英伟达2026股东大会：AI工厂开始赚钱，黄仁勋给出下一轮产业答案

北京时间2026年6月25日凌晨，全球市值最高企业——英伟达召开年度股东大会。

这被视为AI行业从“技术讲故事”转向“商业算回报”的关键节点。

黄仁勋没有过多聊技术参数，而是讲了一套新的商业逻辑：客户买的不再是GPU硬件，而是一座能持续生产收益的“AI工厂”。

配合这一逻辑，会上披露了全年（2026财年（FY2026））2160亿美元营收、65%增速的成绩单，并官宣了季度股息提升约25倍、新增800亿美元回购的资本政策。

这让“AI工厂”的商业逻辑完成了一次至关重要的跨越——从战略叙事到真金白银的回报——并最终形成了闭环。

从卖芯片，到卖AI工厂

黄仁勋指出，计算产业正在经历60年来最深刻的一次变革。

过去60年的模式是“人写软件、计算机执行指令”，计算机只是工具。

而在AI时代，计算机能自主理解、推理、规划、调用工具并完成实际工作——它从“工具”，变成了“会使用工具的助手”。

对应到基础设施层面，数据中心的定位也发生了本质变化：

传统数据中心是“工具仓库”，用来存储和传输文件；

AI时代的数据中心是“AI工厂”，核心产出是Token（制造词元）。Token可以直接转化为代码、方案、设计、自动化服务，每一个Token都是可变现的利润单元。

黄仁勋在演讲中反复强调一句话：有用的AI已经到来，并且能够赚钱。这也是本轮算力需求爆发的底层支撑。

当AI能直接帮企业增收降本，算力采购就不再是“成本投入”，而是“产能投资”——客户买的不是服务器，而是一座能持续产生收益的工厂。

基于这一逻辑，黄仁勋给出了一个长期判断：AI工厂建设是一个数十年维度的长周期工程，可以对标电网、互联网等重大基础设施，甚至有望成为人类历史上规模最大的基建项目。需求不会短期见顶。

他同时将AI产业拆解为“五层蛋糕”架构：能源、芯片与系统、基础设施、模型、应用。

英伟达的定位，是为整个体系提供底层基础设施。其竞争核心不是硬件单价，而是“单位Token的产出成本与吞吐量”——也就是AI工厂的投产效率。

从单芯片到完整AI工厂的三代演进

黄仁勋将AI工厂的效率升级，拆解为三代算力平台的迭代。这不仅是技术路线图，更是一套精心设计的产业落地脉络——每一代产品都对应着一个特定的市场阶段，也隐含着英伟达对竞争壁垒的思考。

第一阶段：从0到1——Hopper架构

Hopper专为大模型预训练而生，完成了AI算力的商业化验证。它回答了“这条路能不能走通”的问题，是上一轮生成式AI爆发的核心底座。

第二阶段：从1到N——Blackwell架构

Blackwell是当前的主力产品，定位为AI工厂的“标准产能单元”，核心要回答的问题是“能不能规模化赚钱”。

黄仁勋引用了SemiAnalysis的第三方测试数据来佐证其效率：最低的单位Token成本，Token吞吐量是第二名的30倍；智能体场景下，能效较Hopper提升20倍。目前该产品已在超大规模云厂商、AI实验室、工业企业及主权国家客户中广泛部署。

但黄仁勋选择在这个节点密集打出第三方背书，释放出一个信号：英伟达正在从“技术领先”的话术，转向“商业回报可量化”的话术，以巩固客户对持续投入的信心。

第三阶段：为智能体时代重建系统——Vera Rubin平台

黄仁勋的判断是，智能体的计算模式与过去截然不同——它需要反复思考、调用工具、检索数据、执行代码，过程中CPU很容易成为瓶颈，让昂贵的GPU闲置。Vera Rubin的应对策略不是单芯片升级，而是一套完整的系统级方案：

Rubin是负责“思考”的GPU，Vera是从零为智能体设计的CPU，再搭配高速互联与全栈软件，实现从单卡到跨数据中心的全链路优化。

当单芯片性能竞争逐渐逼近物理极限，英伟达正把护城河从“一颗芯片”迁移到“一整座工厂”的系统整合能力上。一旦客户按这套封闭优化的架构搭建生产环境，迁移成本将大幅上升。

