Processors and graphics cards, memory and storage, chips and fabs, servers, networking and robotics — the hardware behind everything else.
LG Display launched a 24.5-inch 720Hz gaming OLED panel, while Samsung Display unveiled a 31.5-inch QD-OLED with 4K and 360Hz. Samsung invested 4.1 trillion won in 8.6-generation IT OLED lines, expecting laptop OLED penetration to grow from 6% in 2026 to 19% by 2030.
OData is deploying DeltaFlow, the proprietary liquid cooling system from parent company Aligned Data Centers, at its SP04 facility in Brazil and QR03 in Mexico. The project will cost $630 million across both sites, with phase one installation and testing already underway. The system enables hybrid air and liquid cooling for high-density AI racks.
Aolani said 22,000 Nvidia Blackwell Ultra GPUs are expected to be deployed across two sites in Southeast Asia. The rollout is slated for early 2027.
The International Federation of Robotics estimates around 7,000 autonomous humanoid robots were sold worldwide in 2025 for industrial and professional service use. Many were bought for research rather than productive work.
Leakers Kopite7kimi and Kepler_L2 claim Nvidia will not launch its RTX 60 graphics cards on GR20x "Gaming Rubin" chips until 2028, refuting reports of an RTX 6090 in 2027. The delay is tied to the DRAM shortage, expected to peak in mid-2027.
ABB announced Infinitus, a solution for AI data center infrastructure. The company expects Southeast Asia's data center capacity to grow from 2.8 GW to 9.4 GW by 2035.
Realtek has unveiled the RTL8773J family, the first Bluetooth 7-ready microcontrollers. It pairs a 200 MHz Cortex-M55 core with a Cortex-M33 core, up to 2MB SRAM, 8MB PSRAM and 16MB flash, and supports HDT at up to 7.5 Mbps plus Realtek HDT+ at up to 15 Mbps.
U.S. cloud service providers have raised their enterprise SSD demand forecasts over the past two weeks, accelerating deployments and expanding use across more applications, including QLC drives. TrendForce expects 4Q26 orders to surpass the already high 3Q26 levels, giving NAND suppliers more control over capacity allocation and pricing.
Lattice Semiconductor unveiled its next-generation secure control FPGA series, Mach-N2, built on a 16nm FinFET process. The chips integrate post-quantum encryption to strengthen control-system security.
Leading Interconnect Semiconductor, a subsidiary of Zhen Ding Technology, has started construction of an advanced packaging substrate plant for AI in Bao'an, Shenzhen. Total investment is expected to reach at least CNY12 billion (about US$1.79 billion).
AgiBot introduced AGILE 2.0, an end-to-end perception and control model combining vision, terrain understanding, traversability planning, and whole-body motion. Demos on the Lingxi X2 platform include jumping through a ring of fire, diabolo juggling, and balancing on a large ball. Exact metrics and latencies were not disclosed.
At PrimeBot's presentation in Shanghai on September 20, Suzhou-based chip designer Ricore was named a computing partner for consumer robots. The flagship Radiant D35 chip combines a custom RISC-V core with a dataflow architecture for edge inference of physical AI and is delivered in chiplet packaging.
At a conference in Shenzhen, CIX Technology announced the AGX X2 (AGXcelerator) platform for local inference of AI agents: 80–300 TOPS, 64–160 GB unified LPDDR5 memory, support for models from 7 to 122 billion parameters. It claims energy consumption per token reduced by more than 50% relative to industry baselines.
MediaTek announced the flagship Dimensity CX C10 Max chip for laptops, tablets and convertibles. It has eight big cores with no efficiency cores, a 55 TOPS NPU, an AI ISP and a TSMC 3 nm N3 process. MediaTek claims up to 15% higher multi-threaded performance and up to 19 hours of battery life.
South Korean AI chip developer Rebellions expanded its NPU use to four commercial SK Telecom services. The infrastructure handles about 14 million real-time requests and over 4 billion tokens per day.
A day before Snapdragon Summit, a Chinese blogger showed an engineering sample of Qualcomm's first 2nm chip, Snapdragon 8 Elite Extreme Gen 6. In Geekbench 6, it scored 4327 single-core and 12873 multi-core, trailing A20 Pro in single-core but beating it in multi-core.
The shift of AI from data centers to cars, robots, and wearables is increasing demand for FD-SOI technology. China is building its own supply chain from substrates to fabs and chip design.
Tower Semiconductor will expand silicon photonics, SiGe and advanced packaging capacity in the Hokuriku region, making Japan its largest production base for optical communication chips. The move is driven by growing demand from AI data centers.
TSMC plans to launch four fabs for its 1.4nm process in Taiwan's Central Science Park. Pilot production starts in Q3 2027, mass production in 2028, reaching 120,000-160,000 wafers per month. Estimated wafer price is $35,000, only slightly above $33,000 for the 2nm node.
ASRock has assigned final names to its Z990 chipset motherboard models and started shipping them, designed to support Intel Nova Lake desktop processors. ComputerBase reported the news.
According to Reuters, Chinese memory maker CXMT is preparing to expand from DRAM into NAND flash memory used in SSDs and storage drives. The information is based on three sources familiar with the company's plans.
NVIDIA GPUs are reported to include dozens of embedded RISC-V cores that handle tasks outside shader compute, including functions related to graphics driver operation.
TSMC is building a validation hub for advanced packaging in southern Taiwan. Rapid CoWoS expansion is pressuring equipment and materials suppliers to certify new technologies faster.
SK hynix is developing a 3D DRAM architecture in which memory cell processing is separated from control peripheral circuits, allowing each part to be optimized with a separate manufacturing technology.
Kyocera has completed construction of a plant in Isahaya, Nagasaki, a key site for its semiconductor business. Production of fine ceramics for equipment and chip packages will begin in spring 2027, with investment of about ¥25 billion (€139.8 million).
Chinese chipmaker Hygon is preparing to release an AI chip for robotics. The launch slogan is "Let computing power reach the physical world." Specifications and the announcement date were not disclosed.
Secure Data Recovery analyzed more than 5000 failed HDDs: median operating time was 8687 hours (about 362 days), and 49.8% of drives failed in under a year. The sample includes only drives that already failed, so it does not reflect the real service life of a typical HDD.
Samsung Electronics is preparing to at least double production of HBM4 memory in 2027 amid a persistent memory shortage. The company is accelerating its shift to more expensive products.
The chairman of Wiwynn said that AI hardware design and deployment are driven by growing demands for power, cooling, and interconnect speed, not just compute chips. This is changing the entire server supply chain.
ByteDance plans to spend about 100 billion yuan ($14.9 billion) annually on chip substrates, according to Digitimes. A shortage of advanced packaging substrates is limiting China's AI chip output.
