Citing people familiar with the matter, foreign media reported that with the launch of the iPhone Duo and the strong early sales of the Galaxy Z Fold 8 and Z Fold 8 Ultra, Samsung is accelerating the pace of advancing its new foldable phones. It is reported that the next-generation Galaxy Z TriFold may be released at the end of 2026 or early 2027. In addition, despite facing upward cost pressure, it aims to launch at nearly the same price as the first-generation Trifold. Previous reports stated that Galaxy Z TriFold2 was originally planned to be released at the end of 2027.

According to a report by Korean media, Heraldcorp, 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, and SALT-KV, to address the needs of these evolving workloads. PIM technology can improve the performance and efficiency of high-end service fast-decoding workloads and is an excellent alternative for overcoming existing limitation. HBF technology can provide high bandwidth and large capacity while improving the efficiency of AI services that need to handle long contexts. SALT-KV technology improves processing efficiency based on consideration of the characteristics and reusability of KV cache, and also has sufficient potential to lead the evolution of AI systems. SK hynix stated that it will enhance its technological competitiveness and actively respond to the diverse demands of the AI era.

According to media reports, as AMD relies on TSMC for the production of most of its core chips, and TSMC previously was reported to further raise foundry prices by early next year at the latest, AMD may pass on the increased foundry costs to downstream partners and consumers in order to maintain product gross margin. It is said that AMD has notified several customers of the planned price increase, with the affected products covering AI accelerator chips, consumer GPUs, and motherboard chipsets. Although the sources did not explicitly state whether CPU products will also see price increases, the market speculates that the Ryzen series processors, which are highly dependent on TSMC's advanced processes, may also follow suit with price increase in the fourth quarter. There is currently talk that a 10% price increase remains a possibility, though the actual increase is yet to be confirmed.

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. It is reported that Micron has sent the plant's first shipment of India-made memory modules to Dell Technologies for laptops manufactured and sold in India. 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. Under the new pricing plan, the price of H100 will rise from $3.85 per GPU hour to $4.50, an increase of about 17%; H200 will rise 20% from $4.50 to $5.40; B200 will rise nearly 19% from $7.15 to $8.50; and B300 will rise about 21% from $7.85 per GPU hour to $9.50. The changes also extend to CPU-only compute. AMD EPYC Genoa CPU rates will rise 25% to $0.015 per vCPU-hour from $0.012, while Genoa memory will increase about 41% to $0.0045 per GiB-hour from $0.0032. The ultimate impact on customers' bills will therefore depend on their configurations and usage.

Intel previously mentioned at an earnings conference that its 14A process would enter risk trial production in the second half of 2027, and was subsequently scheduled to enter mass production in the second half of 2028. According to media reports, Intel CEO, Lip-Bu Tan, recently revealed that the 14A process is ahead of its original schedule and will begin trial production in the first quarter of next year. It is expected that if the 14A process enters mass production, it will help expand the company's foundry business and further capture orders from TSMC.

According to media reports, ADATA announced through its Malaysian subsidiary that it will purchase mechanical equipment worth no more than $105.219 million, with the equipment to be used for operational production. In addition, another expenditure covers the lease of data center space, power and other related infrastructure at the Skyvast Ground Zero Campus, with a capacity of approximately 2.016 MW, a lease term of five years starting from the date of installation completion, and a total transaction amount of no more than $19.3536 million. It is said that ADATA's investment this time is mainly aimed at building out AI computing capacity. The company estimates that it will begin contributing to its operation starting in 2027, with the benefit expected to last for three to four years.

According to media reports, OpenAI plans to hold an initial public offering (IPO) next year and is seeking a valuation of at least $1.5 trillion as it draws in new investment. It is reported that investors recently proposed investing in OpenAI at a valuation of $1.2 trillion, but OpenAI believes its valuation should be higher. OpenAI is pushing up its valuation mainly due to the growing number of customers for its coding tool Codex and its latest AI models, GPT-6 Astra and GPT-5.6 Sol. People familiar with the matter said the negotiations are still at an early stage, so the investment size and terms have not yet been finalized.

According to media reports, Meta is collaborating with Broadcom on chip design and with TSMC on manufacturing its custom AI chips, aiming to build out its AI infrastructure and reduce its reliance on Nvidia. Meta is reportedly testing its third-generation in-house product, the MTIA 450 (Arke), which has already successfully run Meta's own AI models as well as models from DeepSeek and Alibaba. The next-generation product, the MTIA 500 (Astrid), is expected to complete its design in about a month. Under the current plan, the MTIA 450 currently being tested will enter data centers in the first half of 2027, while the next-generation MTIA 500 is expected to be deployed before the end of 2027.

Citing foreign media reports, ByteDance's revenue grew about 30% year-on-year to $120 billion in the first half of 2026, but net profit fell by a single-digit percentage from a year earlier to $20 billion, mainly because massive spending on AI infrastructure weighed on profitability. Earlier reports said ByteDance was considering investing as much as $70 billion in data centers and other AI infrastructure in 2026, though that figure is only a potential spending ceiling and the company has not confirmed a budget or released audited 2026 capital expenditure data. ByteDance has also completed a bank loan of about $30 billion to support research and development recently. In addition, people familiar with the matter said that in the first half of this year, overseas revenue accounted for more than 30% of ByteDance's total revenue, the vast majority of which came from TikTok. By comparison, that share was only 25% in 2024 and is expected to be around 30% for the full year of 2025.

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. 

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 prices, 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. A Samsung official said that while the company is indeed internally considering whether to adjust prices due to cost pressure, no final decision has been made yet.

NVIDIA has officially released the RTX PRO 5500 Blackwell workstation GPU on its website, designed specifically for rack-mounted workstation deployments. RTX PRO 5500 adopts the Blackwell architecture, featuring 21,760 CUDA cores and 84GB of GDDR7 memory. The memory supports error-correcting code (ECC) and delivers bandwidth of up to 1,398 GB/s. In addition, RTX PRO 5500 supports PCIe Gen5, which provides double bandwidth of PCIe Gen4, improving data transfer speeds from CPU memory.

Chen Ching-wen, general manager of Team Group, said the memory market remains persistently tight in terms of supply, with prices set to rise further in the fourth quarter, though the pace of increases will moderate compared with the past eight to nine months, shifting from monthly jumps of 15% to 20% to quarterly gains of about 20% to 30%. On the demand side, Chen believes DRAM demand is currently more solid than NAND Flash, with the DRAM shortfall being very pronounced and massive while NAND Flash also faces a shortage, but it is relatively less severe. On the supply side, capacity at memory manufacturers has already been largely locked up by long-term agreements, and the pace of expansion at new plants cannot keep up with demand. Chen said the industry has recently been receiving notices from upstream manufacturers clearly signaling that supply releases in 2027 will be sharply tightened, with capacity at those manufacturers almost fully booked through 2027-2028. It is predicted that the truly severe shortage may hit in the first half of 2027, when the volume of goods available externally will be very limited. DRAM is also Team Group's main product line, accounting for about 75% to 80% of its business, and its operating strategy for next year will focus on safety stock and customer screening.