2026年11月15日 星期日

美國AI 的未來

GCIOSGF Newsletter 會訊:

Reprinted 

美國民主黨在2026年期中選舉大勝,重掌參眾兩院,選民透過選票對現任政府的施政表達了強烈的抗議與不滿,包括對高通膨與生活成本的無法容忍,強烈反對捲入海外戰爭,捍衛醫療保障與基層社會福利,年輕世代與少數族裔的求變訴求。

AI資料中心推升用電需求 居民憂擴建設施轉嫁效應

經濟與民生是選民最關切的痛點。高達近九成的選民將生活成本視為投票的首要考量。物價與汽油價格的持續上漲,讓許多選民將經濟壓力歸咎於現任政府的關稅政策與海外軍事行動。

AI資料中心意外成為基層政治與民生經濟的衝突焦點。它在選戰中的角色不再僅限於科技發展指標,而是與能源消耗、電價上漲及地方環境資源緊密綁定。美國住宅電價自2024年以來已上漲超過10%,部分如俄亥俄州等地的漲幅甚至超過20%。隨著AI資料中心龐大的用電需求持續攀升,電力公司必須擴建發電廠與升級電網。選民強烈擔憂這些昂貴的基礎設施成本,最終會被轉嫁到一般家庭的電費帳單上,引發基層不滿。

AI資料中心的開發在各地遭遇強烈抵制。2026年的期中選舉,地方公投有很高比例直接針對資料中心開發具備「限制擴張」性質的提案,這反映了地方居民對於電力資源排擠、環境衝擊及土地過度開發的深刻疑慮。

美國本土AI資料中心未來挑戰

雖然民主黨奪回國會控制權後,預期將增加對AI產業的聽證與輿論監督,但真正決定AI資料中心建置進度的,是德州、喬治亞州、俄亥俄州與賓州等地的「州級電力核准」與「電費政策」。

州長與地方公用事業委員會的選舉結果,直接牽動了AI巨頭的擴展版圖。這些基層反彈與法規限制並不會完全停止美國的AI建設,但已產生實質的產業阻力。選舉過後,市場預期,未來美國本土的AI資料中心將面臨建設成本提高、專案完工時程延後,以及投資地點必須重新配置等挑戰。

在2026年的美國期中選舉中,全美共有十三個州出現了至少五十項與資料中心開發相關的地方公投。最終,共有至少三十八個地方轄區的選民投票通過了禁止或嚴格限制AI資料中心擴張的提案。這些公投主要集中在俄亥俄州、加州、緬因州、堪薩斯州、伊利諾州與佛羅里達州等地。

基層反彈與法規限制 已產生實質產業阻力

各地通過的具體限制內容可歸納為以下四大類:

(1)限制用電與建築規模:俄亥俄州的Conneaut市與堪薩斯州的Ottawa市,雙雙通過公投,禁止興建尖峰負載或總用電需求超過250MW的資料中心。緬因州的North Berwick則通過了禁止面積超過25,000平方英尺的資料中心開發案。

(2)實施暫停開發期:緬因州的Wells、Peru等城鎮通過了一百八十天的開發與許可凍結期。新罕布夏州的Rockingham County通過長達十二個月的禁令,期間鎮政府不得受理任何資料中心的許可申請。

(3)全面禁建:加州洛杉磯郡的Alhambra市通過Measure DB,修改土地使用規範,全面禁止在市內興建資料中心(支持率高達82%)。佛羅里達州的Edgewater市與俄亥俄州的 Hubbard市,也通過了禁止在市界內批准、建造或營運新資料中心的修正案。

(4)諮詢性反對與阻擋分區變更:在伊利諾州的Joliet Township,選民以高達79%的反對票,在諮詢性公投中明確表態反對市議會核准AI資料中心的建設。

除了地方層級的公投阻力,這股「鄰避效應」也向上延燒至州級選舉。例如在資料中心重鎮德州,支持在全州範圍內暫停資料中心開發的民主黨籍候選人Gina Hinojosa成功當選州長,這預示著AI基礎設施在未來將面臨更嚴苛的政治與法規挑戰。

科技巨頭四大策略調整

為了解決高達24至36個月的電網審批延宕與地方鄰避效應,微軟、亞馬遜和Google等科技巨頭主要採取了以下四大策略調整:

(1)避開一線重鎮,轉向次級市場與海外布局。次級市場往往缺乏現成的深層光纖骨幹,科技巨頭必須額外花費數億美元鋪設長途海底電纜或跨州陸上光纖,才能將偏遠算力連回核心網路。

(2)走向自備電力(BYOP)與電表後(Behind-the-Meter)微電網。這是目前最能立竿見影的解法。超過65%的資料中心營運商正在評估或使用現場發電方案。透過天然氣渦輪機或燃料電池,能將五年的等待期壓縮至一年多。

