competition with collaboration. DealSite reported in June that Samsung’s MX division was testing Micron’s LPDDR5X samples ahead of the Galaxy S26 launch, comparing them with in-house DS products, with results expected by year-end.

Micron also said in a mid-year press release that it is sampling 1γ-based LPDDR5X 16GB modules to select partners and plans to offer capacities ranging from 8GB to 32GB for 2026 flagship smartphones.

DealSite noted that if Micron is named primary supplier again, it would be another setback for Samsung’s DS division. Previous reports, including The Korea Herald, show that the Galaxy S25 used Micron’s mobile DRAM instead of Samsung’s. While Micron has supplied as a secondary partner for over a decade, this would mark its first role as the primary supplier for the Galaxy series, as per The Korea Herald.

Samsung Speeds up LPDDR6 Development

It is interesting to note that, according to Business Post, Samsung is seen as the frontrunner in LPDDR6 development. The company is expected to be the first to bring LPDDR6 to market and plans to showcase it at CES 2026, the report notes.

Built on Samsung’s 1c (12‑nm class) DRAM process, its LPDDR6 reportedly offers an 11% performance boost and 21% better energy efficiency over LPDDR5X. With up to 1.5× higher data bandwidth, it can support real-time large language model (LLM) workloads and high-resolution, high-refresh-rate applications, as per Business Post.

The report indicates that Samsung may start mass-producing LPDDR6 by year-end and could equip select Galaxy S26 models with the new memory, aiming to get a head start on competitors. However, some industry watchers suggest LPDDR6 might debut first in the Galaxy S27 series.

Read more

• [News] SanDisk, Samsung Reportedly Delay NAND Delivery, Hitting Transcend with 50–100% Price Spike
• [News] Samsung MX Faces Galaxy S26 Cost Pressure as Memory Division Reportedly Sticks to Quarterly Pricing

(Photo credit: Micron)

South Korea’s PC SSD market is undergoing significant shifts as major manufacturers raise prices and Micron’s recent announcement. According to South Korean outlet ZDNet, Micron—the world’s third-largest supplier—is preparing to exit the consumer SSD business, a move expected to create openings for Chinese SSD makers that compete heavily on price and volume.

The report notes that Micron plans to withdraw from the consumer SSD segment after February 2026, prompting mid-size Korean PC makers and consumers to seek replacement options. Citing industry sources, the report indicates that Micron’s exit is expected to substantially lower the entry barrier for Chinese SSD manufacturers.

Huawei Expands SSD Portfolio With New Korean Launch

As the report notes, Huawei Korea held a launch seminar on December 2 for domestic partners, introducing three SSD products: the eKitStor Xtreme 201 for desktop PCs and laptops, and the portable eKitStor Shield 210.

Notably, the eKitStor Xtreme 201 is a PCI Express 4.0, M.2 2280 NVMe SSD. The 1TB model delivers peak read and write speeds of 7.4GB/s and 6.7GB/s, respectively—performance comparable to competing offerings, the report adds.

Huawei Boosts SSD Appeal in Korea With TLC NAND

As the report highlights, the previous domestic model, the “eKitStor Xtreme 200E,” struggled to gain traction among Korean consumers because it used QLC NAND flash, which is generally less favored for speed and reliability. In contrast, the Xtreme 201 adopts TLC (3-bits-per-cell) NAND flash.

Citing a Huawei Korea representative, the report notes that the Xtreme 201’s TLC NAND flash provides nearly three times the total write endurance of SSDs from Korean manufacturers with the same capacity.

The report adds that Huawei completed wireless certification for its SSD lineup in Korea at the end of November and plans to begin consumer sales later this month. Official pricing has yet to be finalized.

Meanwhile, as the report notes, SSD prices in Korea are unlikely to ease anytime soon. As of December 9, major SSD suppliers said manufacturers had already informed partners of supply price increases of 20% to 40%, taking effect between late October and early December.

TrendForce observes that in the fourth quarter, market sentiment is shifting from recovery to scrambling for components. Major NAND Flash suppliers are cautious about increasing production due to past volatility, resulting in enterprise SSD output growth significantly lagging demand. With the market tilted in favor of sellers, TrendForce predicts that average enterprise SSD contract prices will rise by more than 25% QoQ in Q4, potentially leading to another industry revenue record.

