KB Securities estimates the global semiconductor market will reach about 2000 trillion won (~$1.44 trillion) this year, with the memory market growing more than threefold to 1100 trillion won (55% of the total market). AI infrastructure investment next year will rise 63% to 1755 trillion won, with memory’s share increasing from 40% to 57%.
Counterpoint Research: global foldable smartphone panel shipments in the first half of 2026 fell 13% year-over-year. The decline is concentrated in the clamshell segment, with the market shifting toward in-fold designs.
Japan's Resonac has grown and processed a 300mm silicon carbide (SiC) single crystal, a step toward mass production of power electronics for electric vehicles, energy, and data centers. Moving from 200mm to 300mm increases chip yield per wafer.
Samsung Electronics is deepening ties with Japan's semiconductor ecosystem, developing large-format 3.xD chiplet packaging for HPC and AI processors. The company is relying on Japanese materials, equipment, and engineering resources amid intensifying chip packaging race.
Gallium nitride (GaN) is becoming a key power semiconductor for AI data centers: rising rack power and the shift to higher voltages require improved efficiency and power density. The technology still lags competitors on cost.
At Huawei Connect 2026, the company introduced the Agentic supercluster architecture, enabling scaling of Ascend compute to 1 million accelerators via UnifiedBus. This expands Huawei's strategy beyond individual chips and storage systems.
Samsung Display, according to Digitimes, is preparing investments of about KRW300 billion in a 12-inch RGB OLED-on-Silicon (OLEDoS) production line. The microdisplay technology is approaching mass production amid demand from XR devices.
Nvidia revealed its AI platform roadmap: mass shipments of Vera Rubin will begin in the second half of 2026, followed by Vera CPU, Rubin GPU, and NVL4, NVL8, and NVL72 platforms. The company is shifting focus to rack-scale solutions.
Apple announced the 2nm A20 Pro chip for iPhone 18 Pro, while MediaTek announced the 2nm Dimensity 9600 Pro. Both manufacturers are betting on agentic AI features instead of chasing peak performance.
Seagate published its Data Infrastructure Readiness Report 2026: enterprise AI strategies are moving from pilot deployments to mass rollout, increasing pressure on data storage and processing infrastructure.
The Taoyuan government plans to create a 40-hectare high-end IC substrate zone in the first phase of the Taoyuan Aerotropolis industrial park. A memorandum of understanding will be signed with substrate makers Unimicron and Kinsus amid growing AI demand.
A team led by Ethan Secor at Iowa State University is developing aerosol jet printing to apply electronic circuits on complex curved surfaces with 10–100 micron precision. The print head is mounted on a robotic arm, and the process is controlled by software and real-time monitoring. Startup Contour Circuits received a $150,000 NASA contract, and the research is supported by NSF.
Chinese company UBTech launched the world's first humanoid assembly factory in Liuzhou: 14,000 square meters, up to 10,000 robots per year, one robot every 10 minutes. The production uses Walker S and Cruzr robots, as well as unmanned carts and an automated warehouse for 112 robots.
China installed 54% of all new industrial robots worldwide in 2024 and produces 94% of the magnets used in them. Senator Dave McCormick said assembling in the US does not solve dependence on Chinese components.
Marvell announced an expanded partnership with GlobalFoundries to increase production of silicon-germanium wafers for optical transceivers and laser modules at its Burlington, Vermont fab. The technology has been validated at up to 200 Gbps per channel for 1.6 Tbps transceivers, with plans for 3.2 Tbps.
UK startup Vsim from Cambridge developed a system for training robots in a virtual environment: walking and grasping skills are learned in minutes instead of days. The software is optimized for GPUs and runs onboard the robot, computing 20,000 scenarios per second ahead.
Anderon, IBM's new quantum chip fab, will receive $1 billion from the U.S. Commerce Department under the CHIPS Act. The funds are intended to accelerate quantum wafer production in the United States.
The memory shortage that began in late 2025 continues through 2026 with rising prices. Jolla expects normalization by 2028, while SK Hynix says demand will exceed supply after 2030. Framework and Jolla are betting on modular designs and discrete chips.
On July 5, 2026, in Incheon, teams B-Human and HTWK Robots played the first-ever RoboCup 11-on-11 humanoid match, with B-Human winning 4:0. In October 2026, Italy's first open RoboCup Humanoid Soccer tournament will be held at Maker Faire Rome with 3-on-3 and 5-on-5 matches.
Samsung is in talks to transfer DRAM production to third-party contractors to free up capacity for higher-priced HBM memory. Partners include Dreamtech in India, Hanyang Digitech in Vietnam, and SFA Semicon in the Philippines.
Delta Electronics announced an integrated power and cooling system for AI factories based on the NVIDIA DSX platform: 800 V DC power delivery with up to 98.5% efficiency and liquid cooling up to 3 MW. The solution combines energy storage, transformers, and rack systems to accelerate data center deployment.
EuroHPC selected Bull to build the Lumi-AI section at Finland's CSC center in Kajaani. The system, using AMD MI430X GPUs, will deliver a 10x performance gain in AI tasks and roughly 2x in FP64 versus the current Lumi-G on MI250X.
Arm is developing its first ready-made server CPU, the Arm AGI CPU, while continuing to sell IP licenses and Neoverse CSS. Work to turn IP into a physical product is underway at Arm's lab in Austin, Texas.
The Chinese Academy of Sciences developed GAA transistors with an on/off ratio above 500,000, theoretically enabling 3nm-level performance on older DUV lithography. However, MEOL process constraints limit SMIC to a density of about 137.8 million transistors per mm², matching TSMC's 5nm node.
Lenovo added the four-socket ThinkAgile VX850 V4 node for resource-intensive in-memory workloads, plus three deployment service tiers and Express bundles for Microsoft, Red Hat, SUSE and Nutanix. Availability dates were not announced.
Sapphire Technologies and Anax Power propose placing modular data centers directly on US gas pipelines: turboexpanders use pressure drops in pipes to generate electricity and cold coolant. Sapphire estimates this cuts energy and cooling operating costs by about 2%.
Lattice Semiconductor has opened free access to Lattice Prompt, an AI tool for FPGA design using natural language. It uses Model Context Protocol and RAG based on validated device documentation and covers the entire development cycle. A 10x productivity increase is claimed, with no independent confirmation.
Lightmatter joined Open CPX MSA and released the Passage L20 CPX module, the first Open CPX module for AI infrastructure with bidirectional optics. It transmits 12.8 Tbit/s (6.4 Tbit/s each direction) and halves the number of fibers and connectors in AI scale-up networks.
onsemi announced its Embedded Power Platform (EPP), an architecture where the silicon wafer forms the package base and combines silicon, SiC and GaN. The platform promises 3-5x higher power density; in a solid-state switch test the solution was 50% smaller and 20% cooler. Sampling for strategic customers starts in 2026, with Subaru among partners.
A HP EliteBook X G3i laptop with a 20-core Intel processor, likely from the Nova Lake family, was spotted in the SiSoftware database. It lists clock speeds of 3–4.6 GHz, 18 MB L3 cache, and four 4 MB L2 blocks.