黄仁勋称Vera将开辟一个全新市场，是英伟达历史上最重要的产品之一——这也意味着，公司未来的营收结构很可能会发生显著变化，值得投资者重点跟踪。

公司治理与资本政策

本次股东大会完成了常规治理事项：10名董事全部连任，高管薪酬方案通过，控制权格局不变。唯一获批的股东提案是将超级多数投票条款改为简单多数，属于渐进式优化。

值得留意的是，英伟达全年（2026财年（FY2026））营收2160亿美元、同比增长65%，数据中心营收1940亿美元、占比近九成（1940/2160≈89.8%）。在这个底盘上，季度股息一次性上调25倍，新增800亿美元回购额度，并承诺每年将50%以上的自由现金流返还股东。

这里有一个值得标记的节奏参照：什么时候该把利润全部砸回研发和扩张，什么时候该启动系统化的股东回报——英伟达画出了一条分界线。对投资者而言，这意味着公司的估值逻辑正从纯“成长股”向“成长+现金回报复合型”迁移。

配合这套叙事，业务架构也做了简化：分为数据中心和边缘计算两大平台，共享同一套架构与软件栈。架构统一意味着生态锁定，这是英伟达护城河的关键一环。

以上内容根据本次股东大会现场音频整理而成，更多细节可精读英伟达2026股东大会中英文逐段对照完整实录。文末附2025年股东大会书面问答材料，可辅助理解英伟达长期经营思路。

Enjoy!

▌英伟达2026年度股东大会中英文对照版全文

Good morning, and welcome to NVDIA's 2026 Annual Meeting of Stockholders. I'm Toshiyuki Hara, Vice President of Investor Relations and Strategic Finance. We are hosting a fully virtual annual meeting again this year. To help the meeting run smoothly, please follow the rules of conduct for the meeting posted on our virtual meeting website.

各位股东，大家上午好。欢迎参加英伟达2026年度股东大会。我是投资者关系与战略财务副总裁Toshiyuki Hara。今年我们继续以全线上形式召开年度股东大会。为确保会议顺利进行，请遵守发布在我们虚拟会议网站上的会议行为准则。

Other members of NVDIA's management who are on the line with me today are Jensen Huang, President, Chief Executive Officer and Director; Colette Kress, Executive Vice President and Chief Financial Officer; and Timothy Teter, Executive Vice President, General Counsel and Secretary.

今天与我一同在线的管理层成员还包括：总裁、首席执行官兼董事黄仁勋（Jensen Huang）；执行副总裁兼首席财务官Colette Kress；以及执行副总裁、总法律顾问兼秘书Timothy Teter。

I would also like to introduce the outside members of our board who are in attendance: Frances Kunreuther, John O. Dabiri, Dawn Hudson, Harvey Jones, Melissa Lora, Steve Dowling, A. Brooke Sewell, Toshiyuki Hara and Mark Stevens.

接下来请允许我介绍出席本次会议的独立董事：Frances Kunreuther、John O. Dabiri、Dawn Hudson、Harvey Jones、Melissa Lora、Steve Dowling、A. Brooke Sewell、Toshiyuki Hara以及Mark Stevens。

Finally, I would like to introduce Sarah Barden from PricewaterhouseCoopers LLP, our independent registered public accounting firm. After the formal portion of the meeting and Jensen's business update, we will set aside some time for questions. I will now turn the meeting over to Tim.

最后，我要介绍来自普华永道会计师事务所（PricewaterhouseCoopers LLP）的Sarah Barden，他们是我们独立注册的公共审计机构。在会议正式议程和黄仁勋的业务更新结束后，我们将留出时间回答股东提问。现在我将会议交给Tim主持。

Thanks, Toshi. Good morning, and welcome to our 2026 Annual Meeting, which will now officially come to order. Jensen will serve as chairman, and I will serve as secretary and conduct the procedural portion of the meeting. First, a few housekeeping items.

感谢Toshi。各位早上好，欢迎参加我们的2026年度股东大会，现在正式开会。黄仁勋将担任本次会议主席，我将担任秘书并主持会议的议程环节。首先说明几项会议安排。

We have opened the online portal for stockholders to vote their shares, and I will make an announcement when the polls are about to be closed. Stockholders may ask questions during the meeting by entering them into the portal.

我们已为股东开放了线上投票通道，在投票即将截止时我会发出公告。股东可以在会议期间通过投票平台提交问题。

As stated in our rules of conduct, the meeting is not to be used as a forum to present views that are not directly related to our business, the business of the meeting, or for matters of individual concern. In fairness to all attendees, we will limit each stockholder to one question and may combine similar questions into a single response.