Read more

• Enterprise SSD Prices and Shipments Surge in 3Q25, Industry Revenue Climbs 28%, Says TrendForce
• [News] Micron to End Crucial Consumer Memory by Feb 2026, Redirects Supply to Enterprise amid AI Surge

(Photo credit: Huawei)

China’s AI chip IPO boom is accelerating, and MetaX is stepping into the spotlight. According to South China Morning Post, Shanghai-based MetaX Integrated Circuits has become the newest standout in China’s stock market, attracting even more retail investor interest than fellow chip maker Moore Threads ahead of its debut.

The report highlights that MetaX garnered 5.17 million retail investors through its online subscription, yielding a final allotment rate of 0.033 per cent, according to a Monday filing. Meanwhile, Moore Threads attracted 4.82 million retail investors ahead of its Friday debut, with a final allotment rate of 0.036 per cent—meaning MetaX’s retail allotment rate was even lower. As the report explains, on the Star Market, institutional investors receive proportional allocations, while retail investors subscribe through a lottery-style system.

MetaX’s Emerging Role in China’s AI Chip Race

Seen as one of China’s leading candidates to challenge NVIDIA’s high-performance AI chips, MetaX is seeking to raise 4.2 billion yuan (US$594 million) from its IPO. Its offer price of 104.66 yuan per share implies a valuation of roughly 41.9 billion yuan, as the report indicates.

The report further points to MetaX’s C600—introduced in July—which features HBM3e and FP8 precision to accelerate AI training while lowering power consumption. Mass production of the “fully domestically produced” chip is expected in the first half of 2026.

MetaX’s chip development rests on a team established in September 2020 and composed primarily of former AMD engineers. As noted by 21jingji, the firm introduced its N100 inference chip in April 2023 and its C500 training-and-inference chip in February 2024. The C500 is now its main commercially shipped product.

MetaX Faces Ongoing Losses Amid Heavy R&D Spending

Meanwhile, as the South China Morning Post indicates, MetaX remains loss-making due to high research and development costs and NVIDIA’s long-standing dominance in the Chinese market. According to the report, the company posted a net loss of 1.4 billion yuan in 2024, even as revenue surged more than tenfold to 743.1 million yuan. In the first nine months of 2025, it recorded a loss of 345.5 million yuan while revenue rose more than 400 per cent from a year earlier. The company previously said it expects to break even in 2026 at the earliest.

Read more

• [News] China’s GPU Trio Rise as NVIDIA Retreats — Decoding Moore Threads, MetaX, and Cambricon
• [News] Baidu Reportedly Explores Kunlunxin Spinoff and Listing Amid a Surge in China Chip IPOs

(Photo credit: MetaX)

According to TrendForce’s latest memory spot price trend report, regarding DRAM, DDR5 and DDR3 spot prices saw a mild pullback this week as the rapid surge prompted some year-end profit-taking among traders, while DDR4 prices continue to climb—particularly 16Gb products. Meanwhile, in NAND, key support lies on the supply side, where suppliers—who have not released additional wafers as year-end approaches—have pushed spot prices to continuous all-time highs amid constrained market activity. Details are as follows:

DRAM Spot Price:

DDR5 and DDR3 prices have eased slightly this week after a sharp rally that proved too aggressive recently. TrendForce observes that some traders are making adjustments to profits as year-end approaches, however it will not influence the overall tight market conditions. DDR4 prices continue to rise, especially the 16Gb products. Recent changes to spot prices have not influenced the forecast that contract prices will continue to rise significantly. The average spot price of mainstream chips (i.e., DDR4 1Gx8 3200MT/s) has increased by 2.00% from US$16.729 last week (December 3) to US$17.064 this week (December 9).

NAND Flash Spot Price:

The wafer market was seen with subsided price hikes but persistently strong prices this week. Due to the excessive hike previously, the market is now under temporal revisions and adjustments, though TrendForce points out that the key support lies on the provision end, where suppliers, who have not released additional wafers as the end of the year approaches, resulted in continuous all-time highs of spot prices under constrained market activities. Spot prices of 512Gb TLC wafers rose by 0.28% this week (December 9), arriving at US$9.634.

Trump has approved H200 sales to China, but the arrangement has raised concerns. According to The Wall Street Journal, the chips permitted for export must undergo a special U.S. security review before shipment. As a result, the H200s—manufactured primarily in Taiwan—would first be sent to the U.S. for national-security screening before being forwarded to China. The report adds that this unusual routing has prompted discussion about whether it is linked to the U.S. government’s requirement for a 25% share of the sale.

