原文標題： Forget chips: The next battleground in the U.S.-China AI race is nuclear power 原文連結：https://reurl.cc/GNqv9y 發布時間：Sept. 20, 2025 at 8:00a.m. ET 記者署名：Christine Ji 原文內容： Investors are paying close attention to the computer-chip war erupting as the U.S. and China race for global leadership in artificial intelligence — yet Wall Street is focused on the wrong battle. 投資人正密切關注美中在爭奪人工智慧全球領導地位時爆發的晶片戰爭——但華爾街卻把 焦點放錯了地方。 The real winner will come down to something much less flashy: energy. Specifically, nuclear energy. 真正的勝負將取決於一個遠不如晶片光鮮亮麗的領域：能源，尤其是核能。 After all, the most advanced AI chips in the world are useless if you can’t turn them on. Nuclear energy is rapidly becoming one of the foremost solutions to power the AI boom — and prevent impending energy shortages. 畢竟，即使是全球最先進的 AI 晶片，如果無法啟動，也毫無用處。核能正迅速成為支撐 AI 熱潮、並防止能源短缺的首要解決方案之一。 The nuclear option 核能選項 “I think that one of the best sources of data-center electricity is going to be from nuclear power, and of this there is very little doubt,” Richard Windsor, founder of the research firm Radio Free Mobile, told MarketWatch. “ If you want reliable electricity without carbon, nuclear is pretty much your only option.” 「我認為資料中心最好的電力來源之一將來自核能，這幾乎沒有疑問，」研究公司 Radio Free Mobile 創辦人 Richard Windsor 告訴《MarketWatch》。 「如果你要可靠且無碳 的電力，核能幾乎是唯一選擇。」 According to Bank of America, data centers used to power AI globally could use more energy than the entire country of Japan by 2026. 根據美國銀行的資料，到 2026 年，用於 AI 的全球資料中心耗電量可能會超過整個日本 。 According to the research provider BloombergNEF, the amount of electricity flowing through global electricity grids is expected to surge 30% as soon as 2030. 根據研究機構 BloombergNEF，至 2030 年，全球電網中的電力流量預計將增加 30%。 Companies like the cloud-infrastructure provider CoreWeave Inc. have identified energy as a critical bottleneck for AI development. 像雲端基礎設施供應商 CoreWeave 這樣的公司，已將能源視為 AI 發展的關鍵瓶頸。 Bank of America research strategist Felix Tran highlighted nuclear energy as an emerging “necessity” for future AI growth due to its low lifetime costs, minimal carbon emissions and continuous baseload power. 美國銀行研究策略師 Felix Tran 指出，由於核能具備低長期成本、極少碳排放，以及可 提供穩定基載電力，它正成為 AI 成長的「必要條件」。 Windsor emphasized its ability provide continuous, round-the-clock power — something that traditional renewables, such as wind and solar, can’t guarantee. Windsor 強調，核能可以提供持續、全天候的電力，而這是傳統再生能源（如風能與太陽 能）無法保證的。 Nuclear technologies 核能技術 Most nuclear plants today use fission to split large atoms like uranium to release heat and generate electricity. 當今大多數核電廠使用「核分裂」，透過分裂如鈾這樣的大原子來釋放熱能並發電。 There are a few small modular reactors (SMRs) in operation in China and Russia , which represent the next generation of advanced, smaller-scale fission technology. 目前在中國和俄羅斯已經有少數「小型模組化反應爐」（SMR）投入運轉，這代表著下一 代較小規模的先進核分裂技術。 But the real “holy grail” will be nuclear fusion, which works by fusing light atoms like hydrogen together at extremely high temperatures to release a massive amount of energy, according to Bank of America. 但根據美國銀行的說法，真正的「聖杯」將是「核融合」——透過在極高溫下融合像氫這 樣的輕原子，以釋放龐大的能量。 ‘Build, baby, build!’ 「建設吧，快建！」 The U.S. government has picked up on the nuclear opportunity with a sense of increasing urgency as it aims to close a reported 10-to-15-year development gap with China, according to the Energy Department. 根據美國能源部，美國政府已意識到核能的機會，並帶著越來越緊迫的心態，力求縮小與 中國據稱長達 10 至 15 年的發展差距。 While U.S. energy capacity overall has been stagnant for over a decade, China is adding the equivalent of the entire U.S. power capacity to its own grid every 18 months. 雖然美國整體能源產能停滯已超過十年，但中國每 18 個月就為自家電網新增相當於整個 美國發電量的電力。 At the current pace, China could overtake the U.S. in nuclear capacity by 2030 , according to BofA’s Tran. 依照美國銀行 Tran 的估計，照這種速度，中國到 2030 年可能在核能發電能力上超越美 國。 In May, President Donald Trump issued four executive orders aimed at kickstarting nuclear investment. 2025 年 5 月，川普總統簽署了四道行政命令，目的是推動核能投資。 