正如我们的会议行为准则所述，本次会议不得用于发表与公司业务或会议议题无关的观点，也不得用于表达个人关切事项。为公平对待所有参会者，每位股东限提一个问题，我们可能会将相似问题合并回答。

After the conclusion of the meeting, we will publish the pertinent questions and our responses on our Investor Relations website.

会议结束后，我们将把相关问题和回复发布在投资者关系网站上。

During this meeting, we may make forward-looking statements based on current expectations. These statements are subject to significant risks and uncertainties, and our actual results may differ materially.

在本次会议上，我们可能会基于当前预期作出前瞻性陈述。这些陈述面临重大风险与不确定性，实际结果可能存在重大差异。

For a discussion of factors that could affect our future financial results and business, please refer to the reports we file with the Securities and Exchange Commission, including our Annual Report on Form 10-K and our Quarterly Reports on Form 10-Q. All of our statements are made as of June 24th, 2026, based on information available to us as of today, and except as required by law, we assume no obligation to update them.

关于可能影响我们未来财务业绩和业务的因素讨论，请参阅我们向美国证券交易委员会提交的报告，包括10-K年度报告和10-Q季度报告。所有陈述均基于截至2026年6月24日的现有信息作出，除法律要求外，我们不承担更新义务。

We will address the matters described in our proxy statement dated May 12th, 2026.

我们将讨论2026年5月12日发布的委托声明书中所述事项。

Following Jensen's business update, the polls will be closed, we will announce preliminary voting results, and then the official portion of the meeting will be adjourned. I have a complete list of the stockholders of record of NVDIA's common stock on the April 27th, 2026 record date for this meeting. I also have an affidavit from Broadridge, certifying that they commenced the mailing of proxy materials on May 12th, 2026. I am appointing Chris Wood of American Election Services LLC to act as the inspector of elections at this meeting.

黄仁勋完成业务更新后，投票通道将关闭，我们将公布初步投票结果，随后会议正式议程即告休会。我持有截至2026年4月27日登记日的英伟达普通股完整股东名册。我还持有Broadridge出具的宣誓书，证明其于2026年5月12日开始寄送委托投票材料。我指定美国选举服务有限公司的Chris Wood担任本次会议的选举监察人。

He will tally the final votes when balloting on all matters is completed. Chris has taken the customary oath of office, and we will file this oath with the records of the meeting. The presence in person or by proxy of a majority of the shares entitled to vote at the meeting will constitute a quorum.

他将在所有事项投票完成后统计最终票数。Chris已宣誓就职，该誓词将与会议记录一并存档。有权在会议上投票的多数股份持有人亲自或通过委托出席，即构成法定人数。

Chris has informed me that a quorum of stockholders is present to conduct today's meeting based on proxies received. If you are eligible to vote and have not submitted your proxy, or if you want to change your vote, please vote online now. You do not need to vote if you have already sent in your signed proxy or voted online or by telephone; your votes will be counted automatically.

Chris已通知我，根据收到的委托投票，出席今天会议的股东已达到法定人数。如果您有投票资格但尚未提交委托，或者希望更改投票，请立即在线投票。如果您已寄回签署的委托书或已通过线上或电话投票，则无需再次投票，您的选票将被自动计入。

The time is 9:04 a.m. Pacific Time, and the polls are currently open for voting. There are seven items of business for this meeting: three management proposals and four stockholder proposals. The management proposals are as follows. Proposal 1: the election of ten nominees to serve as directors until our 2027 Annual Meeting. Proposal 2: the approval of the compensation of our named executive officers for fiscal 2026 as disclosed in our proxy statement. And Proposal 3: the ratification of PwC as our independent registered public accounting firm for fiscal 2027.

现在是太平洋时间上午9点04分，投票通道已开放。本次会议共有七项议程：三项管理层提案和四项股东提案。管理层提案如下：提案一，选举十名候选人担任公司董事，任期至2027年度股东大会。提案二，批准委托声明书中披露的2026财年指定高管薪酬方案。提案三，确认普华永道为2027财年独立注册公共审计机构。

The stockholder proposals are as follows. Proposal 4: a proposal to replace supermajority voting provisions in our charter and bylaws with a simple majority standard. Proposal 5: a proposal to evaluate and report on faith-based community resource groups. Proposal 6: a proposal to evaluate and report on civil rights and non-discrimination relating to DEI. And Proposal 7: a proposal to report and disclose greenhouse gas emissions from the use of the company's sold products.

