On September 23 (local time), the U.S. International Trade Commission (ITC) voted to institute a Section 337 investigation into certain dynamic random access Memory (DRAM) devices and downstream products and components (III).

This marks YMTC’s first substantial injunction victory worldwide since the comprehensive patent battle between the two companies erupted in November 2023.

With the Mid-Autumn Festival and National Day holidays approaching, some Memory industry players have already entered vacation mode, leading to a noticeable decline in market activity this week. Many manufacturers believe that even active price reductions would struggle to stimulate demand, so most have chosen to hold current prices steady. The limited success of modest price concessions by a few firms under performance pressure—evidenced by continued shrinking transaction volumes—further confirms the weak market conditions.

According to a report by Korean media outlet ZDNet, Solidigm, a subsidiary of SK Hynix, is considering constructing its first NAND flash Memory production base in the United States. The company is currently in the stage of coordinating detailed plans, including factory site selection and investment timing.

Lim Eui-cheol, Vice President of SK hynix in charge of Solution AT, recently delivered a presentation at the "AI Infra Summit" held in the United States. He stated that as AI workloads become increasingly diverse, the existing GPU-HBM architecture alone is no longer sufficient to meet new demands. SK hynix is preparing new solutions combining software and hardware, such as PIM (Processing-in-Memory), HBF(High Bandwidth Flash), and SALT-KV (Semantic-Aware Lifecycle Tiering for KV Cache), to address the needs of these evolving workloads.

At Huawei Connect 2026, Huawei Rotating Chairman Wang Tao announced that the R&D progress of the Ascend 960 chip has exceeded expectations, with its overall performance doubling and its release schedule moved up.

According to media reports, Micron expects to assemble and test tens of millions of chips in 2026 at its first semiconductor assembly and test plant in India (the Sanand plant), and to expand capacity to hundreds of millions of units in 2027. Micron CEO, Sanjay Mehrotra, said that the plant's current output capacity already exceeds the Memory required by India's laptop market. Finished DRAM and NAND products were now leaving Micron's facility in Sanand, Gujarat, for customers around the world with the "Made in India" label.

According to media reports, Nebius will raise prices for some of its on-demand computing resources starting October 1, marking the company's second round of price increases since May.

According to Korean media reports, Samsung plans to break away from its previous model of relying entirely on its in-house foundry division for Base Die production. Instead, it is adopting a “dual-track” strategy that also leverages TSMC’s process technology to manufacture Base Dies.

According to Korean media reports citing industry sources, Samsung has decided to outsource the increased production of DDR5 modules and SSDs for PCs, servers, and laptops entirely to Outsourced Semiconductor Assembly and Test (OSAT) partners, rather than expanding its in-house capacity. This move aims to concentrate limited internal resources on high value-added Memory products like HBM.

According to media reports, Intel CEO, Lip-Bu Tan, expects the Memory supply shortage to become more severe next year due to growing AI demand. He noted that some Memory prices have already climbed five to seven times and many corporations are slow to expand their businesses because they can't secure enough Memory. He explained that as demand for high bandwidth Memory (HBM) for AI surges, pressure is growing on the supply of commodity DRAM as well. In addition, Lip-Bu Tan stressed the importance of advanced packaging technology, emphasizing that CPU, AI accelerators, HBM, and I/O chips should be connected within a single package to shorten data travel distance and improve processing speed and power efficiency. Intel is targeting the AI Semiconductor market with its own advanced packaging technology, EMIB.

Micron Technology has announced the successful demonstration of the world's first 512GB DDR5 module on multiple server platforms. Leading semiconductor enablers, AMD and Intel, are both actively validating the module, which will deliver speeds up to 9,200 MT/s. Micron expects 512GB RDIMMs to be in volume production sometime in the second half of 2027 aligned to customer needs.

As we approach the end of the third quarter, resource prices have been on a continuous upward trend for nearly three months. Consequently, prices for industrial-grade DDR5 secondary-grade dies and embedded LP resources have surged to historical highs. However, for finished products like industrial DDR5 SODIMM and LP4X/5X, customer price tolerance is nearing its limit. Faced with shrinking demand, some Memory manufacturers have begun to slow down their resource procurement. The state of "low demand, low stocking, and low shipments" is likely to persist for some time.

Citing mobile communications industry sources, South Korean media reported that a certain mobile carrier has notified retailers that the retail prices of all Galaxy S26 series models will be raised uniformly by 149,600 Korean won starting October 1. Industry insiders believe that Samsung Electronics may find it difficult to maintain its existing pricing structure due to recent rise in mobile application processor unit price, the burden of Memory semiconductor cost brought by high-performance on-device AI feature, and the continued impact of high exchange rates. Samsung Electronics is said to be cautious about officially confirming the retail prices of the Galaxy S26 series.

DeepSeek V4.1 Flash model has recently officially released. It is the smallest model in DeepSeek's new model architecture series and features native multimodal visual understanding capabilities. DeepSeek V4.1 Flash is a 552B-parameter MoE model and significantly reduces the size of the KV Cache. Compared with the previous generation model, demand for HBM has been reduced to 1/4, and demand for SSDs has been reduced to 1/8. In Agent usage scenarios, cache hit costs often account for a relatively high proportion, and the compression of the KV Cache substantially lowers the usage cost of Agent-type tasks.