BYOP與微電網目前最立竿見影

天然氣發電不僅違背了企業的淨零碳排承諾,營運複雜度也極高(需要處理高壓天然氣管線與水冷卻資源)。此外,未來為了減碳,巨頭們還必須額外投資碳捕捉與封存(CCUS)技術。

(3)算力分散化,規避單一地區的審查門檻。對於AI模型訓練而言,這是最沒有效率的妥協。分散式運算打破了GPU叢集力求緊密連結的物理原則。為了解決這點,業界甚至開始將傳統衡量電源效率的PUE(電力使用效率)指標,轉向評估PCE(電力運算效率),以真實反映能源轉化為有效算力的比率。

(4)簽署彈性用電協議以安撫地方公用事業。從結果來看,這些策略非常有效地解決了時間延宕與算力部署卡關的致命危機,但代價是建置成本暴增、底層運算效率折損,以及營運複雜度的大幅提升。

2026年11月14日 星期六

美國容不下黃種人的路線之爭

GCIOSGF  Newsletter 會訊:

Reprinted 

誰在搞黃仁勳?兩家 AI 巨頭的路線之爭

一場圍繞著英偉達創始人黃仁勳的輿論風暴,正在美國科技界和華盛頓政治圈持續發酵。從美國主流媒體接連發布針對黃仁勳的負面報道,到國家安全政策專家公開質疑,再到多位國會議員要求調查,甚至呼籲其辭去英偉達CEO職務,這場風波在短時間內迅速升級。表面上看,這是一起涉及企業治理、私人關係與國家安全的爭議。但隨著新聞報道、政治獻金、智庫關係和政策遊說等線索逐漸浮出水面,一家AI公司的名字反覆出現——Anthropic。這家開發Claude人工智慧模型的公司,近年來迅速崛起為美國最具影響力的AI企業之一。而圍繞在黃仁勳的風暴,恰好發生在美國人工智慧產業路線競爭日益激烈的背景之下。更值得關注的是,多位在事件中扮演重要角色的記者、政策專家和國會議員,都與Anthropic有不同程度的職業、資金或政策連結。這究竟只是一系列巧合,還是美國AI產業權力遊戲的集中體現?黃仁勳與Anthropic之間的矛盾,首先源自於雙方對人工智慧產業發展方向的不同理解。身為英偉達的掌舵人,黃仁勳一直主張依靠技術創新、全球市場競爭和開放生態來推動AI發展。他認為,過於嚴格的晶片出口限制不僅可能壓縮美國科技企業的市場空間,也可能促使海外市場加速建立替代性的技術生態。同時,英偉達一直積極支持開放權重AI模式的發展,希望先進算力能被更廣泛的開發者和企業使用。 Anthropic則代表另一種路線。其CEO Dario Amodei長期強調先進人工智慧的安全風險,支援更嚴格的晶片出口管制,並主張加強對前沿AI模型的監管。兩種路線背後,是截然不同的產業利益結構。英偉達希望盡可能擴大AI算力的市場規模,而Anthropic更強調先進AI系統的安全治理和受控部署。當人工智慧逐漸成為決定未來全球經濟格局的關鍵技術時,圍繞著行業規則的爭奪,已經不再局限於企業之間的商業競爭,而是深入到了華盛頓的政治和監管體系。

媒體報導背後的人脈網絡

此次風暴的重要推手之一,是美國主流媒體的連續通報。 《紐約時報》記者Ana Swanson長期關注美國科技政策與晶片出口限制。她在過去的相關報導中,多次引用Anthropic共同創辦人Jack Clark的觀點。 Jack Clark曾任彭博社記者、OpenAI政策負責人,後來參與創立Anthropic。他不僅熟悉新聞媒體運作,也長期活躍於華盛頓科技政策圈。另一家重要媒體《金融時報》同樣存在值得關注的連結。該報記者Demetri Sevastopulo在黃仁勳事件中,對其私人關係及潛在利益衝突進行了報告。而《金融時報》先前也曾獲得Anthropic有關AI服務政策的重要獨家消息。記者與科技企業保持長期溝通,本身屬於正常的新聞工作。但當一家企業與新聞當事人在重大政策議題上有明顯分歧時,這種信源關係就有必要接受更多檢視。特別是在涉及一家全球科技巨頭CEO聲譽的敏感報告中,消息來源的獨立性與利益關聯,直接影響報告的公信力。華盛頓國家安全圈的關鍵人物隨著事件發酵,美國國家安全政策專家Chris McGuire透過X平台傳播相關報道,並以更具衝擊力的措辭強調黃仁勳的私人關係爭議,使事件迅速成為社交媒體熱點。 McGuire的職業背景引起了外界關注。他曾在拜登政府國家安全委員會任職,當時的上級Tarun Chhabra,如今正擔任Anthropic國家安全政策負責人。類似的人脈交集也出現在喬治城大學及其安全與新興技術中心CSET。 CSET長期研究人工智慧、半導體技術和國家安全問題,成立初期曾獲得Open Philanthropy的大規模資金支持。而該機構聯合創始人Holden Karnofsky後來加入Anthropic。在圍繞黃仁勳的爭論中,部分具有喬治城大學及相關政策研究背景的專家,也相繼參與公開討論。這些關係揭示了華盛頓科技政策圈的一個重要特徵:大型AI企業、政府前官員、智庫研究人員和媒體評論人士之間,存在著高度交叉的人脈網絡。當然,這些職業聯繫不能直接證明相關人員採取了協調行動,但它們足以引發潛在利益關係的討論。

