All published memory-market news.

SK hynix said second-quarter 2026 revenue reached 79.3187 trillion won, with operating profit of 60.5426 trillion won and net profit of 93.9226 trillion won, all company highs. The company linked the performance to sustained AI infrastructure spending, stronger sales of high-value memory products, and quarter-over-quarter price gains in both DRAM and NAND flash. Beyond the headline results, the company said it has signed long-term agreements with around 10 customers and expects memory demand to remain structurally strong as AI deployments expand. It also said HBM4 began mass shipments in the second quarter, HBM4E sample shipments were completed in the first half, and additional capacity projects will be phased in to support demand. For NAND, SK hynix said 321-layer products now make up the largest share of production and should reach about half of domestic capacity by year-end. For RamTrend, the key signal is that AI-driven demand is still supporting firmer pricing and aggressive capacity planning in premium memory categories. That does not automatically mean every memory segment will tighten at the same pace, but the company commentary supports a constructive near-term pricing outlook for HBM, AI server DRAM, and related enterprise storage products.

TechPowerUp reports that SK hynix is preparing to start mass production of LPDDR6 during the second half of 2026, after earlier expectations had pointed to the first half of the year. The report says the new memory is built on SK hynix's sixth-generation 10 nm-class node and is expected to deliver higher throughput than LPDDR5X, with prior indications pointing to speeds up to 14.4 Gbps. It also describes efficiency-focused features such as selective sub-channel operation and dynamic voltage and frequency scaling. For RamTrend, this is a technology and supply-readiness signal rather than an immediate retail pricing event. A commercial LPDDR6 ramp would strengthen SK hynix's position in advanced mobile memory and could influence future component sourcing across smartphones, AI edge devices, and other low-power systems. The reported performance and efficiency gains suggest vendors are prioritizing bandwidth-per-watt improvements as memory demand expands beyond traditional handset use cases.

Tom's Hardware reviewed how sharply retail memory pricing has moved over the past year and described a consumer market still constrained by AI-related demand. The report points to major price increases across DDR5 kits, higher costs for older DDR4 products, and more expensive SSD capacity, leaving DIY PC builders with fewer affordable upgrade options. It also highlights Micron's exit from the Crucial consumer business as a sign that suppliers are prioritizing higher-margin enterprise and HBM demand over retail channels. For RamTrend readers, the main takeaway is that consumer memory relief has not yet arrived. Even where DDR4 remains available as a fallback, pricing has risen enough to limit its value advantage. The article also suggests that system bundles and prebuilt PCs may now offer better cost efficiency than buying standalone RAM at current spot-like retail prices. If supply remains concentrated on AI servers and enterprise customers, consumer DRAM pricing is likely to stay firm rather than normalize quickly.

Tom's Hardware revisited Intel's LGA 1700 platform to compare DDR4 and DDR5 in a modern gaming suite and found DDR4 trailing by roughly 11% to 14% on average, with some titles showing losses above 20% and one cited result above 25%. The article argues that LGA 1700 remains a practical bridge platform because it lets buyers reuse DDR4 while postponing a more expensive move to DDR5. That matters for RamTrend because the benchmark story is being driven by memory pricing, not just enthusiast tuning. The source describes DDR5 as expensive enough to reshape buying decisions, keep older DDR4 platforms relevant, and push consumers to weigh weaker CPUs plus DDR5 against stronger CPUs paired with legacy memory. Even where DDR5 clearly leads on performance and upgrade longevity, elevated DRAM prices are extending the commercial life of DDR4 and delaying full migration for cost-sensitive builders.

Tom's Hardware tested older hard drives, SATA SSDs, and NVMe SSDs in several external enclosures and found that reused drives can still deliver practical performance, especially when paired with the right interface. A SATA SSD remained far faster than a portable hard drive, while an older PCIe 3.0 NVMe model came close to dedicated 10 Gbps and 20 Gbps external SSDs. At the high end, a PCIe 5.0 drive in a USB4 enclosure reportedly matched or exceeded some premium retail options. The article matters because it frames current external SSD pricing as unusually expensive and positions enclosure-based builds as a way to avoid some of that inflation. That does not change NAND supply directly, but it does show how elevated finished-drive pricing is influencing buyer behavior and extending the useful life of older storage hardware. For RamTrend, the clearest takeaway is that price pressure in external SSDs is strong enough to shift demand toward reuse, enclosures, and lower-cost component combinations.

VAST Data said its broader collaboration with AMD now covers a reference stack for AI training and inference built around 6th Gen EPYC CPUs, Instinct GPUs, ROCm, networking hardware, and the VAST AI Operating System. The companies are positioning the platform for persistent inference and agent-style workloads that need large context windows, heavy data movement, and higher concurrency than traditional training clusters. The most relevant memory-market angle is KV cache offload. Instead of holding all inference context inside limited GPU memory, the design uses shared high-performance storage so active workloads can keep more GPU capacity available while still retaining prior context. VAST also said early testing with an Instinct MI355X system showed faster time to first token and higher token throughput when cache data was offloaded, although those results depend on system configuration and workload conditions. The hardware roadmap adds PCIe Gen6 through AMD EPYC 9006 support and uses NVMe SSD-based storage clusters plus AMD Pensando Pollara 400 networking to move data between GPUs and shared storage. If this model gains adoption, it would strengthen demand for low-latency enterprise SSD capacity and storage fabrics in AI deployments, even if it does not directly change DRAM pricing in the near term.

