All published memory-market news.

Researchers at the National University of Singapore have presented CHIPSMORE, a proposed accelerator for serving language-model inference requests. Semiconductor Engineering reports that the design combines compute-in-interconnect techniques with compute-in-memory chiplets. The study targets systems that must handle different inference modes and multiple requests at once, including low-rank adaptation workloads. Moving selected operations nearer to memory can reduce data-transfer pressure, which is a central constraint in many AI inference designs. This is research rather than a commercial product announcement, so it does not change current memory demand or pricing. It is nevertheless relevant to the longer-term direction of AI-memory architectures and the ways chiplet integration may be used to improve inference efficiency.

The hardware-accurate NeoGeo AES+ remake has reportedly been delayed until late 2027 because its maker is struggling to source memory at workable terms. Tom's Hardware says the company connected the decision to an industry-wide RAM shortage and stronger AI demand. One specialist console project is too small to move the memory market, and the account has not been independently verified by RamTrend. Still, it illustrates how constrained memory supply can affect buyers outside the largest data-center and handset customers. The item is best read as an anecdotal shortage signal rather than evidence of a new pricing trend. Broader contract-price and shipment data remain more important for judging market conditions.

Samsung Electronics is reportedly increasing production of mid- and lower-priced Galaxy phones through November, while some competitors reduce exposure to lower-margin models. DigiTimes Daily attributes the shift partly to a more difficult component-cost environment. Memory is a meaningful bill-of-materials item in affordable handsets, so a rise in DRAM and storage prices can pressure margins most strongly in value segments. A producer with procurement scale or more flexible sourcing may be better positioned to maintain shipments while rivals trim plans. The report is a directional indicator rather than a measure of handset-memory demand. It nevertheless shows how higher component prices are beginning to affect production decisions downstream from the memory supply chain.

Adata Technology and Team Group reported record monthly revenue for August, according to DigiTimes Daily. The companies sell memory modules and related storage products into consumer and commercial channels. The report links their inventory-building activity to persistent AI-related demand and a tighter supply outlook. That combination can encourage distributors and module vendors to secure components earlier, particularly when they expect contract prices to keep rising. The figures do not establish a new industry-wide price benchmark on their own, but they support the view that the memory supply chain remains cautious about availability into 2027. RamTrend will watch whether the same pattern appears in component pricing and shipment data.

Desktop add-in-board shipments reached 12.5 million units in the second quarter of 2026, the highest quarterly level since early 2022, according to Jon Peddie Research data reported by TechPowerUp. The report says shipments rose 5.9% from the previous quarter and 7.8% from a year earlier. The growth came despite several rounds of GPU price increases and rising graphics-memory costs. The source says buyers may be bringing purchases forward because they expect prices to rise further, a behavior that can temporarily support demand even in an inflationary environment. The supplied material does not give a breakdown by vendor, memory configuration, region, or board price. The explanation for demand is attributed to JPR commentary rather than established as a causal result.

Kioxia's planned US listing could expand the Japanese NAND maker's access to global institutional investors, according to DigiTimes. The report places the potential listing alongside a sharp recovery in NAND profitability, stronger data-center demand, and a new phase of capacity investment. Improving profitability can give NAND suppliers more flexibility to fund technology transitions and manufacturing expansion. At the same time, new capacity investment must be weighed against the need to preserve supply discipline in a market that has historically been volatile. The supplied source does not state the timetable, valuation, listing venue, financial metrics, or capacity figures. The item is a strategic-financing and market-cycle signal rather than a confirmed capital-markets transaction.

Macronix International, Winbond Electronics and Etron Technology reported strong revenue growth through August, according to DigiTimes. The report says all three memory manufacturers benefited from the continuing market upswing, with executives pointing to tight demand conditions. The companies operate in different areas of the memory market, but their reported momentum adds to evidence that firmer pricing and constrained availability are extending beyond the largest DRAM and NAND suppliers. Revenue growth can reflect changes in shipment volume, product mix, and pricing, so it should not be treated as a pure price measure. The supplied source does not provide each company's revenue figures, product-level sales mix, or comparison with prior periods beyond the general description of strong growth. The item is a company-results signal of a tight memory environment rather than a complete market survey.

