According to Korean media reports, Samsung Electronics delivered a speech under the theme of "Expanding Innovation in the AI Era," proposing speed, memory expansion, and data persistence as three technology directions for AI infrastructure. Specifically, it will increase data processing speed by launching high-performance memory (such as HBM5 and custom HBM), achieve memory scalability through CXL memory modules equipped with DRAM (CMM-D) and memory pooling technology, and enhance performance and extend service life using next-generation SSD data storage and management technology.

AMD has introduced the AMD Helion agentic AI rack, an open blueprint for agentic AI infrastructure based on an Open Compute Project spec-design and powered by AMD compute. The rack is designed to provide partners and customers with an all-in-one solution for enterprise-grade agentic AI. Typical Helion configurations include the CPU-only rack and An all-in-one rack. Among them, a CPU-only rack powered by 6th Gen AMD EPYC™ processors that can support up to 32,768 CPU cores; an all-in-one rack that combines AMD EPYC CPU nodes, AMD Instinct™ MI350P GPU nodes and general-purpose compute on one platform, supporting up to 16 GPUs.

At the Open Compute Project (OCP) Global Summit today, Intel shared new technical details on the Intel GPU D-series, formerly code-named Crescent Island. Intel GPU D-series is designed to improve the economics of sustained, memory-intensive AI inference. Built on Intel's latest Xe3P architecture, the GPU combines up to 480 GB of LPDDR5X memory, 1.5 TB/s of memory bandwidth and 1.3 petaflops of FP8 compute within a 350-watt power envelope. Designed for large language models, long-context workloads and concurrent inference services, Intel GPU D-series supports models with up to 300 billion parameters on a single card and more than one trillion parameters in an eight-GPU system. 

According to industry sources cited by South Korean media, Samsung Electronics is internally developing three types of z-NAND flash memory. In addition to zNAND-O, for which its model has already been publicly demonstrated, these also include zNAND-T and zNAND-I. It is reported that zNAND-T is designed for high throughput, featuring maximized bandwidth to meet the needs of high-capacity, high-performance AI data centers. zNAND-I, meanwhile, maximizes IOPS. Although wider bandwidth allows large amounts of data to be processed at once, higher IOPS means more data I/O operations can be completed per second.

According to the South Korean media outlet ZDNET, SK hynix has introduced the H-TPU architecture. H-TPU is not a tensor accelerator that performs AI computing, but rather a programmable processor dedicated to controlling the SSD data plane and processing transactions. Its key feature is that it abandons the traditional approach of relying on a single large general-purpose CPU. Instead, it subdivides control functions and breaks the processor down into functional units distributed across dozens of small TPU clusters. This architecture maximizes efficiency by splitting the processor by function, thereby overcoming the power and area limitations of general-purpose CPUs. In addition, SK hynix expects that H-TPU-centric controller innovation, combined with HBF (High Bandwidth Flash) emerging as the next-generation memory, will produce strong synergies.

Supply chain sources say Intel has taken the lead in raising prices for multiple generations of PC processors since September, with high-end Panther Lake chips seeing increases of 15% to 20%, while Meteor Lake and Arrow Lake rose by about 5% to 10%. Among them, not the entire Panther Lake series saw simultaneous price increases. This round of adjustment was concentrated on products manufactured by TSMC's N3E process and equipped with high-end GPU tiles. Prices of Qualcomm's Snapdragon X / X2 Elite and other platforms rose by about 10% starting in October. AMD is also reported to be preparing to adjust the prices of some CPUs, with the extent of the increase, scope, and effective date still to be confirmed.

According to foreign media reports, Nvidia is in talks to deepen its investment in ​open-source startup Reflection AI or acquire it. It is reported that Reflection AI is a company founded in 2024 by former DeepMind ​researchers. On the 5th, it released its first open-weight ⁠model, Beam, ​as it seeks to compete in ​coding and agentic tasks with lower-cost Chinese models such as DeepSeek and ​Kimi. Nvidia is already a major financial backer and strategic investor in Reflection AI, having invested $800 million in the startup. The report says the two companies may sign an agreement within the next few weeks, but the negotiations could also fall apart. It added that although the terms of the agreement have not been disclosed, Reflection was valued at $25 billion during its fundraising in March last year.

