All published memory-market news.

DigiTimes reports growing activity among Chinese chip designers in high-speed connectivity for AI servers. Montage Technology expects its third-quarter 2026 profit to more than double, attributing the outlook to stronger DDR5 shipments. Separately, Shenzhen-based Techwinsemi is developing Compute Express Link controllers for memory expansion. Together, the developments point to activity across both memory modules and systems that extend memory capacity; the source does not quantify market-wide shipment or pricing effects.

Credo and AMD have built an emulation platform for memory disaggregation using Credo's OmniConnect Weaver software on AMD's Versal AI Edge Series Gen 2 adaptive SoC. The system is designed to make terabytes-class LPDDR memory available to a programmable-logic XPU over VSR links. The source describes a working emulation platform, not a generally available commercial deployment. It also says the design uses standard organic packaging without a silicon interposer. The announcement highlights an alternative approach to attaching high-capacity memory to AI accelerators, but does not provide shipment or pricing details.

Dell Technologies announced an expanded AI Factory with NVIDIA lineup at GTC Berlin. The additions include a PowerEdge M9882 rack with 256 NVIDIA Vera CPUs, a Groq 3 LPU inference system, an air-cooled two-socket Vera server and a Dell Pro Precision desktop with GB10 and 64GB of memory. Dell says the software is available now, while the new hardware is scheduled to arrive from October 2026 through early 2027. The announcement adds one specific memory configuration to a broader AI infrastructure portfolio; it does not disclose a broader memory supply or pricing change.

Tom's Hardware, citing Reuters, reports that Micron employees at the company's Taoyuan operations in Taiwan plan a one-day strike on October 22. The union is seeking a specific proposal for this year's bonus and a more transparent formula for future profit sharing. The union says the stoppage is meant to encourage an agreement rather than end negotiations. The source characterizes the expected effect on Micron's memory output as limited, but provides no production figures.

FADU introduced its File-Semantic SSD at the 2026 OCP Global Summit. The design carries NFSv3 operations through standard NVMe commands and handles file lookup and mapping inside the drive, aiming to reduce work for host CPUs when GPUs and other accelerators retrieve data. FADU reports about 460,000 operations per second in a preliminary test that maintained throughput as the server was reduced from 16 cores to one. The company expects partner samples in Q3 2027; these are vendor-reported results and a future sampling target.

Tom's Hardware reports that Amazon is offering Samsung's 4TB 990 Pro M.2 SSD for $799.90, compared with a listed prior price of $1,099. The article describes this as a 27% reduction and says the offer works out to roughly 20 cents per gigabyte. This is a retailer-specific promotion that may change; it should not be read as evidence of a broad NAND price decline.

Blocks & Files reports that HPE intends to send its Spaceborne Computer-4, carrying Kioxia solid-state drives, to the lunar surface on Astrolab's FLIP rover. The mission is planned for launch in late 2026 under NASA's Commercial Lunar Payload Services program, with Griffin-1 expected to deliver the rover near the lunar south pole. This is a planned deployment; the source does not report that the launch or landing has occurred.

A DIGITIMES report says Samsung Electronics plans to manufacture its M2800 mobile modem on a 2nm process rather than 4nm. The move is expected to free 4nm capacity primarily for HBM base dies, the control layer at the bottom of stacked memory products. The report gives no capacity figures or expected output, so the scale of any additional HBM supply is not yet clear.

Micron says its Micron Memory Taiwan unit and the Taichung chapter of the Taiwan Micron Memory Union reached agreement after several rounds of talks. The parties signed a labor-management agreement and withdrew mediation that had been scheduled for October 22. Discussions with the Taoyuan union are continuing, and that group has planned a one-day strike for the same date. The outcome therefore reduces immediate uncertainty at one site but does not resolve the separate Taoyuan dispute.