4000萬美元背後的政治影響力

相較於人脈關係,Anthropic在華盛頓的政治資金佈局更引人關注。根據文章引述的政治獻金紀錄,2026年7月,Anthropic旗下政治行動委員會AnthroPAC曾向民主黨參議員Chris Coons捐贈2,500美元。 Coons後來公開支持對黃仁勳相關爭議展開調查。另一位公開要求黃仁勳辭職的共和黨參議員Kevin Cramer,也曾接受AnthroPAC提供的1,000美元捐款。更大規模的資金投入則指向Public First Action。 2026年,Anthropic宣布對這項AI政策倡議組織的累積支持達4,000萬美元。 Public First Action主要推動人工智慧安全政策和公共政策宣傳,其倡議的部分監管方向,與華盛頓支持加強AI技術管制的政治力量存在交集。這意味著,Anthropic不僅透過企業主管直接參與政策討論,還透過政治行動委員會、政策倡議組織和政府關係團隊,在華盛頓建立了廣泛的影響管道。需要強調的是,合法政治捐金並不代表議員的相關行動受到企業指使。 Anthropic也公開表示,向Public First Action提供的資金用於政策宣傳和公眾教育,而非直接影響特定候選人的選舉。但當接受相關捐款的政治人物,對Anthropic的重要政策對手公開施壓時,這些資金和政治關係自然成為值得關注的線索。

從AI企業延伸到國家安全智庫

Anthropic與華盛頓政策圈的聯繫,也體現在公司治理層面。美國新美國安全中心CNAS執行長Richard Fontaine,於2025年加入Anthropic長期利益信託。 CNAS是華盛頓重要的國家安全智庫,長期研究美國技術競爭策略、先進晶片出口限制及全球科技供應鏈安全。而Anthropic的長期利益信託,是其公司治理架構的重要組成部分,擁有參與董事會組成決策的權力。這意味著,Anthropic與美國國家安全政策界之間的聯繫,已經超越普通的企業諮詢關係。同時,Anthropic近年來也吸收了大量前白宮官員及人工智慧政策專家,包括Tarun Chhabra、Ben Buchanan、Elizabeth Kelly等人。這些人才佈局,使Anthropic擁有涵蓋行政部門、國會和智庫系統的政策溝通能力,也進一步強化了其在美國AI監管討論中的影響力。

黃仁勳離場,誰會受益?

黃仁勳不僅是英偉達的CEO,更是美國AI產業發展路線的重要倡議者。英偉達的晶片和運算平台支撐著全球大量AI企業,其領導層的商業策略將直接影響先進算力的供應、開放AI生態的發展以及國際市場的競爭格局。一旦黃仁勳因輿論壓力離開英偉達,新任領導層是否繼續堅持既有的全球商業策略,是否繼續支持開放AI生態,以及是否在監管問題上採取更加保守的立場,都將成為未知數。對於那些長期主張強化AI安全監管、擴大技術管制的企業而言,這種領導變化可能帶來不同的產業政策環境。而Anthropic近年來在華盛頓建立的廣泛政策網絡,使其具備了參與相關規則制定的能力。不過,從可能受益到主動策劃,仍然存在著巨大的證據距離。目前並沒有足夠的公開證據證明,Anthropic直接組織或協調了針對黃仁勳的負面報導和政治施壓。真正值得關注的是,當新聞報導、政策遊說、政治獻金、國家安全智庫和科技企業之間的關係不斷交織,美國人工智慧產業的競爭已經遠遠超出了產品性能、算力成本和模型能力的範疇。未來的AI技術應該更加開放,還是受到更嚴格的監管?全球先進算力的分配,應該多交由市場決定,還是透過政策手段加以限制?大型科技企業又將在多大程度上影響這些規則的製定?圍繞著黃仁勳的爭議,正成為這些深層矛盾的集中體現。