Trump’s July 2025 AI Action Plan emphasized the need to develop AI infrastructure — especially nuclear-energy capabilities — and called on the nation to “build, baby, build!” 川普在 2025 年 7 月的《AI 行動計畫》中強調，發展 AI 基礎建設（尤其是核能能力） 是必要的，並呼籲全國「建設吧，快建！」 And nuclear energy was one of the top discussion topics during Trump’s visit to the U.K. this week, as Trump and British Prime Minister Keir Starmer signed a technology partnership involving investments in AI and nuclear power. 本週川普訪問英國時，核能也是主要討論議題之一。他與英國首相基爾·斯塔摩簽署了一 項技術合作協議，其中涉及 AI 與核能投資。 Can nuclear deliver? 核能能實現嗎？ Digital infrastructure has become critically intertwined with economic and military power, increasingly aligning public and private nuclear initiatives, Sean Farney, vice president of JLL’s Americas data-center strategy, pointed out. JLL 美國數據中心策略副總裁 Sean Farney 指出，數位基礎建設已與經濟和軍事力量緊 密結合，這也使得公共與私人核能計畫日益趨於一致。 Big Tech companies are increasingly dedicating resources to the development of nuclear energy; Nvidia, Microsoft, Meta, Amazon and Alphabet have all made nuclear-energy investments as they scramble to secure megawatts of power in the race to build the most advanced AI models. 大型科技公司正越來越多地投入資源於核能發展；Nvidia、Microsoft、Meta、Amazon 和 Alphabet 都已投資核能，以便在打造最先進 AI 模型的競賽中搶先確保電力供應。 Building nuclear infrastructure is no easy feat. While SMRs are cheaper and quicker to build than traditional energy plants, fusion technology is still nascent. 建設核能基礎設施並不容易。雖然小型模組化反應爐比傳統電廠更便宜、建造更快，但核 融合技術仍處於初期階段。 Benjamin Lee, an engineering and computer-science professor at the University of Pennsylvania, sees potential in the technology but acknowledged pragmatic constraints. 賓夕法尼亞大學工程與電腦科學教授 Benjamin Lee 看好這項技術的潛力，但也承認存在 現實限制。 “Nuclear is not a near-term solution. It may not even be a medium-term solution,” Lee told MarketWatch. “We don’t have an established industry that knows how to build nuclear on time and on budget.” 「核能不是短期解決方案，甚至可能不是中期解決方案，」Lee 告訴《MarketWatch》。 「我們還沒有一個能準時並控制成本建造核電廠的成熟產業。」 By contrast, Tran pointed out that China can build nuclear plants in a three- to five-year time frame. 相比之下，Tran 指出，中國能在 3 到 5 年內建好核電廠。 Strict regulation, high upfront costs and a loss of expertise after a long pause in nuclear-plant building all pose challenges to nuclear development in the U.S. today. 嚴格的監管、高昂的前期成本，以及因長期停滯而流失的專業技術，都是美國核能發展的 挑戰。 However, the urgent nature of the AI race could help alleviate the “ tremendously high regulatory hurdle,” according to Farney, as the Trump administration has been pursuing energy deregulation. 然而，根據 Farney 的說法，AI 競賽的緊迫性可能有助於降低「極高的監管門檻」，因 為川普政府正推動能源去管制化。 With increased government support, Farney estimates that the time to deploy SMRs in the U.S. could be cut down from seven years to three. 在政府加大支持的情況下，Farney 估計美國建置 SMR 的時間可以從 7 年縮短至 3 年。 And with Big Tech companies allocating parts of their massive capital- expenditure budgets to nuclear energy, the industry is seeing a fresh wave of private investment which could accelerate development, according to Lee. 此外，隨著大型科技公司將龐大資本支出的一部分投入核能產業，該產業正在迎來新一波 私人投資潮，這可能加速發展。 It all comes down to “getting a big-enough order book for new nuclear generation,” he said, as repeated construction will be the only way for the U .S. to rebuild its nuclear expertise. 他說，最終關鍵在於「獲得足夠大的新核電訂單」，因為只有持續的建設才能讓美國重建 其核能專業能力。 The stakes are high, as both Big Tech’s financial future and U.S. national security are on the line. But Farney is optimistic that the energy industry will achieve a breakthrough. 賭注極高，因為這同時攸關大型科技公司的財務前景與美國的國家安全。但 Farney 樂觀 地認為能源產業會取得突破。 There’s “a whole lot of money involved,” Farney said, “and an incredible spirit of innovation.” Farney 說：「這裡涉及的資金龐大，並且充滿了驚人的創新精神。」 心得/評論： AI 競賽的下一步是「能源戰」。 誰能先解決 AI 用電的結構性短缺， 誰就可能在算力與 AI 模型競爭中領先。 中國在核能建設效率與國家推動上領先， 美國則依靠川普政策與 Big Tech 資本來追趕。 這不僅是科技與經濟問題，更直接關乎國家安全。 -- ※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 150.117.197.145 (臺灣) ※ 文章網址: https://www.ptt.cc/bbs/Stock/M.1758449896.A.652.html 核電也是綠電 ※ 編輯: TyuzuChou (150.117.197.145 臺灣), 09/21/2025 18:37:14