股东提案如下：提案四，以简单多数表决标准取代公司章程和细则中的超级多数表决条款。提案五，评估并报告基于信仰的社区资源小组相关情况。提案六，评估并报告与DEI（多元、公平与包容）相关的公民权利及非歧视事项。提案七，报告并披露公司售出产品使用过程中产生的温室气体排放。

The board recommends a vote for each of the three management proposals and against the four stockholder proposals. The proxy statement provides the company's responses to the stockholder proposals, and explains why the board recommends a vote against each stockholder proposal.

董事会建议对三项管理层提案均投赞成票，对四项股东提案均投反对票。委托声明书中载有公司对各项股东提案的回应，并解释了董事会建议否决每项股东提案的理由。

We'll now ask each stockholder proponent to introduce their proposal. Each proponent will have three minutes to present. At this time, we would like to introduce John Cheveden to present Proposal 4. Mr. Cheveden, you have three minutes. Please begin.

现在我们请每项股东提案的提案人介绍其提案内容。每位提案人有三分钟陈述时间。首先有请John Cheveden介绍提案四。Cheveden先生，您有三分钟时间，请开始。

Hello. This is John Cheveden. Proposal 4: Majority Vote Standard — the proposal that won 99% NVDIA shareholder support at the 2025 NVDIA Annual Shareholder Meeting.

大家好，我是John Cheveden。提案四：多数表决标准——该提案在2025年英伟达年度股东大会上获得了99%的股东支持。

Shareholders request the board of directors take each step necessary so that each voting requirement, now in the charter and bylaws, that had an explicit or implicit default to state law calling for a greater than simple majority vote, be replaced by a requirement for a majority of the votes cast for and against proposals, or a simple majority.

股东要求董事会采取一切必要措施，将公司章程和细则中所有默许适用州法律中高于简单多数表决要求的条款，替换为以赞成票和反对票的简单多数为准的表决要求。

Shareholders are willing to pay a premium for shares of companies that have excellent corporate governance. The supermajority voting requirements, like those in NVDIA, have been found to be one of six entrenching mechanisms that are negatively related to company performance, according to What Matters in Corporate Governance by Lucian Bebchuk of Harvard Law School.

股东愿意为具有优秀公司治理水平的公司支付溢价。根据哈佛法学院Lucian Bebchuk教授所著《公司治理的关键要素》一书的研究，英伟达现行的超级多数表决要求已被发现是六种与管理层固化相关的负面机制之一，对公司业绩有消极影响。

Supermajority requirements can be used to block proposals supported by most shareholders but opposed by management. The NVDIA board of directors deserves to be harshly criticized for not supporting this proposal. And the chair of the NVDIA Governance Committee, Mr. A. Brooke Sewell, serves against votes for his lack of support for this proposal that won 99% shareholder support.

超级多数表决要求可被用来阻止大多数股东支持但管理层反对的提案。英伟达董事会不支持这项提案，理应受到严厉批评。英伟达治理委员会主席A. Brooke Sewell先生因缺乏对这项获得99%股东支持提案的支持，应被投反对票。

This proposal first received outstanding 91% support at the 2024 NVDIA Annual Shareholder Meeting as a non-binding shareholder proposal. In response to the overwhelming 91% vote, Nvidia put this proposal on the 2025 NVDIA Annual Meeting ballot as a binding Nvidia proposal.

这项提案最初在2024年英伟达年度股东大会上作为非约束性股东提案获得了91%的杰出支持率。面对91%的压倒性投票结果，英伟达将这项提案作为约束性提案列入2025年年度股东大会投票议程。

NVDIA shareholders then gave 99% support to the majority proposal. However, this 99% support translated into 65% based on all

NVDIA shares outstanding.

英伟达股东随后以99%的支持率通过了这项多数表决提案。然而，这99%的支持率按英伟达全部流通股计算仅相当于65%。

Since so many individual shares did not vote, and this proposal needed 67% support based on all shares outstanding, whether or not NVDIA shares voted — the first mistake by the NVDIA Corporate Governance Committee, chaired by Mr. Sewell, was to not make a slight extra effort to get more shareholders to vote in 2025, since the incoming votes were 99% in favor in 2025.