真正值得追問的,是美國 AI 產業的權力之爭

但當同一家公司在如此多的節點上反覆出現,而這家公司又與事件的核心人物存在實質性的政策利益衝突時,我們不能只滿足於一句「這都是巧合」。真正的問題是:這些關係究竟只是美國科技政策圈正常運作的結果,還是有尚未被公開揭露的協調行為?如果有人認為這是一場有組織的行動,就需要提供比人員交集和資金關係更直接的證據,例如實際溝通記錄、協調安排或明確的行動指示。但同樣重要的是,新聞媒體、政策研究機構和政治人物也應接受公眾對潛在利益衝突的檢視。黃仁勳的私人行為應由他自己負責。如果涉及企業治理或國家安全問題,就應依法調查。但對一個人的私人道德評價,不能取代對情報指控的事實核實;圍繞國家安全的合理審查,也不應該因為企業之間的商業競爭而失去透明度。這件事情真正令人關注的,是它觸及了美國人工智慧產業正在發生的一場深層權力競爭。未來最先進的AI技術該如何傳播?誰有權利決定哪些國家、哪些企業和哪些開發者能夠獲得算力?開放模型能否繼續發展?美國AI產業的政策方向,最終應該由市場競爭、技術創新和公開的民主程序決定,還是會受到少數擁有龐大政治資源的科技企業更大的影響?當一家AI公司能夠同時接觸國會、國家安全機構、頂尖智庫和主流媒體時,它的政策影響力究竟該受到怎樣的監督?這些問題,遠比黃仁勳的一段私人關係更重要。我們現在仍然無法證明,Anthropic就是這場危機的幕後組織者。但是,圍繞著黃仁勳的新聞傳播、政策施壓和政治資金網絡中,確實存在一系列值得公開檢驗的聯繫。一個巧合,可以是巧合。兩個巧合,也可能是巧合。但當記者、政策專家、國會議員、政治資金和國家安全智庫的關係不斷匯集到同一個產業利益衝突點時,我們至少應該認真追問,究竟發生了什麼事。所以,真正的問題不是黃仁勳有沒有犯錯,而是:在美國AI產業的巨大權力競爭中,究竟是誰在推動這場風暴,又是誰希望看到黃仁勳失去影響力?這個問題,值得繼續追查。

The Conflict Over the Path Forward for Asians in the United States

01 Who Is Targeting Jensen Huang? A Strategic Clash Between Two AI Giants

A public opinion storm surrounding Nvidia founder Jensen Huang is intensifying across the U.S. tech sector and the political circles of Washington. The controversy has escalated rapidly—from a series of negative reports on Huang in mainstream American media to public skepticism from national security experts, and even calls from several members of Congress for an investigation and his resignation as Nvidia’s CEO. On the surface, this appears to be a dispute involving corporate governance, personal relationships, and national security. Yet, as details regarding media coverage, political contributions, think-tank affiliations, and policy lobbying emerge, the name of one AI company keeps surfacing: Anthropic. The developer of the "Claude" AI model, Anthropic has risen in recent years to become one of the most influential AI firms in the United States. This storm surrounding Huang unfolds against the backdrop of an increasingly fierce competition over the strategic direction of the American AI industry. Notably, several journalists, policy experts, and lawmakers playing key roles in this affair share professional, financial, or policy-related ties to Anthropic. Is this merely a series of coincidences, or is it a concentrated manifestation of a power struggle within the U.S. AI sector? The friction between Jensen Huang and Anthropic stems, first and foremost, from their divergent visions for the development of the AI ​​industry. As the leader of Nvidia, Huang has consistently advocated for driving AI progress through technological innovation, global market competition, and an open ecosystem. He argues that overly stringent chip export restrictions could not only shrink the market reach of U.S. tech companies but also accelerate the creation of alternative, non-U.S. technological ecosystems abroad. Furthermore, Nvidia has actively supported the development of "open-weight" AI models, aiming to make advanced computing power accessible to a broader range of developers and enterprises. Anthropic, by contrast, represents a different path. Its CEO, Dario Amodei, has long emphasized the safety risks posed by advanced AI, supporting stricter chip export controls and advocating for more robust regulation of frontier AI models. Underlying these two approaches are fundamentally different industrial interest structures. Nvidia seeks to expand the market for AI computing power as broadly as possible, whereas Anthropic places greater emphasis on the safety governance and controlled deployment of advanced AI systems. As artificial intelligence increasingly becomes the pivotal technology that will shape the future global economic landscape, the struggle over industry rules has transcended mere commercial competition between firms, extending deep into the political and regulatory apparatus in Washington.