Chinese chipmakers competing with Nvidia are sharply increasing prices for current and upcoming AI accelerators, according to DigiTimes. The report attributes the adjustments to tighter and more expensive high-bandwidth memory. HBM is a critical input for AI accelerators, where memory bandwidth and capacity often determine system performance. Rising HBM costs can therefore move quickly into the price of domestic compute hardware, adding pressure to China's efforts to expand its own AI infrastructure stack. The supplied source does not identify the companies involved, the size of price changes, HBM suppliers, product volumes, or contract terms. It is a market signal of HBM-linked cost pressure rather than a complete measure of China's accelerator market.

Nvidia is reportedly changing the planned HBM configuration for its Rubin Ultra AI accelerator, moving to an eight-high stack design from an earlier twelve-high approach, according to DigiTimes. The report links the change to higher memory costs and a greater focus on bandwidth efficiency and total system economics. HBM stack height is a key design choice for AI accelerators because it affects memory capacity, packaging complexity, power, and cost. A move toward fewer stacks or lower stack height could signal tighter optimization of memory resources as accelerator vendors balance performance against the cost of high-bandwidth memory. The supplied source does not provide a confirmed Nvidia product specification, product timing, memory supplier list, or expected HBM volume change. The report should be read as a roadmap signal rather than a confirmed shift in Nvidia procurement.

Jiangsu Pacific Quartz has obtained qualification for high-purity quartz materials and components in domestic DRAM manufacturing, according to Tom's Hardware. The company already supplies high-purity quartz to chipmaking-tool producers including Lam Research and Tokyo Electron, the article says. High-purity quartz is used in semiconductor manufacturing equipment and consumables, making local qualification relevant to efforts to reduce supply-chain exposure in China's chip industry. The article cites an analyst's view that the domestic DRAM manufacturer may be CXMT, but it does not state that the company confirmed this identification. The supplied material does not disclose the qualified material volume, customer name, production impact, or whether the qualification will replace imports. The development is a materials-supply-chain signal rather than evidence of an immediate increase in DRAM output.

DeepSeek has reportedly ordered 160,000 Huawei Ascend 950DT AI accelerators for a large-scale inference deployment in Inner Mongolia, according to TechPowerUp. The report says each accelerator includes 144 GB of Huawei-designed HBM memory, which would amount to roughly 23 PB across the stated order. The reported deployment illustrates the scale of memory requirements for large AI inference systems. An order of that size would also underline the importance of domestic accelerator and high-bandwidth-memory alternatives in China's AI infrastructure buildout. The order size, deployment plan, accelerator specifications, and total HBM figure are reported by the source and are not independently confirmed in the supplied material. They should be treated as a reported demand signal rather than a confirmed procurement disclosure.

Apple raised prices across much of its iPhone lineup following the September 9 launch of its latest Pro models, according to TechPowerUp. The report says the iPhone 18 Pro and Pro Max start at US$1,199 and US$1,299, respectively, and that several retained models also received US$100 increases without hardware changes. The article attributes the timing to rising DRAM and storage costs as AI data-center demand absorbs memory supply. It also notes that other PC-system makers have adjusted prices as DRAM and SSD costs rose. Apple has not, in the supplied material, provided a detailed cost breakdown or explicitly confirmed that memory is the cause of each price change. The price increases are factual product-market data, while the precise contribution of DRAM and NAND costs remains the report's interpretation.

CXMT is reportedly encountering low yields in HBM3E risk production, according to TechPowerUp's account of South Korean media reports. The article says roughly three of every four HBM3E stacks are defective, implying a yield near 25%. The report attributes the challenge in part to immaturity in through-silicon-via technology, which is central to connecting stacked DRAM dies. It also says CXMT is attempting an eight-high HBM3E configuration while its established competitors are progressing toward newer HBM generations. The yield figures, production status, and technical comparisons are reported rather than confirmed by CXMT in the supplied material. The development should therefore be treated as a reported indicator of the execution gap in China's high-bandwidth-memory manufacturing efforts, not a verified company disclosure.

