あさって　電子立国日本の半導体

NAN GreenMet has announced it will establish "NAN MagneTech," India's first integrated neodymium (NdFeB) magnet manufacturing facility, in Andhra Pradesh, investing ₹12.5 billion (about $130 million) in its first phase. The project is reportedly backed by Japanese technology partners, including Masato Sagawa, one of the inventors of NdFeB magnets, and is positioned as part of India's response to China's dominance of the rare earth magnet supply chain.

At Advancing AI 2026, AMD unveiled "AMD Kria AI Solutions," a new robotics product line. Built around FPGA technology inherited from its 2022 Xilinx acquisition, it claims to enable low-latency control such as 23-degree-of-freedom robotic hand control. Foundation Future Industries' humanoid robots already use AMD's chips.

NVIDIA's flagship GeForce RTX 5090 has kept climbing in price, reaching at least €4,250 in Europe and $4,929.99 at NVIDIA's own online store — roughly double its $1,999 launch MSRP from January 2025. The driver is a GDDR7 shortage, as Samsung, SK Hynix, and Micron prioritize wafer capacity for AI-datacenter HBM, leaving VRAM now accounting for more than 80% of a high-end GPU's bill of materials.

AMD has told board partners it will raise wholesale prices for Radeon RX 9000-series GPU kits (die plus VRAM) by at least 10% starting August 2026, driven by GDDR6 memory prices roughly tripling since late 2025. In China, the RX 9070 XT's official channel guidance price has already crossed $1,000 for the first time, 40% above its March 2025 launch price, while ASUS has raised US listings by as much as 17.5%.

Hitachi announced on July 23, 2026 that it has been selected for a NEDO public program and will begin joint R&D with Intel's Japan subsidiary and the National Institute of Advanced Industrial Science and Technology (AIST) on semiconductor-based "silicon quantum computers." The goal is a 100-qubit device by fiscal 2028 and a 1,000-qubit prototype with quantum error correction implemented by 2030.

Delta Electronics plans to begin low-volume production of Nvidia's next-generation "800V HVDC" power architecture for AI factories in late Q1 2027, ramping up in stages from Q2 to match customer deployment schedules. SemiAnalysis has flagged the possibility that large-scale adoption could slip to 2028 or later, though supply chain sources say this reflects first-generation design validation delays rather than a change to the technology roadmap itself.

TSMC is advancing "CoPoS" (Chip-on-Panel-on-Substrate), a next-generation advanced packaging technology using 310x310mm panels, and plans to begin mass production in the second half of 2028. However, multiple media reports confirm that full commercialization of glass-core substrates is unlikely before 2030. The biggest technical barrier is substrate warpage, and TSMC Chairman C.C. Wei says there are "no shortcuts."

India's government has approved a new scheme called "Semicon 2.0" to strengthen its semiconductor ecosystem. This article summarizes the six-pillar plan, from chip design to workforce development, and the progress already made under Semicon 1.0, based on India's official PIB announcement.

AMD is combining Instinct MI455X, EPYC Venice and Pensando networking in Helios, a rack-scale AI platform built around UALink over Ethernet.

ASML has outlined plans to raise Low-NA EUV lithography system capacity by roughly 30% in both 2027 and 2028, driven by surging AI chip demand. This article summarizes the confirmed Q2 2026 earnings figures and capacity plans based on ASML's own statements and multiple independent reports.