Asia’s semiconductor scene was buzzing during the week of September 1–7, with breakthroughs in leading-edge process technology, pricing shifts that could ripple through the AI and HPC markets, and strategic moves by major players from Japan to China and the United States. Here’s a concise wrap-up of the headlines shaping chips, foundries, and the broader electronics supply chain.
Rapidus claims 2nm milestone on density, surpasses Intel on performance projections
Japan’s Rapidus is pushing hard into the 2nm race with its “2HP” process, and early numbers are eye-catching. Analysts say the node’s logic density is tracking at 237.31 million transistors per square millimeter—slightly ahead of TSMC’s estimated 236.17 million and well above Intel’s 18A at 184.21 million. The company has already taped out a gate-all-around test chip using ASML’s EUV lithography and met key electrical targets. Rapidus is preparing its IIM-1 fab for volume, aiming for around 25,000 wafers per month by 2027. If timelines hold, the Japanese challenger could become a credible alternative for cutting-edge logic.
TSMC plans 2026 price hikes for advanced nodes
Foundry pricing is set to climb in 2026, with TSMC preparing 5–10% increases across advanced processes. The steepest adjustments—around 10%—are expected for high-performance computing and artificial intelligence chips, where capacity remains tight and complexity is rising. While not officially confirmed, company executives have previously signaled the need to offset escalating costs tied to advanced manufacturing and supply chain dynamics. The move could influence device pricing and margins across the semiconductor ecosystem.
Apple ramps automation, shifts costs to suppliers beginning 2025
Apple is accelerating automation across its manufacturing base as it diversifies beyond China and braces for higher tariff-related expenses. Starting in 2025, suppliers will be required to fund their own robotics and automation upgrades to secure new orders, covering everything from iPhones to Macs. The push is designed to reduce labor dependency and improve consistency at facilities in India and Vietnam. Although near-term spending pressure is rising for partners, Apple expects automation to lower long-term costs and stabilize quality.
NAND prices climb again as demand tightens; SanDisk raises tags by 10%
NAND flash is in the middle of a price rebound. SanDisk implemented a 10% increase on September 4 across consumer and channel products—its second hike this year—amid strengthening demand from AI servers and enterprise storage. Spot pricing indicates the sharpest gains are in smaller-capacity parts such as 256GB and below. Meanwhile, lingering oversupply in 1TB devices is tempering price momentum at the high-capacity end. Analysts expect other major suppliers to follow as inventories normalize.
Samsung reignites Taylor, Texas fab buildout with Tesla chip win
Samsung Electronics is restarting investment at its long-delayed Taylor, Texas semiconductor site, allocating roughly US$2.9 billion for equipment after landing a significant order from Tesla. Engineering teams are scheduled to arrive in September and November to establish production lines. Mass production of Tesla’s AI chips is targeted for late 2026 or early 2027, with stabilization of 3nm and 2nm-class processes cited as key milestones. The renewed push could strengthen Samsung’s U.S. manufacturing footprint and its positioning in automotive AI.
China’s YMTC eyes DRAM with CXMT, targets future HBM growth
Yangtze Memory Technologies, China’s leading NAND maker, is moving into DRAM as part of a broader national effort to advance memory technology. The company plans to invest in DRAM equipment by late 2025 and is discussing a partnership with ChangXin Memory Technologies to address the high-bandwidth memory market. CXMT has already achieved mass production of HBM2, while YMTC brings expertise in hybrid bonding—an essential chip-stacking technique for next-generation HBM. The collaboration could help narrow the gap with South Korea’s incumbents faster than many expected.
Alibaba advances AI chip plans, denies massive Cambricon order
Alibaba is progressing on a next-generation AI processor that is reportedly under test and designed to interoperate with Nvidia hardware. Some reports suggested a potential move from overseas fabrication to a domestic foundry, though this has not been confirmed. Separately, speculation that Alibaba urgently ordered 150,000 MLU370 chips from Cambricon was denied by Alibaba Cloud, which stated no such purchase occurred.
The big picture
From Rapidus’ 2nm density breakthrough and TSMC’s anticipated pricing reset to renewed U.S. investments and China’s HBM ambitions, the week underscored how competitive and globally intertwined the semiconductor industry has become. AI, HPC, and advanced memory are at the center of every strategic decision—and the pace shows no sign of slowing.