Tom’s Hardware reviewed Framework’s Laptop 13 Pro and noted that the DIY configuration relies on LPCAMM2 memory on certain Intel-based models rather than standard SO-DIMMs. The review said the memory was harder to source than conventional modules and reported that Framework’s pricing for the tested memory configuration nearly doubled during the test period, rising from about $849 to about $1,600. For RamTrend, the main takeaway is not the laptop itself but the pricing and supply behavior around LPCAMM2. New memory form factors often face thin supplier ecosystems, weaker price competition, and limited spot availability compared with mature notebook memory standards. That can create unusually high end-user pricing even when broader DRAM markets are not moving by the same magnitude. The item suggests that LPCAMM2 remains an immature channel market with elevated buyer risk, especially for repairable or modular laptop designs that depend on less common memory options.

Tom's Hardware highlighted a Newegg bundle built around AMD's new Ryzen 7 7700X3D that includes an MSI Pro X870-P motherboard, one 16GB G.Skill Ripjaws M5 DDR5-6000 module, and a free Cooler Master 240mm liquid cooler for about $588. The reported package discount is $233, which makes the included memory effectively cost around $16 within the deal structure. For RamTrend, the main significance is at the channel and retail level rather than the upstream memory supply chain. The offer suggests continued promotional pressure in enthusiast PC builds, with memory used as a value lever to move CPU and motherboard inventory. Because the bundle includes only a single 16GB DIMM, it is more of an entry configuration than a full dual-channel setup, but it still signals a highly competitive consumer DDR5 environment. This kind of promotion can temporarily anchor buyer expectations lower for mainstream DDR5 kits, even if it does not directly imply a structural drop in DRAM contract pricing.

ASUS has introduced a new server lineup based on 6th Gen AMD EPYC 9006 processors, spanning dual-socket and single-socket designs for enterprise, AI inference, and cloud workloads. The higher-density systems are described as supporting 32 DIMM slots and MRDIMM configurations, alongside PCIe Gen6 and large E3.S NVMe storage footprints. From a memory-market perspective, the main takeaway is sustained platform support for advanced DDR5 deployments in performance-sensitive infrastructure. ASUS did not disclose pricing or an availability date, so the announcement does not provide a direct pricing signal, but it does reinforce ongoing enterprise demand for high-capacity, high-bandwidth server memory.

MSI and Colorful distributor pricing in China shows broad increases across Nvidia's RTX 50-series lineup, with some models moving sharply higher versus official MSRP and MSI's prior listed prices. The source ties the latest increases to higher VRAM costs and a tighter component environment, with GDDR6 and GDDR7 memory cited as part of the pressure. For memory-market watchers, the key signal is not just higher GPU shelf prices but the claim that memory content is materially contributing to those increases. If graphics memory costs remain elevated, board partners may continue pushing through higher prices on cards with 8GB, 12GB, 16GB, and above, reinforcing inflation across gaming and AI-adjacent hardware segments.

ChangXin Memory Technologies closed its first trading day in Shanghai far above its IPO price, raising fresh attention around China’s expanding DRAM industry. The company disclosed major spending on DRAM process upgrades, next-generation DRAM research, and wafer line improvements, with no dedicated HBM project named in the prospectus. That matters because CXMT is adding meaningful DRAM capacity while still trailing leading HBM suppliers in yield and cost structure. If its expansion stays focused on standard and server DRAM, the likely effect is stronger competition in conventional memory segments rather than a near-term shift in high-bandwidth memory supply. The article also points to already-booked output and large server DRAM agreements, suggesting capacity is being absorbed by major buyers even as CXMT scales.

CXMT completed a major IPO on Shanghai's STAR Market on July 27, raising roughly CNY57.9 billion, or about US$8.6 billion. According to the source, the listing is the largest IPO so far on that board and gives the company additional resources to expand manufacturing capacity, improve process technology and speed up new product work. That matters for RamTrend because new capital at a leading DRAM producer can support a broader spending cycle beyond the chipmaker itself. The reported knock-on effect includes stronger demand for semiconductor equipment, materials, packaging services and memory-module suppliers tied to China's DRAM ecosystem. The article does not provide a production timetable or output targets, so the near-term effect on DRAM pricing remains uncertain. Even so, the funding strengthens CXMT's ability to scale and could increase competitive pressure in future memory supply if expansion plans convert into meaningful wafer output.

