SK Hynix to Double Production Capacity Over the Next 5 Years to Ease AI Memory Shortage

SK Hynix Chairman Chey Tae-won told reporters in Taipei that the company plans to double its memory chip wafer production capacity over the next five years. The global shortage of memory chips, which are essential to the AI industry, could persist until 2030, and SK Hynix intends to accelerate investment to close the supply-demand gap. While the investment amount was not disclosed, the expansion is expected to be a large-scale one aimed at meeting surging demand centered on HBM (high-bandwidth memory).

LLM Agents Write Security Patches but Leave Vulnerabilities Intact — CVE-Bench Study

A study called "CVE-Bench," which tested security patch generation with five state-of-the-art LLM models against 20 real-world CVEs across 18 Python projects, is drawing attention. Across a total of 300 runs under three prompt conditions, agents were found to pass hidden tests based on the maintainers' own fixes while still leaving the underlying vulnerability in place. The study applied a strict evaluation criterion: a vulnerability that is 90% patched is still a vulnerability. The results highlight key limitations of relying on LLMs for security fixes.

EU Set to Join US-Led Chip Alliance "Pax Silica"

Reports indicate that the EU is moving toward joining "Pax Silica," a US-led chip alliance formed to counter China's AI competition. The alliance, which aims to restructure the semiconductor supply chain, shows the geopolitical side of the AI race becoming increasingly explicit. Europe's participation suggests that the fragmentation of the global semiconductor ecosystem is likely to deepen.

MediaTek Announces Major AI Hiring Push, Also Behind NVIDIA's Spark PC Chip

Taiwanese chipmaker MediaTek has announced a large-scale hiring drive as it moves into the AI space in earnest. The company, which is also contributing to the adoption of NVIDIA's new PC chip "Spark," is optimistic about order visibility in its data center business for the coming years. Amid growing concerns about job losses in the AI era, the chip industry is in fact seeing intensifying competition for talent.

Qwen 3.6-35B Achieves Fast Inference on Intel Arc B70 Pro

Community reports show that the Qwen 3.6-35B-A3B model achieved fast inference on the Intel Arc B70 Pro, with prompt processing at 977 tk/s and a 262k context window. The Intel Arc series is demonstrating practical performance as a non-NVIDIA GPU option, expanding the choices available for local AI environments.

Generative AI Answers Deemed "More Trustworthy Than Word of Mouth" — Hakuhodo Survey

According to a survey by Hakuhodo, the use of generative AI for shopping has reached 24.6% of consumers (about 1 in 4). Particularly noteworthy is the emergence of consumers who rate generative AI answers as more trustworthy than word of mouth. The data suggests that AI's role in information-seeking behavior is shifting from a supplementary tool to a primary information source.

---