As the report indicates, citing sources, because the Constitution prohibits export taxes, officials must design the arrangement carefully to avoid it being characterized as one, and routing the chips through the U.S. allows the government to classify the 25% charge as an import tariff rather than an export tax.

Uncertainty Over Responsibility for the 25% Charge

Economic Daily News notes that the question of who will absorb the 25% revenue-sharing fee has drawn considerable attention. Sources cited by the report indicate that tariffs are typically borne by the buyer, meaning NVIDIA is expected to take on the associated costs, with contract manufacturers handling final assembly also likely to feel the impact. By contrast, the effect on upstream semiconductor suppliers is expected to be relatively limited.

Notably, since the chips are expected to be manufactured by TSMC, the foundry giant’s response has drawn attention. As noted by Economic Daily News, TSMC did not comment. The report adds that the company has previously stated at shareholder meetings and earnings calls that “the buyer bears the related costs,” and while such fees may more easily affect price-sensitive consumer products, TSMC has not yet observed any change in customer behavior.

Security Concerns and Enforcement Challenges

Meanwhile, South Korean outlet Maeil Business Newspaper notes that experts are skeptical about how much this measure actually enhances security, arguing that the real issue lies in where the chips ultimately end up in China and how they are used. Likewise, The Wall Street Journal notes that Trump said AI chips would be supplied only to approved buyers, though risks of smuggling and diversion have made such controls difficult to enforce.

TrendForce predicts that China’s high-end AI chip market will grow by over 60% in 2026. The Chinese government is likely to keep supporting local AI chip independence, helping top IC design companies benefit from government and corporate projects. Consequently, domestic AI chips might increase their market share to about 50%. At the same time, imported products like Nvidia’s H200 and AMD’s MI325 are expected to hold nearly 30% of the market.

Read more

• China’s CSPs and OEMs likely to Actively Procure H200, but Domestic AI Chip Development Continues to Accelerate, Says TrendForce
• [News] Trump Greenlights NVIDIA H200 Export to China: Chip Reportedly 6x H20 Power, with 25% Fee

(Photo credit: NVIDIA)

While Washington approved NVIDIA’s H200 exports to China on December 8th, Beijing reportedly acted first. According to the Financial Times, China has, for the first time, added domestic AI chips to its official government procurement list, with the Ministry of Industry and Information Technology approving processors from leading local players including Huawei and Cambricon.

The Financial Times notes that the move, aimed at boosting the use of homegrown semiconductors in China’s public sector, could generate billions of dollars in extra revenue for local chipmakers. Significantly, the report points out that the announcement came just ahead of Trump’s decision to ease U.S. export restrictions, permitting NVIDIA to ship H200 chips to “approved customers in China.”

Interestingly, the reason Trump greenlit H200 exports to China, as per Bloomberg, was based on an assessment that Huawei can rival NVIDIA more closely than previously thought. U.S. officials cited by the report concluded that by 2026, Huawei could produce several million Ascend 910C accelerators—designed to compete with NVIDIA’s chips—up sharply from June’s estimate of just 200,000 units this year.

China’s Push for Chip Independence

While China’s updated procurement list has not been publicly released, several government agencies and state-owned enterprises have already received the guidance, the Financial Times suggests, adding that this also marks the first time they have been given formal written instructions.

It is worth noting that in recent years, domestic microprocessors intended to replace AMD and Intel chips, along with operating systems designed to substitute Microsoft Windows, have been added to the government-approved procurement list, the report adds.

Beijing’s latest move underscores its confidence that domestic AI chips have matured enough to challenge U.S. counterparts—a rationale that mirrors the reasoning behind Trump’s greenlight for H200 exports—following years of focused investment in the sector.

According to TrendForce, China’s high-end AI chip market is projected to grow over 60% in 2026. Continued government support for domestic AI chip development is expected to benefit leading IC design firms, potentially boosting domestic chips’ market share to around 50%. Imported products such as NVIDIA’s H200 and AMD’s MI325 are projected to hold roughly 30%, TrendForce notes.

