All published memory-market news.

Higher memory prices are adding pressure to smartphone and notebook shipments in 2026, according to DIGITIMES. The publication expects continued AI-infrastructure investment to keep memory quotations rising into 2027. In response to higher retail costs and weaker buying interest, hardware makers including Samsung are leaning on software differentiation, AI integration, after-sales services, cost controls, and promotion of older-generation models. The supplied summary does not quantify shipment or price changes.

SK hynix memory products have been placed in orbit aboard a South Korean satellite following the country's fifth Nuri rocket launch, according to DIGITIMES. The payload includes DRAM and Universal Flash Storage devices, and the satellite has established communication with ground stations. The mission is intended to enable reliability testing in space; no test results have been reported, so the products' performance in orbit remains unknown.

Kioxia has introduced its LD4-series solid-state drives in the E1.L form factor, with configurations reaching 122.88TB, according to the company and Tom's Hardware. The drives use BiCS8 3D QLC NAND and are aimed at high-density hyperscale deployments, including read-intensive data repositories and object storage. The supplied report does not provide performance measurements, so no speed comparison can be made.

The changing capacity of next-generation GPUs is altering when AI operators turn to memory offload, according to a DIGITIMES analysis. Workloads that once exceeded on-package high-bandwidth memory can now run directly on newer accelerators. External capacity is therefore becoming more relevant for deployments with unusually large memory requirements and for operators seeking to extend the service life of existing GPUs. The report describes a shift in use cases, not a quantified change in total memory demand.

AMD has described a memory-profiling feature planned for Zen 6 that samples data accesses and reports their physical addresses and source, including DRAM or external memory such as CXL. The goal is to help Linux place frequently accessed pages in faster tiers. In AMD's tests on a system with 256 logical CPUs and 256GB of CXL memory, some workloads improved more with the profiler-assisted approach than with the existing method, while Redis showed no gain and another benchmark regressed slightly. The pghot kernel changes remain in development, and the results are AMD's own measurements.

GlobalFoundries and TSMC have signed a five-year manufacturing agreement worth $2 billion for silicon interposers used in TSMC's CoWoS-S packaging ecosystem. GlobalFoundries plans to expand its Malta, New York facility, with volume production scheduled to ramp in the first half of 2028. The interposers provide dense connections between compute chiplets and HBM stacks in AI and high-performance computing packages. The companies have not disclosed the planned output capacity.

Kioxia has introduced the LD4 Series, its first E1.L SSD using eighth-generation BiCS FLASH QLC NAND, according to StorageReview. The long-form-factor drives target read-intensive workloads in dense 1U servers and are being sampled to selected customers at 15.36TB and 30.72TB. Kioxia says the underlying design has been validated for capacities up to 122.88TB, though the source does not say those larger versions are currently available. The launch brings higher-density QLC storage to hyperscale-oriented systems.

TrendForce's latest notebook outlook identifies component inflation as a growing risk to demand in 2027. Its baseline forecast calls for a low-single-digit decline in global shipments. If DRAM and CPU prices remain elevated and manufacturers pass more costs to buyers, the report says the decline could reach the high single digits as consumers extend replacement cycles. TrendForce also notes that earlier purchases and improved CPU availability supported shipments in 2026, potentially pulling demand forward. These are conditional forecasts, not confirmed shipment results.

Valve's September 2026 Steam Hardware Survey shows 32GB configurations taking the largest share of reported gaming-PC memory setups. They represented 42.22% of surveyed systems, while 16GB machines accounted for 37.82%. The survey also recorded smaller but growing shares for 24GB and 48GB configurations, at 2.56% and 1.46%, respectively. These figures describe the survey's participating systems and should not be read as a full-market PC shipment census.

Samsung Electronics is undertaking a $4.08 billion project in Vietnam to expand memory testing beyond South Korea, according to DigiTimes. The report describes the investment as a new back-end and export location amid tight DRAM and NAND supply. The supplied source does not specify a construction schedule, site capacity, product mix or how much testing volume the facility will add, so the project's near-term effect on memory availability is not yet clear.

