AI崛起對台灣AI伺服器產業帶來的商機與投資機會

Abstract

隨著計算能力的進步、大數據的普及，透過機器學習，2017年後LLM(GPT-2、GPT-3)出現為AI多樣化應用奠定基礎，2022年12月ChatGPT開啟生成式AI的時代，生成式AI能自動、快速地創建和生成新的內容，如圖像、文字、程式、影片、聲音和3D模型等，AI技術發展，將從目前生成式AI，現階段已進入能與人互動並執行任務的「代理AI」，接下來則是為機器人和真實應用的「實體AI」，AI將成為電力和網路後的第三代基礎建設。Marketsand Markets預測AI市場價值將從2024年的2,146億美元，成長至2030年的13,391億美元，複合年成長率（CAGR）為35.7%；Gartner預計2022~2027年AI軟體支出複合年成長率將成長19.1%。預估至2030年AI資料中心建置將達1兆美元，企業將加速採購GPU及ASIC。 生成式AI帶來各行各業的典範移轉，造成市場對於AI的需求快速暴增，各大CSP及AI業者為搶奪未來AI市場的龐大商機，提升算力成為首要任務，因而展開AI SERVER的軍備競賽，台灣不僅在PC及SERVER OEM/ODM擁有豐富的組裝經驗及完整的零組件供應鏈，更可以提供獨一無二的先進製程幫GPU及ASIC代工，因此在產業供應鏈完整且具垂直整合優勢的利基下，在AI產業崛起的關鍵時刻，率先獲得最大的商機。 從上游晶片、中游組裝到下游應用，台廠在這波AI浪潮中扮演著不可或缺的角色。無論是晶圓代工，散熱、供電、PCB等關鍵零組件，還是伺服器代工龍頭，都讓台灣成為全球AI供應鏈的重要陣地，在AI產業百花齊放的年代，只要稍微有點關聯性，過去2年股價表現表現都是遠遠超過大盤，可說是滿地開花、雞犬升天。惟在經過過去2年的激情後，絢爛將歸於平淡，市場將以更嚴格的標準來檢視AI SERVER供應鏈的續航力，投資標的的選擇將更具挑戰性，因此未來誰能夠憑藉著營運績效、競爭力與獲利能力，在AI Server各個領域的供應鏈中脫穎而出，將是未來布局AI投資機會的致勝關鍵。 AI涉及的供應鏈相當廣泛，這些公司在AI技術研發和應用上都扮演關鍵角色。整個AI產業包括製造伺服器的晶片及主要IC的上游，製造關鍵零組件的中游及伺服器代工廠的下游。由於台灣AI SERVER供應鏈的子產業眾多，因此在上、中、下游分別挑選具有群聚效應(即產業中具代表性的廠商不只一家)的AI SERVER供應鏈子產業，做為比較的基礎。以上述標準選擇出台灣AI SERVER供應鏈中較具代表性的產業，其中在上游選擇的是矽智財(IP)產業(世芯、創意、智原)，中游選擇的是散熱產業(奇鋐、雙鴻、力致)及銅箔基板(CCL)產業(台光電、台燿、聯茂)，下游則是選擇AI SERVER組裝產業(緯穎、廣達、鴻海、緯創)，具有代表性的公司共13家。 透過上述13家公司的經營績效、營運前景、財務數字的預估及公司訪談，了解各家公司基本面。此外，AI SERVER規格迭代快速，因此產品規格變化所帶來的商機將影響AI供應鏈未來的營運前景。另外透過過往的營運績效及財務數字分析，篩選出其中的佼佼者。透過基本面分析、產品規格變化所帶來的商機、營運績效分析及財務數字分析這四項標準，挑選出各個子產業中最具競爭力及潛力的公司。作為主要的投資建議標的。 透過基本面分析、未來產品規格的變化、營運績效及獲利能力的分析，分別由AI SERVER產業的上、中、下游的供應鏈篩選出世芯、奇鋐、台光電及緯穎，作為各個子產業的首選投資標的。最後再透過NPV淨現值法，將各別公司未來現金流折現，計算出各別公司合理價值，再與目前股價比較，計算出潛在報酬率，以供未來投資參考。

