The surge in artificial intelligence (AI) applications has pushed cloud service providers to increase capital expenditures on data centers, prompting major memory manufacturers to shift production towards high-margin, high-performance products such as HBM and DDR5. This strategic realignment has led to a sharp reduction in the supply of niche memory types, especially those used in network communication devices, exacerbating price hikes and supply shortages that are expected to impact Taiwanese vendors through the first half of 2026.
Taiwan's semiconductor materials sector delivered a mixed performance in December 2025, highlighting a widening gap between advanced and mature segments of the supply chain. Upstream silicon wafer suppliers showed signs of stabilization but not a full rebound, while regenerated wafers and advanced-process-related consumables outperformed. By contrast, traditional process materials such as photomasks remained under pressure, while packaging materials showed clear divergence in response to AI-driven demand.
TSMC used its earnings call for the fourth quarter of 2025 to deliver one of its most bullish outlooks in years, arguing that artificial intelligence (AI) is reshaping the semiconductor industry's growth trajectory rather than inflating a short-lived bubble.
SK Hynix plans to bring forward the opening of a new semiconductor factory by three months and will also begin operating another new plant in February 2026, as surging memory demand strains global supply, a senior executive told Reuters.
National Institutes of Applied Research (NIAR) chairman Cheng-wen Wu has called on the Electronic and Optoelectronic System Research Laboratories (EOSL) of the Industrial Technology Research Institute (ITRI) to take up the mantle and connect with the wafer-level and chip-level platforms at the NIAR's Taiwan Semiconductor Research Institute (TSRI).
Taiwan's government is stepping up cross-ministerial support to build a domestic silicon photonics ecosystem, with a focus on strengthening local capabilities in co-packaged optics materials and supporting industrial deployment under its long-term chip innovation strategy.
The US government confirmed this week the conditions for exporting the H200 AI chip to China, reportedly including third-party lab testing before shipment to ensure compliance with AI technology standards. Additionally, the number of AI chips sold to Chinese customers cannot exceed 50% of those sold to US customers. Chinese buyers must also prove they have implemented sufficient security measures and that the chips will not be used for military purposes.
Chinese artificial intelligence startup Z.ai, formerly known as Zhipu AI, has unveiled GLM-Image, a multimodal AI model trained entirely on Huawei's Ascend chip platform, marking a notable milestone in China's push to build a self-reliant AI computing stack. The company said GLM-Image was developed using Huawei's Ascend 800T A2 servers and the MindSpore AI framework, making it the first publicly disclosed multimodal model trained on this domestic hardware platform.
Taiwan Semiconductor Manufacturing Company (TSMC) reported consolidated revenue of approximately NT$1.4609 trillion (US$46.21 billion) and net income after tax of about NT$505.74 billion for the fourth quarter of 2025. For the full year, the company achieved revenue of NT$3.909 trillion and net profit after tax of NT$1.717 trillion, exceeding market expectations.
As the global memory market enters a prolonged upcycle, South Korea's two leading memory chipmakers, Samsung Electronics and SK Hynix, are expected to post record operating profits in 2026. Combined operating profit could exceed KRW300 trillion (approx. US$210 billion), roughly double the level recorded in 2025.
As Micro LED moves into non-display application opportunities, PlayNitride Technologies announced a partnership with Brillink Technologies, a leader in two-dimensional array optical coupling technologies, to jointly develop a next-generation optical interconnect platform capable of meeting the needs of AI and high-performance computing (HPC). The collaboration integrates three core segments: high-efficiency Micro LED emitter arrays, ultra-high-sensitivity GeSi avalanche photodiode (APD) arrays, and two-dimensional array vertical couplers (2D-AVC). Compared with traditional active electrical cables (AEC), the platform is expected to achieve longer transmission distances, lower power consumption, and higher bandwidth density.
More coverage