02 The Network of Connections Behind Media Coverage

A significant driving force behind this controversy has been a series of reports from mainstream American media outlets. *The New York Times* reporter Ana Swanson, who has long covered U.S. tech policy and chip export restrictions, has frequently cited the views of Anthropic co-founder Jack Clark in her past reporting. Clark—a former Bloomberg journalist and head of policy at OpenAI—is not only well-versed in the workings of the news media but has also long been an active figure in Washington’s tech policy circles. Another major outlet, the *Financial Times*, also presents noteworthy connections; its reporter, Demetri Sevastopulo, covered the controversy surrounding Jensen Huang, focusing on his personal ties and potential conflicts of interest. The *Financial Times* had previously secured exclusive scoops regarding Anthropic’s policies on AI services. While maintaining ongoing communication with tech companies is a standard part of journalistic practice, such relationships warrant closer scrutiny when a company and the subjects of a news report hold starkly different views on major policy issues. This is especially true for sensitive reporting that touches upon the reputation of a CEO of a global tech giant, where the independence—or potential conflicts of interest—of one’s sources directly impacts the credibility of the coverage. As the situation unfolded, Chris McGuire, an expert in U.S. national security policy, amplified these reports on the platform X. Using more provocative language to highlight the controversy surrounding Jensen Huang’s personal ties, he helped turn the matter into a viral topic on social media. McGuire’s professional background has drawn attention: he previously served on the National Security Council during the Biden administration, reporting to Tarun Chhabra—who now serves as the head of national security policy at Anthropic. Similar intersections of professional networks can also be found at Georgetown University and its Center for Security and Emerging Technology (CSET). CSET, which conducts long-term research on artificial intelligence, semiconductor technology, and national security, received substantial initial funding from Open Philanthropy. One of CSET’s co-founders, Holden Karnofsky, later joined Anthropic. Amid the controversy surrounding Jensen Huang, several experts with ties to Georgetown University and related policy research circles have also participated in public discussions. These connections highlight a key feature of Washington’s tech policy landscape: a highly interconnected network spanning major AI companies, former government officials, think-tank researchers, and media commentators. While these professional ties do not in themselves prove coordinated action, they are sufficient to prompt questions regarding potential conflicts of interest.


The Political Influence Behind $40 Million in Funding

Beyond personal networks, Anthropic’s political financial footprint in Washington warrants even closer attention. According to political contribution records cited in the article, Anthropic’s political action committee (PAC), AnthroPAC, donated $2,500 to Democratic Senator Chris Coons in July 2026. Coons subsequently voiced support for an investigation into the controversy involving Jensen Huang. Republican Senator Kevin Cramer, who publicly called for Huang’s resignation, also received a $1,000 contribution from AnthroPAC. A much larger sum was directed toward Public First Action; in 2026, Anthropic announced that its cumulative support for this AI policy advocacy group had reached $40 million. Public First Action focuses on AI safety and public policy advocacy, with some of its proposed regulatory approaches aligning with the political factions in Washington that favor stricter controls on AI technology. This indicates that Anthropic has established extensive channels of influence in Washington—not only through its executives’ direct participation in policy discussions but also via its PAC, policy advocacy groups, and government relations teams. It should be emphasized that legal political contributions do not imply that a lawmaker’s actions are directed by a corporation. Anthropic has also stated publicly that its funding for Public First Action is intended for policy advocacy and public education, rather than for directly influencing the election of specific candidates. However, when politicians who have received such donations exert public pressure on Anthropic’s key policy rivals, these financial and political ties naturally become noteworthy leads.


04 From AI Firms to National Security Think Tanks


Anthropic’s connections within Washington’s policy circles also extend to its corporate governance. In 2025, Richard Fontaine, CEO of the Center for a New American Security (CNAS)—a prominent Washington-based national security think tank—joined Anthropic’s Long-Term Benefit Trust. CNAS has long focused on U.S. technology competition strategy, export controls on advanced chips, and the security of global technology supply chains. Anthropic’s Long-Term Benefit Trust is a core component of its corporate governance structure, holding the authority to participate in decisions regarding the composition of the board of directors. This signifies that the relationship between Anthropic and the U.S. national security policy establishment goes well beyond standard corporate advisory arrangements. Furthermore, in recent years, Anthropic has recruited numerous former White House officials and AI policy experts, including Tarun Chhabra, Ben Buchanan, and Elizabeth Kelly. This strategic talent acquisition grants Anthropic the capacity to engage with the executive branch, Congress, and the think tank ecosystem, thereby strengthening its influence over U.S. regulatory discussions on AI.


05 If Jensen Huang Steps Down, Who Stands to Benefit?


Jensen Huang is not only the CEO of NVIDIA but also a pivotal advocate for the development trajectory of the U.S. AI industry. NVIDIA’s chips and computing platforms underpin a vast number of AI companies globally; consequently, the strategic decisions made by its leadership directly influence the supply of advanced computing power, the evolution of open AI ecosystems, and the competitive landscape of international markets. Should Jensen Huang depart from NVIDIA—perhaps due to public pressure—it remains unclear whether his successors would maintain the existing global business strategy, continue to support open AI ecosystems, or adopt a more conservative stance on regulatory issues. For companies that have long advocated for stricter AI safety oversight and expanded technological controls, such a leadership transition could foster a different industrial policy environment. Meanwhile, the extensive policy network Anthropic has cultivated in Washington positions it to play a role in shaping relevant regulations. That said, there remains a significant evidentiary gap between merely benefiting from such a scenario and actively orchestrating it. Currently, there is insufficient public evidence to prove that Anthropic directly organized or coordinated negative press coverage or political pressure directed at Jensen Huang. What truly warrants attention is the increasingly complex web of relationships—linking news media, policy lobbying, political contributions, national security think tanks, and technology firms—which suggests that competition within the U.S. artificial intelligence industry has extended far beyond mere product performance, computational costs, or model capabilities. Should future AI technology be more open, or should it be subject to stricter regulation? Should the global distribution of advanced computing power be left primarily to the market, or should it be constrained through policy interventions? And to what extent will major tech companies influence the formulation of these rules? The controversy surrounding Jensen Huang has become a focal point for these deeper, underlying tensions.

