Taiwanese supply chains seize opportunities in silicon photonics and CPO.

2025/11/21

United Daily News
November 19, 2025
By Economic Daily News, Liao Hsien-Lung

The AI industry atmosphere is gradually shifting from "frenzied anticipation" to a "profit verification" phase. NVIDIA's financial report has become a key indicator for evaluating the fundamentals of the AI industry. Nomura Investment Trust stated that if NVIDIA's financial report exceeds expectations, it will help alleviate market concerns about an AI bubble. As the global AI industry enters a profit verification period, Taiwan's supply chain has demonstrated resilience and forward-looking planning. Although the market experienced short-term volatility due to interest rate and data releases, the long-term trend of AI remains clear. AI-related applications and interconnect technologies continue to drive industry upgrades. Investors are advised to focus on major suppliers in Taiwan's semiconductor and optical communication supply chains, and to consider buying into actively managed market capitalization ETFs on dips to seize the next wave of AI growth opportunities.

Chang Chi-wen, Vice President of Investment Strategy at Nomura Securities Investment Trust, analyzed the discussions between AMD and NVIDIA, noting their continued optimism regarding AI growth over the next five years. This optimism stems from the increasingly broad applications of AI, ranging from enterprise-level databases and software services to personal use in diverse applications such as text generation and video/image processing. The growing number of AI users and the expanding scenarios driving demand for AI computing power are fueling the continued expansion of global AI data centers. This will create new growth momentum for GPUs, cooling systems, power supplies, and even memory, a previously cyclical industry.

Chang further analyzed that this year's price surge in memory (DRAM, SSD) is not only due to the exponential growth in AI-generated data volume, but also, crucially, because AI is creating a completely different "new supply and demand pattern" in the memory industry. This wave of demand is no longer driven by traditional brand server manufacturers (such as Dell and HPE), but rather by the three major cloud service providers (Amazon, Microsoft, and Google). This trend is expected to be more stable and sustainable than in the past. Furthermore, in recent years, major South Korean manufacturers have shifted their production capacity to HBM (Hardware as a Whole Machine), and Taiwanese manufacturers have managed their production capacity systematically over the past few years, resulting in the current oligopoly of five major Taiwanese and South Korean manufacturers, all of whom possess significant bargaining power. As long as AI continues to develop, the memory industry is expected to experience a more sustained upward trend.

Lin Hao-hsiang, fund manager of Nomura Taiwan Enhanced 50 Active ETF (00985), stated that global AI development has shifted from a "computing power race" to an "interconnectivity race." With the emergence of bottlenecks in telecommunications transmission, optical interconnects have become the new "transmission nerves" of the AI era. Especially in the inference market, the demand for high bandwidth, low latency, and extreme performance far exceeds that of the training phase, making silicon photonics and CPO (Co-Packaged Optics) the core of hardware innovation. Currently, silicon photonics technology has entered mass production and continues to upgrade, becoming a high-speed engine; while CPO (Content-Based Optical Processing) brings the optical engine directly close to the GPU, sparking a revolution in ultimate packaging. NVIDIA's latest Rubin and Rubin Ultra adopt a decoupled inference architecture, paired with NVLink 6.0 with up to 3.6TB/s bandwidth. The AI inference market is projected to exceed $120 billion by 2026, more than five times the size of the training market. Silicon photonics and CPO technologies are entering a critical phase:

2024–2025: Mass production of 800G to 1.6T silicon photonics modules.

2026–2027: Explosive growth of CPO and 3.2T optical engines.

After 2028: Optical interconnect technologies above 6.4T become mainstream.

Chang Chi-wen mentioned that Taiwan's supply chain is actively deploying resources, with comprehensive investment from wafer foundry and packaging/testing to optoelectronic components. TSMC (2330), ASE (3711), MediaTek (2454), Quanta (2382), and other leading companies have established the "Silicon Photonics Industry Alliance" (SiPhIA) to promote optical interconnect and CPO (Co-Packaged Optics) technologies and seize the high-speed data center transmission market. Meanwhile, companies such as Advanced Semiconductor (3081), New Energy (2455), Accton Technology (2345), and CSOT (4979) are accelerating the mass production of 800G to 1.6T silicon photonics modules. It is expected that 2026 will see a surge in CPO and 3.2T optical engines, and after 2028, 6.4T technology will become mainstream. The Taiwanese supply chain is expected to play a core role in the global AI inference market. The target is for the silicon photonics market to reach US$64 million by 2028, with CPO expanding to US$500 million. Companies like Advanced Semiconductor and Newtech have already mass-produced 800G to 1.6T silicon photonics modules and are accelerating the development of InP laser chips to capture the global demand for optical communication driven by AI inference. AI inference is pushing the demand for high-bandwidth, ultra-low-latency interconnects between AI servers, making optical interconnects the next-generation data center's transmission nerve. Taiwan's supply chain possesses a complete link from wafer to packaging, and from laser to module, entering a new era of interconnect competition. Based on its silicon photonics mass production capabilities, and with a focus on CPO technology packaging, a boom period is expected between 2026 and 2028, potentially ushering in an AI inference market opportunity exceeding US$100 billion.

Zhang Jiwen believes that although there are many noises in the short-term market, they are not turning points that will change the long-term trend of AI. Currently, the market has corrected from 28,334 points in early November to 27,397 points on November 14th, a drop of 3.3%, which is still relatively limited overall. Furthermore, from an industry perspective, semiconductors and AI experienced a brief pullback due to a shift in mainstream focus, but traditional industries such as cement, food, textiles, and finance showed signs of catching up. This is key to the market's ability to consolidate and rise again, moving towards a more mature stage. Investors should avoid chasing highs in the short term and take advantage of the recent volatility to review their holdings, weeding out weaker performers and focusing on industries with strong growth potential in 2026, such as advanced semiconductor processes, heat dissipation, and power management.

【Disclaimer】
The content of this article only represents the author's personal views and has nothing to do with Creating.
The content, text description and originality have not been verified by this site. This site does not make any guarantee or commitment to this article and all or part of the content, authenticity, completeness, and timeliness. It is for readers' reference only. Please verify the relevant content yourself.

Creating Nano Technologies, Inc.
59 Alley 21 Lane 279, Chung Cheng Road, Yung Kang City, Tainan, TAIWAN
TEL:886-6-2323927 FAX:886-6-2013306 URL: http://www.creating-nanotech.com

BACK TO LIST