Nvidia and Microsoft have opened preorders for Windows laptops built around Nvidia's RTX Spark platform, according to StorageReview. The report says laptop availability is scheduled to begin October 16, with compact desktop systems expected in November. The RTX Spark design combines a GPU and CPU with 128GB of unified memory, according to the source's product description. Nvidia also previewed a Windows version of DGX Station for release in the fourth quarter. The article does not provide a complete list of laptop configurations or pricing.

INNO3D has announced the AGS-6220V2, a 6U rack server designed for AI and high-performance computing. The company's specification lists support for up to 32 DDR5 registered DIMMs and eight PCIe Gen 5 x16 slots for NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. The system uses two fifth-generation Intel Xeon Scalable processors. INNO3D did not provide a maximum memory capacity or pricing in the supplied announcement.

Elon Musk has clarified that Tesla's AI5 chip is expected to use 96GB of LPDDR5 memory, according to DigiTimes. That follows an earlier post in which he said capacity would be reduced to 72GB. Musk said the larger amount avoids leaving Tesla with an unusually low memory configuration compared with other designs. The source does not specify production timing or the number of AI5 chips planned.

Dell has announced an XPS 16 Creator Edition built around Nvidia's RTX Spark N1X platform, with 128GB of unified memory, according to StorageReview. The laptop is listed for preorder at $3,799.99. Dell also announced a compact Creator Edition desktop using the same silicon and a Windows version of its GB300 workstation. The laptop illustrates how high-capacity unified memory is being packaged into creator-class portable systems; the report does not identify the memory supplier or provide details on availability beyond the preorder announcement.

HPE has received its first order for the AMD Helios AI Rack, with cloud provider Vultr committing $1.2 billion to deploy the systems in U.S. data centers, according to StorageReview. The rack-scale platform is built around AMD's 72-GPU MI455X system, with HPE providing infrastructure components. The commitment is a large deployment signal for AI computing capacity and its associated high-bandwidth memory demand, although the supplied report does not state HBM quantities or a memory procurement schedule.

HPE has announced four initial ProLiant Gen13 servers based on AMD's sixth-generation EPYC processors. The lineup includes the two-socket DL585a with support for eight GPUs and a single-socket DL525 that scales to 256 CPU cores and a 16-channel memory configuration, according to StorageReview. HPE also introduced two Open Compute Project systems. The announcement underscores continued growth in memory-channel capacity for AI and data-center servers, although the supplied report does not specify memory capacity or supported DIMM configurations.

Micron and Netlist have settled a patent dispute involving Netlist's memory technology. Under the agreement described by DigiTimes, Micron will license the patents for a total of $600 million over five years, bringing the complaint to a close. The report does not specify the patents' technical scope, royalty structure beyond the stated total, or any effect on Micron's product roadmap. The agreement is a significant legal and licensing development for two memory-sector companies, but not a direct indicator of chip pricing or supply.

Nvidia's DGX Spark lineup now includes a 64GB configuration listed at $4,999, according to ServeTheHome. The outlet also reports that the 128GB GB10 model has moved to $6,950, attributing the increase to memory pricing. The lower-capacity option gives buyers a less expensive entry point, while the reported price change on the larger system illustrates how memory costs can flow into integrated AI hardware. No component-level memory price breakdown was provided.

Kingston Technology says TrendForce's revenue ranking gives it 62% of the worldwide third-party DRAM module market for 2025. The company also says it retained the leading position for a 23rd consecutive year. Kingston points to enterprise, data-center and industrial customers as increasingly important buyers seeking reliable supply and ongoing technical support. The share figure is attributed to the company's account of TrendForce data; it refers to third-party modules, not the entire DRAM market.

Nvidia is adding a 64GB version of its DGX Spark local-AI system for buyers whose workloads do not require the existing 128GB configuration. The new system retains the GB10 platform, and the source says planned 64GB systems from several PC makers will start at $4,999 when they launch October 23. The existing higher-capacity systems remain part of the lineup for larger models and other memory-intensive tasks. The move offers another capacity and price point for local AI hardware, rather than indicating a change in memory technology.