Samsung Electronics has opened an advanced packaging research center in Yokohama, according to DigiTimes. The report says the move brings development work closer to Japanese suppliers of materials and equipment that are important to high-bandwidth-memory packaging. Advanced packaging has become a major differentiator in the HBM market because the technology depends on integrating memory stacks, logic, interconnects, and thermal solutions. The reported center may help Samsung coordinate development with its supplier base as it works to narrow the HBM market-share gap with SK Hynix. The supplied source does not disclose the center's staffing, investment, specific supplier relationships, technology milestones, or effect on HBM output. It is a supply-chain and R&D positioning signal rather than evidence of an immediate production increase.

Apple has reportedly entered a long-term NAND flash supply agreement as it seeks more reliable volume from memory manufacturers, according to DigiTimes. The report says the company is prioritizing supply guarantees over its historical emphasis on securing lower component prices. The reported strategy reflects a market in which AI infrastructure demand is tightening memory availability. For large device makers, long-term agreements can reduce exposure to constrained supply, but they can also limit flexibility when demand or prices change. The supplied source does not identify a confirmed supplier, contract duration, volumes, pricing, or the precise terms of any agreement. The report names Kioxia as a potential recipient, so this should be treated as reported supply-chain positioning rather than a confirmed bilateral contract announcement.

Kioxia has ruled out deeper manufacturing ties with SK Hynix and is reportedly resisting another aggressive round of NAND price increases, according to DigiTimes. The report says the company is instead emphasizing longer customer contracts and disciplined capacity expansion. The approach comes as data-center demand keeps memory supply tight, the report says. Longer contracts can give suppliers and customers more predictability when spot conditions are volatile, while controlled capacity growth can help producers avoid abrupt changes in market balance. The supplied source does not identify the customer contracts, pricing terms, capacity figures, or timeframe for Kioxia's production plans. The item is a supplier-strategy signal rather than a comprehensive measure of NAND pricing conditions.

Samsung Electronics and ASML have announced an expanded strategic partnership centered on High-NA EUV lithography and advanced semiconductor manufacturing technologies. The companies say the work is intended to accelerate the development and deployment of next-generation capabilities. For Samsung, deeper cooperation around leading-edge lithography is relevant to its broader semiconductor roadmap, which spans both logic and memory technologies. High-NA EUV is a potential enabler for continued process scaling where patterning complexity and precision become more demanding. The supplied announcement does not identify joint-development milestones, fab locations, equipment volumes, investment amounts, or memory-product deployment dates. It is a strategic manufacturing signal rather than a confirmed change to Samsung's DRAM or HBM output.

ASUS and Galax have raised prices for selected GeForce RTX 50-series graphics cards, according to TechPowerUp. The report says the increases range from roughly US$22 to US$74 in China depending on the model. The article attributes the adjustments to rising GDDR7 costs and says similar pressure had already prompted earlier changes among Nvidia board partners. Graphics-card pricing can reflect the cost of both GPU silicon and high-performance graphics memory, making GDDR7 a relevant input for the wider hardware market. The supplied material does not include contract prices, memory-kit purchase terms, or a complete list of affected products and regions. The item is a channel-pricing signal tied to reported GDDR7 cost pressure rather than a direct measure of the memory market.

TSMC has reportedly asked materials and equipment suppliers to develop microbump bonding and underfill processes at the 5-micron class for advanced HBM packaging, according to DigiTimes' account of an ETNews report. The reported work suggests that conventional microbumps may continue to be scaled for high-bandwidth-memory integration rather than being displaced immediately by hybrid bonding. Packaging interconnect density is increasingly important as AI accelerators combine multiple dies and memory stacks. The supplied report does not give a production timeline, supplier list, qualification status, or confirmation of deployment in a specific HBM generation. It is a packaging-roadmap signal rather than evidence of a confirmed manufacturing transition.

TrendForce estimates that DRAM industry revenue reached nearly US$154.73 billion in the second quarter of 2026, up 59.5% from the previous quarter, according to a report carried by TechPowerUp. The report attributes the revenue increase to higher traditional DRAM contract prices and supply expansion that continued to trail demand growth. The source also points to AI-server demand as a driver of higher shipments for HBM3E, LPDDR5X, and high-capacity RDIMMs. Growth in AI training and inference workloads can increase demand across multiple memory products rather than only one segment. The supplied article does not include the underlying regional, supplier, or product-level revenue breakdowns. The figures should be interpreted as TrendForce's market-research estimate rather than audited industry totals.