2025年5月8日，美國華盛頓特區國會山，OpenAI首席執行官山姆・奧爾特曼(Sam Altman)、Advanced Micro Devices（AMD）董事長兼首席執行官蘇姿豐、CoreWeave聯合創始人兼首席執行官 Michael Intrator和微軟公司副董事長兼總裁布拉德·史密斯在美國參議院商務、科學和交通委員會舉行的題為 "贏得人工智能競賽：強化美國在計算與創新方面的能力"的聽證會上就座作證。

進入2025年以來，美國鞏固人工智能（AI）主導權的動作不斷，尤其是受Deepseek出圈刺激，布局節奏明顯加快。

特朗普政府以制定新“人工智能行動計劃”等戰略規劃為牽引，面向社會廣泛徵集意見，激發美國國內掀起AI戰略大討論。4月25日，美國政府公開了前期收集到的意見“萬言書”，總數達10068份。蘭德公司、新美國安全中心等知名智庫也集中發力，發佈大量聚焦AI尤其是所謂競贏中國的報告。微軟、AMD、OpenAI等科技巨頭也在5月8日美國國會參議院召開的“贏得全球人工智能競賽”聽證會上各抒己見。綜觀此輪“朝野大辯論”及美國政府近期的政策動向，辯論中的“主流意見”正逐步轉化為政策實踐，下一步動作已呼之欲出。

對內

進一步放鬆國內監管，激發創新競爭活力，着力補齊自身“短板”。相關專家指出，中美AI之爭本質上是發展之爭，當前美國雖仍具有一定領先地位但並不穩固。而且AI競爭剛剛開始，當前的領先並不重要，重要的是未來能夠持續領先。因此他們建議在“揚長”的同時必須更加積極地“補短”，而所謂的“短”主要集中在以下幾方面：

一是能源。AMD董事長蘇姿豐表示，AI產業是能源密集型產業，擴展規模需要大量的能源供給。當前美電基建項目審批繁瑣，建設滯後，無法滿足AI發展需求。微軟總裁史密斯建議，美國政府應合并聯邦電網審批機構，並應用AI數字工具來提升電網建設效率，同時擴建綠色能源發電廠，實現能源結構多元，以充足的電力供應保障AI技術長遠發展。

二是數據。OpenAI、微軟和谷歌等科技巨頭領導人均表示，美國現行涉及版權、隱私及專利保護的法律和政策均制約美企AI模型獲取用於訓練的數據。美國政府不僅應向本國AI企業開放聯邦政府數據訪問權，以便其獲取高質量、規模化的官方數據集用於AI模型訓練，還應大力推動歐盟等其他國家和地區調整版權保護領域的法律和政策框架，允許AI模型“參考和提煉”更多具有知識產權的數據，擴充合法訓練數據來源。

三是人才。據美國“進步研究所”（IPF）統計，美國頂尖AI公司中，近60%是由高技能移民參與創立。本土技術人才短缺已成制約美國AI發展的最大挑戰。黃仁勛也表示，全球AI研究人員有半數是中國人，美國本土人才優勢岌岌可危。為此，各界建議美國政府應放寬移民政策，提高人才留美比例，適度放鬆對中國公民的審查和限制，吸引中國人才赴美工作。同時，應完善STEM(科學、技術、工程、數學)教育體系，將美國AI教育下沉至K-12（基礎教育）階段，使其成為學齡兒童通識教育的一部分。此外，應打造產學研聯合培養模式，提高高校畢業生就業適配率，尤其要重視填補電力和數據運維人才缺口，推出技術人員職業培訓方案。

目前特朗普政府已根據各方建議在“短板”領域展開布局，並簽署多項行政令要求部署核反應堆、加強電網建設、推動AI通識教育培養等，意欲在政策層面為美國AI基礎設施建設及人才培養掃清障礙。

對外

在對華遏制的基礎上，着力從全球生態布局的高度不斷強化主導優勢。美國業界一致認為，中美兩國領先的一方將能制定標準並藉此佔據最終優勢，落後的一方則將陷入完全的被動，而未來制勝的關鍵還在於能否最終主導全球AI生態。鑒於此，美國下一步對外政策重心將集中體現在以下兩方面：