According to media reports, Meta has recently presented an experimental memory compression solution called CRAM (Compressed RAM) at LPC 2026. The solution aims to keep compressed data in memory form, thereby reducing the software overhead associated with traditional Swap path, rather than simply replacing ZRAM or zswap. The core idea of CRAM is to provide hardware-compressed memory to Linux as a special type of NUMA memory, instead of emulating it as a block storage device. It is said that, in pure read scenarios, CRAM's throughput performance is up to 452 times higher than that of the current mainstream ZRAM, and its actual operating speed is already extremely close to native DRAM.

Micron Technology's board of directors has approved an expansion of its share buyback plan, effective in December, according to reports. In its 10-K filing, Micron stated that the board approved an increase in the maximum amount of common stock that may be repurchased at its discretion, effective December 9, 2026, to $35.16 billion. Micron had previously authorized a $10 billion share buyback program, of which approximately $2.2 billion remained unused. 

According to media reports, the entire Xiaomi 17 Ultra series saw a price increase today. The starting price of the standard version rose from 6,999 yuan to 7,999 yuan, while the starting price of the higher-positioned Leica version rose from 7,999 yuan to 8,999 yuan. Pricing for all related versions currently sold on Xiaomi's official website has been adjusted. Meanwhile, Huawei also adjusted prices for its nova16 series. The recommended retail prices of the nova16 and nova16 Pro were raised by 400 yuan across the board, and the nova16 SE by 200 yuan, with corresponding channel procurement prices also raised in tandem. Prices for the nova16z and nova16ultra remain unchanged. It is said that the core driver of this round of price increase is the continued surge in memory chip prices, which has sharply eroded the overall profit margins of smartphone companies. To cope with cost pressure and sustain operations, they have no choice but to raise product prices.

MediaTek's consolidated revenue for September was NT$55.399 billion, down 13.69% MoM and up 1.97% YoY. Cumulative consolidated revenue for January-September totaled NT$469.391 billion, up 5.30% YoY. Third-quarter consolidated revenue reached NT$168.058 billion, up 10.43% QoQ and 18.27% YoY, surpassing the upper end of its original guidance of NT$159.8 billion, beating the company's earlier expectation and setting a new single-quarter record.

After investigating recent rumors of reduced HBM capacity, Counterpoint Research predicts that most of the HBM4 used in Nvidia's next-generation AI platform "Rubin" will maintain a 12-layer configuration. Previously, industry insiders speculated that Nvidia had asked major HBM suppliers to reduce the number of stacked layers from 12 to 8. Some analysts even suggested that ASICs could be downgraded to 4 layers. However, Counterpoint Research analysis argues that in the field of AI computing, ensuring memory bandwidth and capacity is more important than reducing cost. They explained that HBM4 for the Rubin platform is in fact moving toward the 12-layer architecture, while 4-layer products have not been confirmed in the ASIC market yet.

Dell has unveiled the XPS 16 Creator Edition, the first XPS laptop equipped with NVIDIA RTX Spark. It is reported that this model offers two RTX Spark N1X configurations: the higher-end version features a 20-core Grace CPU and a 6,144-core Blackwell RTX GPU, while the lower-end version has an 18-core CPU and a 5,120-core GPU. In terms of memory, it is equipped with up to 128GB of LPDDR5X unified memory, with the CPU and GPU sharing the same memory pool.

Citing industry sources, South Korean media reported that Samsung's VD division, which oversees its TV business, has decided to procure Winbond's LPDDR4 and LPDDR4X chips starting from 2027, with related quality certification testing completed in the first half of 2026. Industry observers believe that as major manufacturers gradually phase out LPDDR4 and LPDDR4X production, supply of the previous-generation memory required for TV products is becoming increasingly tight, prompting Samsung to actively seek alternative supply sources. Notably, although Winbond is a new supplier for Samsung, it is said that the VD division is considering configuring about 50% of its TV products with Winbond DRAM in 2027.

TSMC released its September 2026 revenue report on October 8. Data showed that TSMC's consolidated revenue for September 2026 was approximately NT$511.86 billion, down 0.6% month on month and up 54.6% year on year; cumulative revenue for January to September 2026 was NT$3.90 trillion, surging 41.1% compared with the same period last year. According to statistics, cumulative consolidated revenue for the third quarter reached NT$1.49 trillion, rewriting a record high for a single quarter, up about 17.6% quarter on quarter and about 50.9% year on year.