CONNECT WITH US
Jul 9
What China's rumored limited reopening to Nvidia's H200 implies for US-China chip contest and Beijing's drive for self-reliance
China's rumored tentative plan to allow a handful of its largest artificial intelligence (AI) companies to purchase a small number of Nvidia H200 chips has implications that extend well beyond a single procurement decision. The plan is possibly, though not only, for the shifting balance between US export leverage and Beijing's push to reduce its dependence on foreign silicon.

Samsung Electronics, SK Hynix and Micron Technology are taking different manufacturing paths for the logic base die in HBM4, opening a new front in the competition over advanced memory performance, yield and supply.

Micron Technology's decision to lift its planned US investment to more than US$ 250 billion through 2035, unveiled July 9 alongside the first concrete pour at its Clay, New York, site, is fundamentally a bet on where the AI era's most contested commodity, memory, will be manufactured.

GigaDevice Semiconductor expects first-half 2026 net profit to reach about CNY6.9 billion (US$960 million), a 1,099% increase from a year earlier, after tight memory chip supply lifted both shipment volumes and prices.

GlobalWafers and Micron have signed a 10-year supply agreement, alongside US$500 million in strategic financial support, in a move that underscores how artificial intelligence (AI), high-performance computing (HPC), data centers, and advanced memory demand are reshaping semiconductor sourcing. The deal extends US supply chain localization beyond wafer production and into upstream critical materials.

Chunghwa Precision Test Tech president Scott Huang said on July 7 that the company's biggest challenge over the next five years will be capacity as artificial intelligence demand surges across the semiconductor supply chain. He also urged Taiwan to increase investment in quantum technology and nuclear fusion, saying both fields could shape the island's long-term industrial future.

Meta Platforms plans to begin manufacturing its Iris AI accelerator in September 2026 while securing long-term supplies of memory, storage and optical equipment for a computing expansion expected to reach 14 gigawatts in 2027.
CSP ASIC demand drives structural growth in high-speed interconnects
Jul 10, 10:19
As cloud service providers (CSPs) continue to ramp up capital expenditures, demand for high-speed interconnects within data centers is accelerating. Multiple research firms forecast that leading CSPs will sustain high double-digit capex growth in 2026, with roughly half of the increase driven by data center expansion.

High-bandwidth memory (HBM) prices could more than double in 2027 as Nvidia's Rubin platform drives demand, HBM4 raises production costs, and long-term agreements lock up an increasing share of global DRAM capacity, according to memory industry sources.

China's graduate job market is shifting toward semiconductors, materials, and manufacturing, with global implications for supply chains, AI development, and critical minerals. Fresh salary data show computer science and software engineering losing ground as strategic industries draw more talent, while hard tech majors gain pay advantages across the country.
Global Mixed-mode Technology reported June revenue of NT$743 million (US$23.16 million), down 0.71% from May but up 7.19% from a year earlier. The Taiwanese PMIC maker said second-quarter revenue reached NT$2.238 billion, up 5.99% from the first quarter and down 0.46% year on year, while first-half revenue totaled NT$4.349 billion, a 1.30% decline from a year earlier.

Mitsubishi Chemical and Japan Steel Works are preparing a fresh capacity expansion in gallium nitride (GaN) substrates, aiming to capture growing demand from next-generation power semiconductors used in EVs, inverters and data center power systems.