06 The Real Question: A Power Struggle in the U.S. AI Industry

When the same company repeatedly appears across so many different nodes—especially when that company has a substantive conflict of policy interests with the central figure in these events—we cannot simply dismiss it all as a "coincidence." The real question is: are these connections merely the byproduct of the standard functioning of the U.S. tech policy ecosystem, or is there evidence of coordination that has yet to be disclosed? If one posits that this is an organized campaign, then evidence more direct than mere overlapping personnel or financial ties—such as actual records of communication, coordinated arrangements, or explicit directives—would be required. Equally important, however, is that news media, policy research institutions, and political figures should be subject to public scrutiny regarding potential conflicts of interest. Jensen Huang must be held accountable for his own personal conduct; if issues concerning corporate governance or national security arise, they should be investigated according to the law. Yet, moral judgments about an individual cannot substitute for the factual verification of allegations involving intelligence or security. Furthermore, legitimate national security reviews must not lose their transparency simply because of commercial competition between corporations. What is truly significant here is that this situation touches upon a profound power struggle currently unfolding within the U.S. artificial intelligence industry. How should the most advanced AI technologies be disseminated in the future? Who has the authority to decide which nations, companies, or developers gain access to computing power? And can open-source models continue to evolve? Should the policy direction of the U.S. AI industry be determined by market competition, technological innovation, and transparent democratic processes—or will it be increasingly shaped by a handful of tech firms wielding vast political resources? When an AI company gains simultaneous access to Congress, national security agencies, top-tier think tanks, and mainstream media, what kind of oversight should govern its policy influence? These questions are far more significant than any single personal connection involving Jensen Huang. While it cannot currently be proven that Anthropic is the orchestrator behind this crisis, there are indeed a series of links—spanning media narratives, policy lobbying, and political funding networks—that warrant public scrutiny. One coincidence might just be a coincidence; two could also be a coincidence. But when the paths of journalists, policy experts, members of Congress, political donors, and national security think tanks all converge on a single point of industrial conflict, we are obligated to ask: what is actually happening here? Ultimately, the real issue is not whether Jensen Huang made a mistake, but rather: who is driving this storm within the high-stakes power struggle of the U.S. AI industry, and who stands to gain from seeing Jensen Huang’s influence diminish? This is a question that deserves further investigation.

2026年11月13日 星期五

回答下游廠商的未來

 GCIOSGF. Newsletter 會訊:

Reprinted 

不管誰接班,最成功的劇本,大概就是複製庫克。庫克從賈伯斯手上接過蘋果,任內讓公司市值成為接班時的約十三倍。川普也直接點名,希望由他接手輝達。對股東而言,這當然是個有吸引力的故事。

但我想提醒,庫克能讓蘋果股東賺錢,不代表他會讓輝達供應鏈繼續賺規格升級的錢。

蘋果主要做消費者生意,輝達做的是企業生意。你嫌iPhone升級太慢,頂多換一支Android手機,不會自己開公司做手機。輝達那幾個大客戶卻不一樣。Google、亞馬遜、微軟與Meta,有錢、有人,也有能力自己設計晶片。你慢下來,客戶就有更多理由自己做。

庫克的本事,是把既有產品、生態系與供應鏈經營得更賺錢。我認為,這套經營方式對新技術的態度比較謹慎,跟黃仁勳不斷拉高功耗、封裝與機櫃規格,逼著整條產業鏈往前跑,是兩種不同的生意。

專業經理人接班,股價未必跌。控制成本、提高現金流、買回股票,都可能讓股東高興。但台灣供應商吃的是新規格、新設備、新產能的飯。輝達把錢還給股東,跟把錢拿去推下一代機櫃,對台灣是完全不同的兩套帳。

因此,我最在意的,是那些照著黃仁勳的規格擴廠、買設備,甚至已經把2028年獲利算進股價的公司。設備訂了,錢花了,當初拍板的人卻走了。接班人還願不願意照原來的速度往前衝?

輝達股東可能更賺,台灣供應鏈卻可能少賺

正在出貨的產品,短期未必受到明顯影響。真正承受變數的,是押注Rubin Ultra、Feynman及後續平台升級,靠未來規格拉高售價與毛利的業者。眼前的訂單可以照交,下一代的生意卻可能重新談。

輝達的技術不是黃仁勳一個人做的,但機櫃規格要多激進、產品要推多快、供應商要提前投資多少錢,高度倚賴他拍板。晶片升級跟機櫃升級那是兩件事。晶片能做多小、多快,主要看台積電的製程走到哪裡。從3奈米到2奈米,時間表大致掌握在台積電手上,就算輝達換了老闆,製程照樣往前走。機櫃不一樣。一個櫃子要塞幾顆GPU、用什麼方式連在一起、電怎麼供、熱怎麼散,沒有一家代工廠能替輝達決定。從72顆GPU一櫃,到下一代Kyber要塞進數百顆,這條路是黃仁勳一手往前推的。晶片升級有台積電撐著,機櫃升級卻少了最敢拍板的人。

創辦人可以堅持一項多年不賺錢的技術,職業經理人卻得向董事會解釋,為什麼不能先做比較穩、比較容易賺錢的版本。機櫃一定要這麼複雜嗎?新封裝一定要現在上嗎?能不能沿用既有設計,再賣一年?

