亞馬遜核電協議評測：20年合約解AI缺電危機？ | Amazon Nuclear Deal Review: Can It Fix AI's Power Crisis?

亞馬遜核電協議評測：20年合約解AI缺電危機？ | Amazon Nuclear Deal Review: Can It Fix AI's Power Crisis?

By Kit 小克 | AI Tool Observer | 2026-10-02

🇹🇼 亞馬遜核電協議評測：20年合約解AI缺電危機？

亞馬遜在2026年10月初與電力公司Constellation Energy簽下一紙核電協議，20年合約總值約37億美元，鎖定馬里蘭州Calvert Cliffs核電廠690MW電力，目的是餵飽AI資料中心越滾越大的用電缺口。媒體把這起核電協議捧成「AI缺電救星」，但攤開合約細節，故事沒那麼簡單。

亞馬遜的核電協議到底買了什麼？

這筆交易買的是既有電廠的既有產能，不是蓋一座新核電廠。

合約期間：20年，約37億美元
電力來源：Calvert Cliffs核電廠兩座反應爐
取得電量：690MW既有產能，另資助190MW新增機組
轉折：亞馬遜上月才因居民反對取消廠區旁蓋資料中心的計畫，這次改成「只買電、不蓋廠」
附加效益：長約讓Constellation有財務底氣，替兩座反應爐申請延役到2034、2036年後

為什麼AI公司都在搶核電？

因為AI資料中心用電成長速度，已經超過電網蓋新電廠的速度。美國資料中心一年用電量已達約176TWh，佔全美用電量4.4%；全球資料中心用電量預估從2025年的485TWh，2030年翻倍逼近950TWh。AI缺電不是危言聳聽，核電因發電穩定、不受天氣影響，自然成了科技巨頭首選。

核電協議解得了AI缺電危機嗎？

批評者的答案保守許多：就算用最樂觀的小型模組化反應爐（SMR）估算，新增核能一年也只能生出約15TWh，對比2025年美國資料中心需求推估的312.6TWh，根本杯水車薪。更現實的是成本——新核電廠昂貴又耗時，費用常轉嫁到一般用戶電費上。Gallup調查顯示，70%美國人反對自家社區蓋新AI資料中心，理由正是電費上漲、用水暴增與噪音。

這跟一般AI用戶有什麼關係？

短期內不會改變你用ChatGPT或Claude的體驗，但長期可能反映在電費帳單上。這類核電協議多半是財務操作——幫既有電廠續命、鎖定長期供電，而非短時間生出新電力。「AI用核能供電」的行銷故事聽起來乾淨，實際離大規模商轉還有好幾年。這類基礎建設好不好用，得等真正蓋出新產能才算數，好不好用，試了才知道。

🇺🇸 Amazon Nuclear Deal Review: Can It Fix AI's Power Crisis?

Amazon signed a 20-year nuclear power deal with Constellation Energy in early October 2026, a $3 billion contract locking in 690MW from the Calvert Cliffs nuclear plant in Maryland, plus funding for 190MW of new capacity. Headlines framed it as the answer to AI's power crunch — but the fine print tells a more modest story.

What Did Amazon Actually Buy?

Amazon bought capacity from an existing plant, not a brand-new reactor.

Contract length: 20 years, worth roughly $3 billion
Power source: Calvert Cliffs' two existing reactors
Capacity secured: 690MW from current output, plus funding for 190MW of new generation
The twist: Amazon canceled plans last month to build a data center next to Calvert Cliffs after local opposition — this deal buys power only, with electricity still flowing to the regional grid
Side benefit: the long-term contract gives Constellation financial backing to seek license renewals for both reactors through 2034 and 2036

Why Is Every AI Company Chasing Nuclear Power?

Because AI data center electricity demand is growing faster than grids can build new plants. US data centers already consume roughly 176 TWh a year — about 4.4% of total US electricity — and global data center consumption is projected to nearly double from 485 TWh in 2025 to 950 TWh by 2030. The AI power crunch is not hype; it's a real constraint utilities deal with daily, and nuclear's steady, weather-independent output makes it the obvious pick for Big Tech.

Does This Nuclear Power Deal Actually Fix AI's Power Crunch?

Critics are skeptical. Even under optimistic assumptions, new small modular reactors (SMRs) would add only about 15 TWh a year — a sliver compared to the roughly 312.6 TWh US data centers were projected to need in 2025. The bigger issue is cost: new nuclear capacity is expensive and slow to build, and utilities often pass those costs on to ordinary ratepayers. A Gallup survey found 70% of Americans oppose new AI data centers in their own communities, citing rising electricity bills, water use, and noise.

What Does This Mean for Everyday AI Users?

Nothing changes in how ChatGPT or Claude responds to you today — but it may show up on your electricity bill eventually. Deals like this are mostly financial engineering: keeping existing reactors running and locking in long-term supply, not conjuring new power out of thin air. The "AI runs on clean nuclear" pitch sounds great in a press release, but real new capacity is still years away. 好不好用，試了才知道 — whether this infrastructure bet actually pays off won't be clear until the new capacity is actually built.

Sources / 資料來源

常見問題 FAQ

亞馬遜核電協議買了多少電力？

亞馬遜鎖定馬里蘭州Calvert Cliffs核電廠690MW既有產能，並資助190MW新機組，合約為期20年、總值約37億美元。

核電真的能解決AI缺電危機嗎？

短期內很難。就算樂觀估算，新建小型模組化反應爐（SMR）一年只能多生產約15TWh，相較美國資料中心動輒超過300TWh的用電需求，只是杯水車薪。

這類核電協議會讓我的電費變貴嗎？

有可能。新建核電產能成本高昂，電力公司常將建廠費用部分轉嫁到一般用戶電費上，這也是不少社區反對新建AI資料中心的主因之一。

亞馬遜為什麼不自己蓋新核電廠，而是買現有電廠的電？

蓋一座新核電廠動輒耗時十年以上，遠水救不了近火；買下既有電廠產能並協助電廠申請延役，是更快拿到穩定電力的做法。

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