一是“優化”對華遏制措施。

美國各界認為，之前的芯片出口管制政策已對美國搶佔全球芯片市場產生阻礙：一方面間接推動了中國高端芯片的自主化進程，另一方面也制約了美企拓展海外市場的腳步。英偉達首席執行官黃仁勛近期接受專訪時認為，若中國能以華為芯片替代英偉達芯片實現AI訓練的全鏈路自主化，中國或可在全球範圍內對美國技術進行全面替代，進而削弱美國技術優勢。因此，對華措施要追求“平衡”與“實效”，不能一味“殺敵一千，自損八百”，要在限制發展與佔有中國市場份額及把控技術發展路線等方面找到“最佳”平衡點。

二是“加快”全球AI生態布局。

蘭德公司技術與安全政策中心計算團隊負責人海姆認為，全球產業主導權的關鍵往往在於能否進行大規模技術推廣，而非實驗室里的技術突破，為此必須將美國AI技術與服務推向全球更多地方。近來種種跡象表明，美國政府已汲取意見，迅速採取行動搶佔全球市場。特朗普政府正開展“以芯換資”，通過放鬆芯片制裁與盟國開展投資及關稅談判，以便充分利用技術優勢和政策工具佔領全球市場。目前美國已同海灣國家就出口芯片及建設數據中心達成一致。除政府外，科技巨頭也成為美國開拓海外市場的“排頭兵”。近來特朗普中東之行的“科技天團”格外惹眼，巨頭們還以“星際之門”為牽引不斷拓展全球AI基建市場，OpenAI繼“阿聯酋星際之門”（Stargate UAE）項目之後，還計劃在韓國、新加坡、澳大利亞、印度和日本等地尋求進一步合作的機會。

OpenAI chief executive Sam Altman; Lisa Su, CEO and chair of Advanced Micro Devices (AMD); Michael Intrator, co-founder and CEO of CoreWeave; and Brad Smith, vice chair and president of the Microsoft Corporation, are seated to testify before a Senate Commerce, Science, and Transportation Committee hearing titled Winning the AI Race: Strengthening US Capabilities in Computing and Innovation, on Capitol Hill in Washington, DC, US, May 8, 2025.

Since the beginning of 2025, the United States has moved steadily to consolidate its dominance in the field of artificial intelligence. Corresponding deployments have accelerated, particularly after DeepSeek's sensational debut.

Driven by such strategic planning as the newly proposed “Artificial Intelligence Action Plan,” the administration of Donald Trump is soliciting opinions from the general public, triggering nationwide discourse about America's AI strategy.On April 25, the U.S. government published 10,086 pieces of advice it had collected in the first stage. Such well-known think tanks as RAND and CNAS have also issued a great number of AI-focused reports, concentrating specifically on outcompeting China. Tech giants such as Microsoft, AMD and OpenAI aired their views at a U.S. Senate hearing on May 8 aimed at “winning the AI race.” Judging from the nationwide response and the Trump administration's recent policy orientation, the mainstream opinions that emerged in the process are gradually turning into policy practices. These, in turn, provide credible clues about what the next steps will be.

Domestically, in addition to a further relaxing of oversight to inspire innovation and competitiveness, the focus will be on addressing areas where America comes up short. China-U.S. AI competition boils down fundamentally to development ability, an area where the U.S. still holds certain advantages. But those advantages are no longer secure. The competition over AI has just begun, and current advantages are not what matters most. Rather, the key question is who will lead sustainably in the future. Therefore, while acknowledging the current U.S. advantages, China feels a pressing need to proactively make up for U.S. weaknesses. These are mainly concentrated in the following aspects:

As AMD's President Lisa Su noted that the AI industry is energy-intensive, and expanding it will require tremendous energy supplies. U.S. power infrastructure, which is currently mired in cumbersome approval procedures and slow construction, will not meet the needs for AI development.Microsoft President Brad Smith suggested the U.S. government should consolidate its federal grid approval agencies and improve construction efficiency by means of AI digital instruments. At the same time, it should expand green energy power plants to achieve energy diversification and ensure long-term development of AI technologies supported by sufficient driving power.

