The resilience of global goods trade amid tariff escalation and an energy shock is real, but it rests on a narrow base. A breakdown of ITC Trade Map export data shows that a handful of AI hardware categories, led by servers, processors and memory chips, account for well over a third of the growth in world exports, and that the gains are flowing overwhelmingly to a small group of East Asian exporters and their North American customers.
Tesla CEO Elon Musk first said on X that the capacity of the AI5 chip memory would be cut in half to 72GB, but later clarified that AI5 will use 96GB of LPDDR5 instead. The change comes as Tesla seeks to avoid becoming the only company using what Musk described as a minimum-memory LPDDR5 configuration.
Elite Semiconductor Microelectronics Technology (ESMT) reported September consolidated revenue of NT$7.461 billion (approx. US$234 million), up 470.23% from NT$1.308 billion a year earlier but down 5.62% from August. For the first nine months of 2026, revenue reached NT$43.363 billion, up 343.91% year-on-year.
AI data center demand for LPDDR5X DRAM has continued to expand through 2026 and is expected to remain strong into 2027, tightening supply for PCs and smartphones and keeping device prices under pressure.
Samsung Electronics is reportedly seeking 2027 prices for its next-generation high-bandwidth memory (HBM) that are more than three times the price of its current HBM3E product, industry sources said. The increase would come as its main HBM product moves from HBM3E to HBM4. Analysts said the shift should further boost the company's profitability, as it leans on what it calls industry-leading performance.
Samsung Electronics plans to buy DRAM from Taiwan's Winbond Electronics starting in 2027 and is considering using it in about half of its televisions, South Korean publication The Bell reported, citing industry sources. The move comes as major memory makers phase out the LPDDR4 and LPDDR4X chips that TVs still depend on.
As the number of AI computing transistors integrated within a single CoWoS package rises rapidly, DIGITIMES observes that moving massive volumes of data to compute chips fast enough is becoming increasingly critical to fully utilizing available computing power.
Nanya Technology is stepping up investment in new factory capacity to expand packaging and testing as some customers move into trial production for specification-based designs. The company said it has won approval from the National Science and Technology Council's (NSTC) park review committee to set up a 3D wafer packaging and testing plant in Pingtung Science Park, aiming to capture demand for high-performance, high-bandwidth memory solutions driven by fast-growing AI applications.
Climax Technology said elevated memory prices are pressuring its gross margin in the second half of 2026, while longer customer validation cycles for substitute memory parts are also dampening its outlook. The wireless communications company said its overall second-half outlook remains conservative.
Semiconductor startup Volantis has raised US$88 million in Series A funding to develop an AI inference architecture that uses photonic interconnect technology to address memory bottlenecks in AI chips. The round, announced on October 1, 2026, and reported by Reuters, is aimed at breaking through the memory wall that has limited performance in current systems.
Micron Technology's revenue and profit for the fourth quarter of fiscal 2026 both beat expectations, and the company expects capacity supply to be even tighter in 2027 and 2028 than in 2026. Donghui Lu, vice president of Micron's DRAM front-end manufacturing operations in Asia and chairman of Micron Taiwan, said AI is driving strong memory demand and the company is expanding across its four major manufacturing bases in Taoyuan, Taichung, Tongluo, and Tainan.
Facing a prolonged memory supply crunch, Tesla CEO Elon Musk has announced steep cuts to the memory specifications of the company's AI5 and AI6 chips to secure enough memory for Optimus mass production. Musk said the downgrades will not affect the humanoid robot's performance.
AI-driven memory demand is surging, and Micron has recently accelerated capacity expansion across multiple regions worldwide, including Taiwan, the US, Japan, Singapore, and India. Donghui Lu, vice president of Micron's DRAM Asia frontend operations and head of Micron Taiwan, said the company's capex for construction over the next few years will be enormous, ranking it among the world's biggest construction investors, and stressed the keyword "together" as it works with Taiwanese equipment and materials suppliers to enter overseas markets.
Micron is using an externally manufactured base die in NVHBM, a custom high-bandwidth memory (HBM) product it is co-developing with Nvidia, as the company bets that customization can unlock added value even when manufacturing is outsourced. The move underscores how HBM competition is expanding beyond memory stacks into logic circuitry and system integration.
The Science Park Review Committee under Taiwan's National Science and Technology Council (NSTC) approved 13 investment projects at its 35th meeting on the 1st, including Nanya Technology's plan to enter the Pingtung area of the Southern Taiwan Science Park and build advanced 3D semiconductor packaging and testing operations.
Samsung Electronics is expected to report another operating loss at its Mobile eXperience and network business in the third quarter of 2026, even as strong Galaxy Z foldable sales support handset demand. The preliminary earnings release is due in early October 2026, with the pressure coming from sharply higher memory prices and limited ability to raise device prices.
Topco Scientific said its Japanese subsidiary, Shunkawa, officially opened its Fukuoka office on October 1, making it the company’s third base in Japan alongside its Tokyo headquarters and Kumamoto office. The move is designed to leverage Fukuoka’s location and industrial clustering to strengthen real-time technical support and new business development in Kyushu’s semiconductor hub.
Micron Technology has sharply raised 2027 prices for high-bandwidth memory (HBM), the stacked DRAM packaged with AI accelerators. The increase comes after a year in which fixed contract prices left HBM largely outside the memory price surge.
Applied Materials announced that Kioxia will join its Equipment and Process Innovation and Commercialization (EPIC) Center in Silicon Valley, California. The two companies will collaborate on developing advanced memory cells and structures, multi-chip stacking architectures, and new materials engineering technologies to improve the memory density and performance required in the artificial intelligence (AI) era.
Micron told investors on September 30 that it cannot say when memory supply will catch up with demand. It expects the shortage to get worse over the next two years, even though additional cleanroom space is expected from late 2028.
Applied Materials and Kioxia are joining forces at a Silicon Valley research hub to develop next-generation memory technologies that could shape AI systems used worldwide. The collaboration underscores efforts to boost memory density, speed, and energy efficiency, which are increasingly important as global demand for data-intensive computing and advanced chip packaging rises.
The memory market is entering the fourth quarter of 2026 with a widening split between weak consumer demand and intense server buying, underscoring how artificial intelligence workloads are reshaping supply. Spot prices for DDR5 server modules continue to rise, while some lower-grade products are softening in the channel.
Micron Technology's fiscal fourth quarter showed a memory maker whose growth now comes mainly from price rather than volume, with customers committing cash years in advance to secure supply. Revenue for the three months ended September 3, 2026, reached US$54.2 billion, up 31% from the previous quarter, and GAAP gross margin widened to 86.8%.
Samsung Electronics' HBM4 ramp is giving its foundry business a timely boost, as rising demand for the memory product translates into more work for the company's 4nm logic lines.