Synaptics has announced its Astra SR200 series of microcontrollers, which it describes as xMCUs combining heterogeneous compute, AI acceleration and integrated high-bandwidth custom DRAM. The company says the memory can scale to hundreds of megabytes, allowing larger multimodal models to run on devices such as wearables, cameras and industrial systems. The announcement also describes a Torq AI engine with Google's Coral NPU. These are vendor specifications; the release gives no independent performance comparisons.

In an SK hynix-hosted article, OXMIQ Labs CEO Raja Koduri describes data movement as a growing constraint as AI workloads move toward inference and agentic systems. The piece argues that system performance depends on how compute, memory, networks and storage work together, and discusses memory-oriented approaches including HBM, hybrid bonding and processing-in-memory. It is a technology perspective rather than a quantitative product or market announcement.

In comments summarized by TechPowerUp from a Wall Street Journal interview, Dell said memory availability, not a lack of customer demand, is holding back some AI-server deliveries. The report cites roughly $95 billion in undelivered AI-server orders and says Dell expects meaningful new memory-fab supply only around 2029 or 2030 because facilities take years to build. These are attributed estimates, not a guarantee of industry-wide timing; the source also notes that Dell's procurement strategy for consumer products is unclear.

The supplied DIGITIMES summary describes a widening gap between gains in AI compute and the pace of memory-bandwidth growth. It says this can constrain both performance and energy efficiency, and points to HBM, optical interconnects, advanced packaging and 3D stacking as approaches being explored to bring data closer to processing. The source summary provides no comparative benchmarks or implementation timeline.

A report says Kioxia plans to focus on proprietary NAND technology and solutions tailored to customers as demand for AI storage increases. The approach favors higher-value development over aggressive expansion of production capacity, with the stated aim of limiting exposure to memory-market swings. The supplied summary does not quantify planned output, investment, or customer demand.

A hobbyist converted two 16GB DDR5-5600 SODIMMs into a desktop kit running at DDR5-7800. The source reports that the donor modules cost $120 and that the full parts bill came to about $175. This is an individual hardware-modification project, not evidence of a broader change in memory supply or retail pricing.

SK hynix presented its Memory-Centric AI Rack System, or MARS, at the OCP Global Summit. The design separates memory from individual servers and pools it across a rack through Compute Express Link (CXL) fabrics. SK hynix says the approach is intended to address capacity and cost constraints that limit the scaling of large AI models. The supplied material does not include deployment details or quantified performance results.

SK hynix is strengthening its metrology and inspection work as memory structures become more three-dimensional. The source identifies both 3D DRAM architectures and increasingly complex NAND stacking as cases where conventional inspection methods face new demands. It does not specify new tools, investment levels or production milestones, but the development highlights process-control needs as memory manufacturers add structural complexity.

ORICO has entered the laptop market with the Obook 15, a Windows 11 system based on AMD's Ryzen 7 6800H. Memory options are 12GB or 24GB at 6,400 MT/s, while internal storage configurations include a 256GB or 512GB SSD and a 1TB option. The listed US prices are $689.99 for 12GB/256GB, $749.99 for 12GB/512GB and $899.99 for 24GB/1TB. A second M.2 slot is specified for storage expansion. The laptop is listed as available in the United States and the Philippines.

Gartner's estimate put third-quarter PC shipments at 55.4 million units, a 22.4% decline from a year earlier. Electronics Weekly reports that higher memory and storage costs are now showing up in PC pricing. The supplied excerpt does not include a breakdown of how much of the shipment decline or retail-price change is attributable to memory, so the figures should not be read as a measure of memory demand alone.

Hanmi Semiconductor presented several packaging systems at the 2026 SEDEX exhibition. The lineup included the TC Bonder 4 for HBM4 production and a Wide TC Bonder that the company is targeting for launch in 2027. Hanmi also introduced five 2.5D packaging systems designed for larger AI-chip packages. The announcement reflects ongoing investment in packaging tools for memory and AI devices, but gives no production-volume or customer details.