With the advancement of computing power and the proliferation of big data, machine learning has paved the way for diverse AI applications since 2017, especially with the emergence of large language models (LLMs) like GPT-2 and GPT-3. In December 2022, ChatGPT marked the beginning of the generative AI era. Generative AI can autonomously and rapidly create new content such as images, text, code, videos, audio, and 3D models. As AI technology evolves, it is moving beyond generative AI into the current phase of "agent AI"—capable of interacting with humans and executing tasks—and eventually towards "embodied AI" for robotics and real-world applications. AI is expected to become the third foundational infrastructure following electricity and the internet. MarketsandMarkets forecasts that the value of the AI market will grow from USD 214.6 billion in 2024 to USD 1.3391 trillion by 2030, with a compound annual growth rate (CAGR) of 35.7%. Gartner estimates that AI software spending will grow at a CAGR of 19.1% from 2022 to 2027. By 2030, AI data center infrastructure investment is projected to reach USD 1 trillion, driving enterprises to accelerate procurement of GPUs and ASICs. Generative AI is catalyzing a paradigm shift across industries, rapidly driving up demand for AI. Major cloud service providers (CSPs) and AI companies are racing to expand computational capacity, sparking an AI server arms race. Taiwan, with its rich experience in PC and server OEM/ODM and its complete component supply chain, also offers advanced semiconductor manufacturing for GPU and ASIC foundry services. Thanks to this vertically integrated and comprehensive industrial ecosystem, Taiwan is seizing early opportunities in the AI revolution. From upstream chips, midstream assembly, to downstream applications, Taiwanese companies are playing an indispensable role in the AI wave. Whether it's wafer foundry, critical components like thermal modules, power supply, PCBs, or leading server assemblers, Taiwan has become a vital hub in the global AI supply chain. Over the past two years, even companies with peripheral relevance to AI have seen their stock prices outperform the broader market—truly a "rising tide lifts all boats" scenario. However, after the past two years of excitement, the market is entering a more sober phase, applying stricter standards to assess the sustainability of the AI server supply chain. Investment selection is becoming more challenging. In the AI server ecosystem, the supply chain is vast, with companies playing critical roles in technology development and application. The entire industry spans upstream (chip and key IC manufacturing), midstream (key component manufacturing), and downstream (server assembly). Given the diversity of Taiwan’s AI server-related industries, this analysis selects representative sub-industries with clustering effects (i.e., more than one notable company per sector) in each of the upstream, midstream, and downstream segments. Using this standard, the selected representative sectors of Taiwan’s AI server supply chain are: • Upstream: IP (Intellectual Property) industry — including Alchip (世芯), Global Unichip (創意), and Faraday (智原) • Midstream: Thermal solutions — Chicony Power (奇鋐), Auras (雙鴻), Taisol (力致); and Copper Clad Laminate (CCL) — Taiwan Union Technology (台光電), Elite Material (台燿), Iteq (聯茂) • Downstream: AI server assembly — WiWynn (緯穎), Quanta (廣達), Foxconn (鴻海), and Wistron (緯創) A total of 13 key companies are selected. Through company interviews and analysis of business performance, future prospects, and financial forecasts, this study aims to assess the fundamentals of each company. Due to rapid iterations in AI server specifications, the business opportunities driven by spec evolution will directly affect the outlook for the supply chain. By evaluating past performance and financial data, the top performers in each segment can be identified. Using four criteria—fundamental analysis, business opportunities from product spec changes, operational performance, and financial metrics—the most competitive and promising company in each sub-industry is selected as a primary investment candidate. The selected top companies in each supply chain segment are: • Upstream: Alchip (世芯) • Midstream: Chicony Power (奇鋐) and Taiwan Union Technology (台光電) • Downstream: WiWynn (緯穎) Finally, by applying the Net Present Value (NPV) method to discount each company’s projected future cash flows, their fair values are calculated. Comparing these fair values with current stock prices yields their potential returns, serving as a reference for future investment decisions.