這些問題都合理。但對已經押下設備與產能的台灣廠商,每一個合理的問題,都可能變成一筆不太合理的成本。

所以我擔心的,不是明天GPU突然做不出來,而是下一代機櫃不再那麼急著升級。原本一次到位,改成分階段;原本今年定案,改成明年再看。新老闆把風險管好了,供應商期待的成長卻不見了。

輝達可以說產品路線圖不變,但方向不變,不代表速度不變。供應商要看的,是規格何時定案、採購何時簽下、數量何時確認。路線圖上的箭頭,不能拿去付設備款。

一旦升級放慢,獲利與估值會一起承壓。原本明年可以賣更貴的零組件,現在得晚一年;原本市場願意給40倍本益比,現在只肯給30倍。輝達的股價可能因為經營更穩健而上漲,台灣供應商卻因為成長變慢而下跌。

專業經理人可以替輝達股東創造價值,卻未必會替台灣供應鏈保住黃仁勳留下的夢。


2026年11月11日 星期三

Claude新制 Claude New Policy

GCIOSGF Newsletter 會訊:

Reprinted 

 經常往返中國大陸出差或旅遊的 AI 工具使用者需特別注意!AI 巨頭 Anthropic 宣布將於明(12)日起實施全新版《地區支援政策》(Supported Regions Policy)。新制不僅針對中國大陸當地的企業與個人進行存取封鎖,更將限制範圍擴大至「所有身處不支援地區」的人員。屆時即便使用者持台灣帳號、使用台灣 eSIM,或是具備其他國家國籍,只要實際踏入中國大陸境內,都將無法繼續使用 Claude 相關服務。

去中國旅遊注意!明天11/12起新規上路「不看國籍轉機出差都受影響」

根據 Anthropic 官方公布的政策規範,新版地區限制將於 2026 年 11 月 12日施行。

本次限制涵蓋的服務範圍極廣,包含:

.官方平台: Claude.ai 網頁端、Claude Code、Cowork。

.開發與企業端: 官方 API、開發者平台。

第三方通路: 透過公有雲平台(如 AWS、GCP 等)以及授權經銷商所提供的 Claude 模型服務。

Anthropic 官方明確表示,系統判定是否符合使用資格時,除了驗證使用者「當下實際所在的地理位置」之外,亦會同步審查企業的註冊地、總部所在地以及背後實際的股權控制關係。

人在中國大陸即無法使用!帳號註冊地與個人國籍皆不豁免

本次政策更新對台灣及跨國企業員工影響最為直接。Anthropic 於規範中明確指出,任何人在未開放服務的地區境內使用服務均屬違規,包含任職於支援地區企業的員工,也不得因個人國籍或帳號原始註冊地而獲得豁免。

這意味著,過去許多台灣使用者習慣透過海外帳號或行動漫遊(如台灣 eSIM)在中國大陸境內連線使用 Claude 的方式,在 11 月 12 日新制上路後均屬違反官方服務條款。使用者若於中國大陸出差、旅遊或短暫轉機期間強行連線,將面臨無法使用服務或帳號遭系統直接停權(Ban)的風險。

企業背景與股權審查升級!中國企業員工人在美國同樣受限

除了使用者的地理位置外,新版政策亦大幅提高對企業實體背後架構的審查強度,凡於不支援地區註冊或設有總部的企業,旗下員工即使實際位於美國、台灣等合法支援地區,仍無法使用 Claude 服務。

若一家企業直接或間接由不支援地區的個人或實體持有過半股權(或具備實質控制權),即使該公司登記於美國、新加坡等合規國家,其員工與相關使用者仍會被列入禁播範疇。

官方補充強調,此項控制權限制主要針對「企業實體與股權關係」,並不代表所有在海外(如美國)工作的中國籍個人會自動失去使用資格。

台灣市場仍正常支援!專家建議出差前做好備份

事實上,Anthropic 早於 2025 年 9 月即已開始收緊對不支援地區企業的銷售限制,當時禁止由相關地區企業持股超過 50% 的公司使用服務。本次修訂則是將個人所在地、公司註冊地、總部位置與最終控制權進行全面整合。

目前台灣仍屬於 Anthropic 官方明訂的合規支援地區,國內使用者在台灣境內皆可正常使用。不過,鑑於新規將於 11 月 12 日生效,專家建議經常需要跨國出差或前往中國大陸的工作者,出發前應先確認自身帳號屬性與企業架構,並妥善備份工作資料,以免因地理位置切換造成業務中斷。

Claude 新版地區政策常見問題

Q1:我是台灣人,拿台灣電信的漫遊卡去中國大陸出差,可以使用 Claude 嗎?

