跳到主要內容

台積電加碼美國1000億美元:AI晶片需求推升Q2獲利飆77% | TSMC Adds $100B to Arizona: AI Demand Drives 77% Profit

By Kit 小克 | AI Tool Observer | 2026-07-24

🇹🇼 台積電加碼美國1000億美元:AI晶片需求推升Q2獲利飆77%

台積電再砸1000億美元:美國總投資衝上2650億美元

台積電(TSMC)7月16日宣布,將再對美國亞利桑那州追加投資1000億美元,把當地總投資額推升到2650億美元。這是繼2024年、2025年多輪加碼後最大手筆的一次,也把亞利桑那廠區的規模擴大到10座晶圓廠、2座先進封裝廠與1座研發中心,未來將新增至少4座採用2奈米以下製程的產線。這波AI晶片需求熱潮,正在重寫全球半導體版圖。

Q2財報:獲利飆77%,全靠AI基礎建設

台積電同時公布第二季財報:淨利年增77.4%,達新台幣7065.6億元(約219.8億美元),營收年增36%,來到新台幣1兆2703.8億元。管理層直言,成長「幾乎全部來自AI基礎建設客戶」——也就是輝達(Nvidia)、超微(AMD)等公司為訓練與推論晶片下的巨額訂單。台積電也把2026年資本支出預算從原本的520億至560億美元,一口氣調高到600億至640億美元,並預期全年營收成長將超過40%。

為什麼是亞利桑那,而不是台灣加碼?

這次擴產選在美國本土,反映的是地緣政治現實:美國《晶片法案》補貼、客戶(尤其是輝達與蘋果)要求供應鏈分散、加上美中科技戰持續延燒,都讓台積電必須在美國建立完整的先進製程能力,而不只是把舊製程搬過去。值得注意的是,亞利桑那擴廠一旦全數完工,用水量預估將讓台積電成為鳳凰城當地最大的工業用水戶,當地政府與居民對水資源議題已開始關注。

對台灣與供應鏈的意義

對台灣而言,最先進的製程(2奈米以下)與研發核心目前仍留在新竹、高雄,美國廠主要承接量產與部分先進封裝,短期內不必過度擔心「去台化」。但長期來看,AI晶片需求是否能持續支撐台積電同時在台灣、美國、日本、德國多地擴產,是接下來幾季財報要觀察的重點——一旦AI資本支出降溫,這些巨額投資的回收壓力將直接反映在毛利率上。

重點整理

  • 台積電追加1000億美元投資亞利桑那,美國總投資達2650億美元
  • 亞利桑那廠區規劃10座晶圓廠,新增4座2奈米以下製程產線
  • Q2淨利年增77.4%,營收年增36%,成長主要來自AI基礎建設訂單
  • 2026年資本支出上修至600億至640億美元

好不好用,試了才知道。


🇺🇸 TSMC Adds $100B to Arizona: AI Demand Drives 77% Profit

TSMC Bets Another $100B on Arizona as AI Chip Demand Hits Records

Taiwan Semiconductor Manufacturing Co. (TSMC) announced on July 16 that it will invest an additional $100 billion in Arizona, pushing its total U.S. commitment to $265 billion — the largest single addition yet after several rounds of expansion since 2024. The Arizona campus will now include 10 fabs, two advanced-packaging facilities, and one R&D center, with at least four new lines dedicated to 2-nanometer-and-below process technology. The AI chip boom is reshaping the global semiconductor map in real time.

Q2 Earnings: 77% Profit Jump, Almost Entirely From AI

TSMC also reported blowout Q2 results: net profit rose 77.4% year-over-year to NT$706.56 billion (about $21.98 billion), on revenue up 36% to NT$1.27 trillion. Management said growth came "almost entirely" from AI infrastructure customers — namely Nvidia and AMD, placing massive orders for training and inference chips. TSMC raised its 2026 capital expenditure guidance from $52-56 billion to $60-64 billion, and now expects full-year revenue growth above 40%.

Why Arizona, Not More Capacity in Taiwan?

The U.S.-based expansion reflects geopolitical reality: CHIPS Act incentives, customer pressure (especially from Nvidia and Apple) to diversify the supply chain, and the ongoing U.S.-China tech rivalry all push TSMC to build genuine leading-edge capacity on American soil, not just legacy nodes. One overlooked side effect: once fully built, the Arizona site's water demand is projected to make TSMC the largest industrial water user in Phoenix, a detail local officials and residents are starting to scrutinize.

What It Means for Taiwan and the Supply Chain

For Taiwan, the most advanced nodes (below 2nm) and core R&D stay in Hsinchu and Kaohsiung for now — the U.S. fabs mainly handle mass production and some advanced packaging, so fears of "hollowing out" Taiwan's chip industry remain overstated in the near term. The bigger question for the next few quarters: can AI chip demand keep justifying simultaneous mega-expansions in Taiwan, the U.S., Japan, and Germany? If AI capex cools, the payback pressure on these investments will show up directly in TSMC's gross margins.

Key Takeaways

  • TSMC adds $100B to Arizona, bringing total U.S. investment to $265B
  • Arizona campus now plans 10 fabs, including four new 2nm-and-below lines
  • Q2 net profit up 77.4% YoY, revenue up 36%, driven by AI infrastructure orders
  • 2026 capex guidance raised to $60-64 billion

好不好用,試了才知道。

Sources / 資料來源

延伸閱讀 / Related Articles


AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

留言

這個網誌中的熱門文章

Cursor vs GitHub Copilot vs Claude Code:AI 程式助手大比拼 | AI Coding Assistants Compared: Cursor vs GitHub Copilot vs Claude Code

Stanford 研究登上《Science》:11 個 AI 模型有 47% 機率說你對,即使你錯了 | Stanford Study in Science: AI Models Validate Harmful Behavior 47% of the Time — Sycophancy Is a Real Problem

Claude Code 實測:AI 幫你寫程式到底行不行? | Claude Code Review: Can AI Really Code for You?