Advantech unveiled a broad set of AI and edge infrastructure platforms based on AMD's 6th Gen EPYC 9006 processors, covering GPU servers, modular data-center systems, edge deployments, network appliances and server boards. The lineup targets AI, HPC, storage, telecom and industrial workloads rather than a single niche deployment model. For the memory market, the notable point is the platform feature set: support for DDR5-8000, MRDIMM up to 12,800 MT/s and CXL 3.1 memory expansion, alongside high-density E1.S and E3.S NVMe SSD storage. That combination increases the addressable demand for advanced server memory and storage components in systems designed for data-intensive AI and edge workloads. The announcement does not quantify purchase volumes or immediate supply commitments, so the near-term pricing effect is limited. Still, the product direction reinforces broader infrastructure demand for high-bandwidth server DRAM configurations and enterprise SSD capacity tied to AI build-outs.

TerraMaster says its new D8 Hybrid 2 combines four M.2 NVMe SSD slots with four SATA HDD or SSD bays in one USB 3.2 Gen 2 enclosure. The company positions the unit for creators, backups, media libraries and NAS expansion, with support for up to 152 TB depending on the installed drives. It lists pricing at $299.99 in the United States and £299.99 in the UK. The company also says the enclosure can reach up to 1020 MB/s with a single NVMe SSD and up to 980 MB/s from a four-drive HDD array depending on RAID setup. For RamTrend, the significance is limited but real: products like this reinforce demand for hybrid storage architectures that pair flash for active workloads with larger-capacity media for archive use. That is more relevant to SSD attach rates and storage mix than to near-term memory pricing.

Nvidia shared additional technical information about its Vera data center CPU, including unofficial SPEC CPU 2026 comparisons and a detailed look at the chip’s Olympus core. From a memory-market perspective, the key disclosures are the use of SOCAMM2 LPDDR5X, support for up to 1.5 TB of capacity, speeds up to 9600 MT/s, and aggregate memory bandwidth of up to 1.2 TB/s. The company is positioning Vera as part of a broader AI infrastructure stack rather than as a standalone CPU challenge to mainstream cloud server processors. That matters for memory markets because it reinforces Nvidia’s preference for tightly integrated, high-bandwidth memory subsystems inside AI platforms. The announcement does not directly change DRAM pricing, and the benchmark claims are still unofficial. Even so, it adds to the case that advanced low-power memory formats and high-density server memory designs will remain important in AI system architecture, especially where power efficiency and bandwidth are critical.

A new Tom's Hardware report points to Chinese retail listings showing 64GB DDR5-5600 RDIMMs built with CXMT memory priced close to comparable modules based on Samsung and SK hynix chips. In the cited example, the CXMT version was slightly more expensive than similar branded alternatives on JD.com, which runs against the expectation that a new Chinese DRAM source would automatically create a low-cost option. The immediate takeaway for memory buyers is that extra chip supply does not necessarily translate into lower module prices at retail. Module validation, distributor margins, seller strategy, and tight market capacity can absorb any upstream cost advantage before it reaches end customers. If these early listings are representative, CXMT's arrival may improve sourcing flexibility for manufacturers before it materially changes street pricing for registered DDR5 memory.

LG Chem said it has started supplying semiconductor stripper material to Amkor Technology for use in a new production line, marking the company’s first commercial deployment of this product in semiconductor manufacturing. The material is used to remove photoresist and residue after circuit patterning, and LG Chem says the customized version for Amkor cuts removal time by about 50% versus existing products after passing qualification for volume production. For RamTrend, the main significance is indirect: the deal adds capacity and competition in advanced packaging materials at a time when AI investment and HBM demand are pushing more activity into high-end packaging flows. That does not directly change DRAM or NAND pricing today, but it supports the broader supply chain needed for memory products that depend on advanced packaging and substrate-level process control.

Developers and hardware modders have extended monitoring support for NVIDIA's GeForce RTX 50 series so software can read temperature data from individual GDDR7 memory modules. The update builds on earlier work that exposed additional Blackwell sensor data and now reaches thermal sensors embedded in the memory ICs. The source highlights the RTX 5090 as an example, describing a 16-chip GDDR7 layout with 2 GB per module. It also notes that board designs using memory on both sides of the PCB may not map cleanly to the same number of readout points. For RamTrend, the main significance is better operational insight into GDDR7 behavior under load, which may help enthusiasts, overclockers, and board partners evaluate cooling and memory performance. The item is more relevant to product monitoring and validation than to near-term memory pricing.

DigiTimes reports that Samsung Electronics intends to break ground in October 2026 on an expansion of its Onyang semiconductor packaging facility in South Korea. The project is valued at KRW 1.3 trillion, roughly US$884 million, and is aimed at supporting high-bandwidth memory mass production from May 2029. For RamTrend, the message is straightforward: HBM competition is not only about wafer capacity, but also about the packaging infrastructure needed to bring advanced memory to market. Because the planned production start is several years out, the pricing effect is more relevant to long-term supply expectations than to current spot conditions.

According to a DigiTimes item citing The Elec, Samsung is preparing a die-to-wafer hybrid bonding production line with roughly 50 bonders at its Pyeongtaek campus. The line is intended for next-generation high-bandwidth memory and logic chips. For RamTrend, the significance is clear: hybrid bonding is an important manufacturing step for advanced memory packaging, so added tooling plans can signal Samsung's intent to strengthen its position in HBM for AI-driven demand. The item does not establish an immediate price change, but it does point to continued investment in the production technologies needed to support future high-end memory output.