Semiconductor investment, AI chips, advanced packaging and data center infrastructure dominated the week, with South Korea facing US pressure, India scaling chip manufacturing, and Apple, Qualcomm, MediaTek and Taiwan suppliers advancing new AI and 2nm strategies. Below are the most-read DIGITIMES stories from the week of August 24-30, 2026.
US voices first complaint over South Korea's KRW800T chip cluster as investment pressure mounts
The US has informally expressed dissatisfaction with South Korea's KRW800 trillion (≈ US$581 billion) semiconductor cluster plan, according to sources cited by Chosun Ilbo, as Washington intensifies demands for Korean chipmakers to invest and build memory fabs in the US. The complaint is not against the project itself, but against Seoul's slow progress on concrete US investment plans. The issue surfaced during Minister Kim Jung-kwan's August 16 visit, deepening worries among chipmakers about a double burden.
Ablecom takes control of Jochu to expand AI server rack, liquid cooling business
Ablecom Technology said it joined a private placement for Jochu Technology and became the listed company's largest shareholder, giving the Taiwan-based chassis and thermal solutions provider added capacity for AI server racks and liquid cooling products. The investment is meant to help meet rising global data center demand for high-density computing. Jochu, which already makes server racks and is expanding in Vietnam, will work more closely with Ablecom on production and technology development.
India's CG Power targets 70% utilization at larger semiconductor packaging plant
CG Power and Industrial Solutions is targeting 70% utilization at its larger semiconductor packaging plant within about four years, with potential annual revenue of around US$500 million once the facility scales, according to CNBC. The estimate depends on product mix, demand and pricing. The company's CG Semi joint venture with Renesas and Stars Microelectronics is investing more than INR76 billion (≈ US$796 million) over five years, with G1 already in commercial production and G2 moving ahead of schedule.
MediaTek wins Google TPU deal on three key strengths
DIGITIMES analyst Jim Chien said Taiwan's IC design industry is set to grow 10% in 2026, driven by AI and AI ASIC projects from MediaTek, GUC, Alchip, Phison and Silicon Motion. He said MediaTek's first-half 2026 revenue should be roughly flat because AI revenue starts contributing later, but the company won Google's TPU project through CPU, GPU and NPU integration, high-speed I/O and SerDes expertise, and advanced process and back-end design experience.
Qualcomm's HBC play pulls Samsung and SK Hynix into a new race beyond HBM
Qualcomm's High Bandwidth Compute architecture is drawing Samsung Electronics and SK Hynix into a new memory race beyond HBM, as the two Korean suppliers continue their HBM4 and HBM4E roadmaps while exploring Qualcomm's approach. The architecture moves compute beneath the memory stack using TSVs, aiming to reduce data movement and improve bandwidth and efficiency. Qualcomm says AI250 could deliver 133TB/s effective bandwidth per card, with AI300 targeted for 2028, though commercial proof remains untested.
India chip manufacturing push exposes production-test skills gap, NI says
NI executives say India's semiconductor expansion is exposing a shortage of engineers with production-testing experience, even as the country builds strong design and validation capabilities. Baskar Ceri said demand for production-test systems should rise over the next three to five years as OSAT projects scale and more manufacturing comes online. NI is already working with Kaynes Technology and says this is the third attempt at semiconductor manufacturing in India, with real traction this time, but talent gaps remain significant.
Apple's M6 marks a transitional step before wider 2nm adoption
Apple introduced the M6 and M5 Ultra chips without a launch event, with the M6 becoming its first processor built on TSMC's 2nm process and debuting in the new Mac mini. Industry watchers say the chip is mainly a transitional step that validates 2nm design rather than a major near-term revenue driver. The M6 is limited to a base version, while higher-end models are expected to wait for the M7 in the first half of 2027.
Article edited by Levi Li