Second is data. OpenAI, Microsoft and Google executives believe all current U.S. laws and policies involving copyright, privacy and patent protection prevent U.S. AI models from receiving data for training. The government should not only grant U.S. AI companies the right to access federal databases — thereby acquiring high-quality, large-scale official data for training the AI models but should also proactively promote other countries and regions, such as the European Union. This would allow adjustments to their legal and policy frameworks in the field of copyright protection, so that AI models would be able to reference and extract more intellectual property data and expand the data sources for their legal training.

Third is talent. Nearly 60 percent of leading AI companies in the U.S. were created with participation by high-skill immigrants, according to the Institute for Progress (IFP). Shortages of local tech talent have become the biggest challenge constraining U.S. AI development. Jensen Huang, the head of Nvidia, also noted that half of global AI researchers are Chinese, and that the advantage of indigenous U.S. talent is on a shaky ground. Therefore, many people have suggested the U.S. government should further loosen its immigration policies, raise the ratio of talent staying in the U.S., properly relax checks and constraints on Chinese citizens and try to attract Chinese talent to work in the United States.At the same time, it should improve the educational regime for STEM disciplines (science, technology, engineering, math) and start AI education in the K-12 (primary education) stage, making it a part of general education for school-age children. It should create a training model that integrates production, education and research; increase the employment matching rate for college graduates, with a particular focus on addressing the talent undersupply for power and data operation maintenance; and come up with vocational training programs for tech workers.

The Trump administration is already making deployments in several areas of weakness based on those suggestions, and the president has signed multiple executive orders to bring on nuclear reactors, enhance power grid construction and promote general AI education, intending to clear the way at the policy level for U.S. AI infrastructure construction and talent cultivation.

Internationally, the administration wants to build on U.S. containment of China and focus on continuously strengthening advantages from the level of global ecosystem deployment. The U.S. business community is convinced that either China or the U.S. — whichever assumes the leading position — will be able to make the rules and thus have the ultimate advantage. The party lagging behind will sink in complete disadvantage. So the key to future success lies in the ability to finally dominate the global AI ecosystem. For this reason, the future U.S. foreign policy will emphasize the following two aspects:

• “Optimizing” measures for containing China.

The U.S. business community believes previous export controls on chips have already prevented the U.S. from getting a better position in global markets. They indirectly promoted the process of Chinese self-reliance regarding high-end chips and have fettered U.S. companies' expansion in overseas markets. Nvidia CEO Huang said in a recent interview that if China could replace Nvidia chips with Huawei chips and realize full-link autonomy in AI training, China may be able to carry out an all-around substitution of U.S. technologies globally, thereby undermining U.S. tech advantages.So U.S. measures against China should pursue “balance” and “practical effects,” rather than modeling the Chinese adage of single-mindedly “killing 1,000 enemies while losing 800 of one's own.” The U.S. needs to find an “optimum” equilibrium point between containing development and securing market share in China, as well as controlling the route of technology development.

• Accelerating a global AI ecosystem.

The head of the computing team at RAND's Center for Technology and Security Policy pointed out that the key to global industry dominance usually lies in the ability to promote technology at large-scale, rather than technological breakthroughs in the lab. Hence the U.S. must promote its AI technologies and services to more places in the world. There have been various signs the U.S. government has adopted this idea and is moving swiftly to grab market share globally. The Trump administration is carrying out “chips for capital”, engaging in investment and tariffs negotiations with allies through loosening chips sanctions, taking full advantage of U.S. technological advantages and policy tools to occupy global markets. It has so far reached agreements with Gulf states on chips export and data center construction.Besides the U.S. government, tech giants have also become part of the vanguard in expanding U.S. overseas markets. Trump's entourage of ultra-strong tech-giants on his recent trip to the Middle East was especially eye-catching. In addition, taking advantage of the Stargate initiative, tech giants are continuously expanding global markets for AI infrastructure construction. OpenAI, for one, following its Stargate UAE project, is planning to seek opportunities for further cooperation in such countries as South Korea, Singapore, Australia, India and Japan.