不行。新版政策採「實際地理位置」為判定基準,只要人位於不支援地區(如中國大陸境內),即不符合使用資格,使用漫遊或海外帳號仍可能面臨功能鎖定或帳號停權。 

Q2:如果只是在中國大陸的機場轉機,可以使用 Claude 處理工作嗎?

不建議。轉機期間只要設備連接至當地基地台或機場網路,即進入不支援地區範圍,隨時可能觸發 Anthropic 的地區封鎖機制。

AI tool users who frequently travel to or from mainland China for business or tourism should take note! AI giant Anthropic has announced a new version of its Supported Regions Policy, effective tomorrow (November 12). The new policy not only blocks access for businesses and individuals in mainland China but also expands the restrictions to include "all individuals located in unsupported regions." At that time, even if a user holds a Taiwanese account, uses a Taiwanese eSIM, or holds citizenship of another country, they will be unable to use Claude-related services once they actually enter mainland China.

Take note if you're traveling to China! New regulations take effect tomorrow, November 12 – "No matter your nationality, transit and business trips will be affected."

According to Anthropic's official policy, the new region restrictions will take effect on November 12, 2026.

The scope of services covered by this restriction is extremely broad, including:

• Official platforms: Claude.ai web page, Claude Code, Cowork.

• Developer and enterprise platforms: Official API, developer platform.

Third-party channels: Claude model services provided through public cloud platforms (such as AWS, GCP, etc.) and authorized resellers.

Anthropic officially stated that when determining eligibility, in addition to verifying the user's "current actual geographical location," the system also simultaneously reviews the company's registered location, headquarters location, and actual equity control relationships.

Users residing in mainland China are prohibited from using the service! Account registration location and nationality are not exempt.

This policy update has the most direct impact on Taiwanese users and employees of multinational corporations. Anthropic's regulations clearly state that anyone using the service in a region where it is not available is in violation of the rules, including employees of companies in supported regions, and no exemptions will be granted based on nationality or the original registration location of the account.

This means that many Taiwanese users who previously used overseas accounts or mobile roaming (such as Taiwan eSIM) to connect and use Claude in mainland China will be violating the official terms of service after the new regulations take effect on November 12th. Users who attempt to connect while on business, traveling, or during a short layover in mainland China will face the risk of being unable to use the service or having their accounts directly banned by the system.

Company Background and Equity Verification Upgraded! Employees of Chinese Companies in the US Are Also Restricted

In addition to the user's geographical location, the new policy significantly strengthens the verification of the structure behind the company entity. Employees of companies registered or headquartered in unsupported regions, even if physically located in legally supported regions such as the US or Taiwan, will still be unable to use Claude services.

If a company is directly or indirectly held by an individual or entity in an unsupported region with more than half of the equity (or has substantial control), even if the company is registered in compliant countries such as the US or Singapore, its employees and related users will still be banned.

The official statement emphasizes that this control restriction mainly targets "corporate entities and equity relationships" and does not mean that all Chinese individuals working overseas (such as in the US) will automatically lose their eligibility.

Taiwan market remains supported! Experts recommend backing up your data before business trips.

In fact, Anthropic tightened sales restrictions on companies in unsupported regions as early as September 2025, prohibiting companies with more than 50% ownership in those regions from using the service. This revision comprehensively integrates personal location, company registration location, headquarters location, and ultimate control.

Currently, Taiwan remains a compliant supported region officially designated by Anthropic, and users within Taiwan can use the service normally. However, given that the new regulations will take effect on November 12th, experts recommend that workers who frequently travel internationally or to mainland China confirm their account attributes and company structure before departure and properly back up their work data to avoid business interruptions due to geographical location changes.

Claude New Regional Policy FAQs

Q1: I am Taiwanese and use a Taiwan Telecom roaming card when traveling to mainland China for business. Can I use Claude?

No. The new policy uses "actual geographical location" as the criterion for eligibility. If the user is located in an unsupported region (such as mainland China), they are ineligible to use the service. Using a roaming or overseas account may still result in feature lockout or account suspension.

Q2: Can I use Claude for work if I'm just transiting through an airport in mainland China?

Not recommended. During a transit, as long as the device connects to the local base station or airport network, it enters an unsupported region and may trigger Anthropic's region blocking mechanism at any time.

GCIOSGF

美國AI 的未來

GCIOSGF Newsletter 會訊: Reprinted  美國民主黨在2026年期中選舉大勝,重掌參眾兩院,選民透過選票對現任政府的施政表達了強烈的抗議與不滿,包括對高通膨與生活成本的無法容忍,強烈反對捲入海外戰爭,捍衛醫療保障與基層社會福利,年輕世代與少數族裔的求變訴...