Read more

• [News] Trump Greenlights NVIDIA H200 Export to China: Chip Reportedly 6x H20 Power, with 25% Fee
• [News] Inside Huawei’s Patent Playbook: GPU Dominance and Sub-2nm Ambitions Under Sanctions

(Photo credit: Huawei)

U.S. President Trump has approved the export of NVIDIA’s H200 chips to China. According to Commercial Times, industry sources say the move could accelerate inventory digestion across the AI supply chain and stimulate new orders for Taiwanese assembly and packaging providers. OEMs such as Gigabyte and Asus, along with major OSAT firms like ASE, are expected to be among the first to benefit.

As the report indicates, the H200 is built on TSMC’s 4nm process and belongs to NVIDIA’s previous-generation product line. Demand fell sharply over the past year due to export restrictions, resulting in elevated inventory. Sources note that most of this inventory sits with NVIDIA and Taiwanese module assemblers, meaning that once restrictions are lifted, the first beneficiaries will be the Taiwanese firms handling assembly and shipment.

The report adds that U.S. and European customers have already transitioned to Blackwell (the B-series) chips, reducing the significance of the H-series in the high-end market and making it a more suitable alternative for China.

As for whether NVIDIA will increase wafer starts for the H-series, the report notes that the likelihood is low, as the Blackwell line commands higher ASPs and remains in strong demand, and with N4P capacity already tight, resuming production of older chips would divert scarce manufacturing resources.

H200 Revival: From Packaging to Testing, Taiwan Suppliers See Opportunity

If the H200 returns to the market, it could lift utilization rates for advanced packaging and testing capacity. According to the report, institutional investors note that Taiwanese firms such as ASE, Powertech, and KYEC—strong in high-speed transmission, high-frequency packaging, and burn-in testing—would benefit most directly as AI and cloud customers in China resume procurement. The report also adds that because the H200 adopts HBM stacked packaging, it could help restart the pull-in cycle across Taiwan’s AI supply chain.

In addition, the report notes that although major AI server ODMs—including Wistron, Foxconn, Quanta, and Inventec—have been shifting their core products from the Blackwell platform to its successor, the Blackwell Ultra series, OEMs such as Gigabyte and Asus continue to fulfill some H200 orders in markets like Southeast Asia and the Middle East.

Meanwhile, Economic Daily News notes that Inventec is part of the supply chain for Chinese cloud service providers such as Alibaba and Baidu. Institutional investors cited by the report expect that following the H200 export approval, new L6 (motherboard) design and manufacturing orders will emerge, helping drive an increase in Inventec’s server shipments.

TrendForce predicts that China’s high-end AI chip market will grow by over 60% in 2026. The Chinese government is likely to keep supporting local AI chip independence, helping top IC design companies benefit from government and corporate projects. Consequently, domestic AI chips might increase their market share to about 50%. At the same time, imported products like NVIDIA’s H200 and AMD’s MI325 are expected to hold nearly 30% of the market.

Read more

• [News] Trump Greenlights NVIDIA H200 Export to China: Chip Reportedly 6x H20 Power, with 25% Fee
• China’s CSPs and OEMs likely to Actively Procure H200, but Domestic AI Chip Development Continues to Accelerate, Says TrendForce

(Photo credit: NVIDIA)

As major PC/NB brands—including Dell and Lenovo—have reportedly warned clients of imminent price hikes, the memory market’s supply–demand imbalance is worsening rapidly. Commercial Times reports that the NAND Flash shortage is morphing into a full-fledged “dry-year,” prompting multiple PC/NB OEMs to plan SSD capacity downgrades in 2026 to keep device prices from spinning out of control.

According to memory module makers cited by the report, industry peers have inventory that will only last until 1Q26, with some expected to run out of stock as early as March. industry sources further note that the scramble for NAND since Q4 has been “the toughest in nearly a decade,” warning that by 2Q26, the market may face a scenario where there’s simply no stock left to ship.

Thus, supply-chain sources cited by the report say PC OEMs are now reworking their 2026 product specs as NAND allocations fall short and costs surge. Standard SSD configurations such as 512GB are reportedly being cut to 256GB, and 1TB models are being scaled back to 512GB.

NAND Price Surge at Full Throttle

Commercial Times also brings up that NAND prices are rising at an unusually rapid pace, adding Micron’s latest quotes jumped nearly 50% in a single month, raising fears that suppliers may hike prices again before shipment. To avoid this, many are shifting to open-order (floating price) arrangements, the report adds. Industry players cited by the report admit that while orders are pouring in, taking them at fixed prices could actually deepen losses.

TrendForce predicts that average enterprise SSD contract prices will rise by more than 25% QoQ in Q4, potentially leading to another industry revenue record.

NAND Supply Strained by Conservative Expansion and AI Demand

TrendForce observes that tight supply reflects cautious NAND Flash makers, still hesitant to expand after years of volatility, leaving enterprise SSD output far behind soaring demand. Commercial Times adds that Samsung, SK hynix, and Micron have yet to announce capacity expansions—understandable given memory manufacturing’s heavy capital intensity, six-month equipment lead times, and seven- to ten-year payback periods.

At the same time, CSPs are boosting inventories to secure SSDs for costly AI server deployments, effectively crowding out NAND supply for PCs, TrendForce notes.

Read more

• [Exclusive] Memory Crunch Hits PCs: Dell Hikes Prices 15-20% Mid-December, Lenovo from January 2026
• [News] Smartphone Memory Inventories Reportedly Fall Below 4 Weeks, Low-End Devices in the Crosshairs

(Photo credit: Dell’s X)

Google has announced major progress on two categories of AI-powered smart glasses—one equipped with a display and the other focused on audio—setting the stage for head-to-head competition with Meta as early as next year.

In a blog post published Monday, Google said that the first co-developed AI glasses are scheduled for release in 2026. Early hardware partners include Samsung Electronics, Warby Parker, and Gentle Monster, though the final industrial design has not yet been revealed.

Google’s prototype showcases hands-free interaction and supports functions such as photography, navigation, calls, and object recognition, all powered by the Gemini large model. The display variant will offer both monocular and binocular optical configurations and support overlay applications including Maps AR navigation and floating Meet windows. The audio-first version emphasizes lightweight design with an appearance close to conventional eyewear, running Gemini on a paired smartphone to reduce device weight and extend battery life.

Google is also collaborating with Chinese AR firm Xreal on an independent smart-glasses project codenamed Project Aura, built on the same Android XR system and capable of standalone operation, though requiring an external battery. The company additionally rolled out software updates for Samsung’s Galaxy XR headset, including a new “Travel Mode” designed for use in cars and airplanes.

(Photo credit: Google)

As China’s top memory firms race to advance next-gen DRAM and NAND, domestic chip equipment makers are stepping up for a slice of the booming HBM market as well. According to ijiwei, ACM Research has introduced several solutions optimized for HBM production and plans to release HBM4-compatible cleaning systems by 2026.

Among the equipment already hitting the market, ACM’s flagship Ultra ECP 3D system is drawing particular attention. The platform reportedly handles TSV (Through-Silicon Via) copper filling—a critical step in HBM production. As ijiwei highlights, orders began ramping in H2 2024, landing a heavyweight win with Samsung among its customers.

According to ijiwei, the tool reportedly delivers copper filling for TSV structures exceeding 500 layers in 3D NAND configurations, achieving a remarkable 99.8% yield rate. The equipment is specifically designed to handle the high aspect ratio electroplating requirements essential for HBM fabrication, the report adds.

Beyond Samsung, ACM counts other memory giants as customers. Ijiwei notes that ACM’s SAPS megasonic single-wafer cleaning technology tackles TSV deep-hole cleaning, with the platform already deployed in HBM production lines at SK hynix and Micron. Meanwhile, its TEBO megasonic cleaning technology, which addresses the unique cleaning demands of complex 3D structures like HBM, has passed validation at tier-one chipmakers, with four units already shipped, the report suggests.

ACM Research’s HBM Push

According to eastmoney.com, ACM Research is advancing rapidly in the HBM market. The company plans to launch HBM4-compatible cleaning equipment by 2026, handling 16-layer stacked wafers with 2,048 TSVs. In Q3 2025, it secured an HBM packaging equipment order worth over 200 million RMB from a leading global memory manufacturer, with delivery expected in the first half of 2026, the report notes.

Additionally, ijiwei reports that the company has established a dedicated HBM lab at its South Korea R&D center, signaling aggressive plans to expand its footprint in the fast-growing HBM equipment market.

Read more

• [News] Three Semiconductor Makers Accelerate the Development of HBM & Advanced Packaging Equipment
• [News] ACM Research Launches Panel-Level Etching Tool, Expanding Its FOPLP Porfolio

(Photo credit: ACM Research)