由于大量个人股东未参与投票，而该提案需要基于全部流通股的67%支持率才能通过——由Sewell先生领导的英伟达治理委员会犯下的第一个错误，就是没有在2025年额外付出一些努力来动员更多股东投票，毕竟2025年的已统计选票中99%都是赞成的。

NVDIA apparently, as the company, did not care whether its shareholders voted or not, in order for Nvidia to defeat its own proposal that was getting 99% shareholder support. Now the absurd Nvidia position is to not support this proposal, which received 99% support at the 2025 Annual Meeting. Nvidia is throwing water in the face of shareholders by not supporting a proposal that received 99% support at the 2025 Nvidia Annual Meeting. And Mr. Sewell does not deserve to be the Nvidia Lead Director. Please vote for the proposal that received 99% support at the 2025 Nvidia Annual Meeting: Majority Vote Standard, Proposal 4.

英伟达作为公司，显然不在乎股东是否投票，以便让公司自己的提案——一项获得99%股东支持的提案——被否决。如今英伟达的荒谬立场是：不再支持这项在2025年度股东大会上获得99%支持率的提案。英伟达这是在当面泼股东冷水，不支持一项刚刚在2025年获得99%股东支持的提案。Sewell先生不配继续担任英伟达首席独立董事。请投票支持在2025年英伟达年度股东大会上获得99%支持率的提案：多数表决标准，提案四。

Thank you. At this time, we would like to introduce Inspire Investing, which has provided a prerecorded statement presenting Proposal 5, presented by Tim Schwarzenberger, an authorized representative of the proponent. Operator, please begin the statement.

感谢。现在请播放Inspire Investing预先录制的提案五陈述视频，由提案人授权代表Tim Schwarzenberger发言。请播放。

Hello, my name is Tim Schwarzenberger. I'm a portfolio manager and the Director of Corporate Engagement at Inspire Investing. Inspire is the largest provider of Christian ETFs and represents William Cunningham, the filer of Proposal 5.

大家好，我是Tim Schwarzenberger，担任Inspire Investing的投资组合经理兼企业参与事务总监。Inspire是最大的基督教ETF提供商，代表提案五的提交人William Cunningham。

Companies from American Express to Apple allow religious employees to form employee resource groups or community resource groups. Why? Because employees of faith are entitled to the same respect that other employees enjoy. They should be able to bring their whole selves to the company they work for. And Nvidia, one of the most innovative and ambitious companies in the world today, is no exception.

从美国运通到苹果公司，许多企业都允许有宗教信仰的员工组建员工资源小组或社区资源小组。为什么？因为有信仰的员工理应享有与其他员工同等的尊重，他们应当能够完整地做自己，投身于所服务的企业。而英伟达，当今天下最具创新精神和雄心的公司之一，同样不应例外。

Nvidia CEO Jensen Huang has said that Nvidia aims to, quote, "be empathetic to the experience of all" and, quote, "act to make Nvidia a place of opportunities." That commitment to equality and opportunity is an excellent goal. But if everyone truly belongs, then religious belief should be part of that vision. Nvidia has community resource groups for employees based on gender, race, and veteran status, but not faith.

英伟达首席执行官黄仁勋曾表示，英伟达的目标是"体恤所有人的经历"，并"努力将英伟达打造成一个充满机遇的地方"。这种对平等与机遇的承诺是一个卓越的目标。但如果每个人都真正有归属感，那么宗教信仰也应纳入这一愿景。英伟达已设有基于性别、种族和退伍军人身份的员工社区资源小组，却没有基于信仰的小组。

In its opposition statement to this proposal, Nvidia notes that the company does not tolerate harassment or discrimination based on an employee's religion. But this response, essentially restating the basic standard set forth for workplaces under Title VII of the Civil Rights Act, is strange for a company that refers to other types of diversity as crucial to Nvidia's mission of opportunity.

英伟达在对该提案的反对声明中表示，公司不容忍基于员工宗教信仰的骚扰或歧视。但这一回应本质上只是重申了《民权法案》第七章对工作场所的基本要求，对于一家将其他类型的多元化视为实现机遇使命之关键的公司而言，这样的回应显得奇怪。

The company has seen fit to extend resources to groups organized around race and gender. But when it comes to people of faith, who make up the vast majority of American employees, they have to be content with baseline protections against harassment and discrimination. This is not about promoting any particular viewpoint. It is about neutrality. When companies allow some forms of identity-based organization but not others, they risk creating the appearance of ideological favoritism.

公司认为有必要向围绕种族和性别组建的团体提供资源支持，但对于占美国员工绝大多数的有信仰人群，却只满足于提供反骚扰和反歧视的底线保护。这不是在推崇某种特定观点，而是在追求中立。当公司允许某些基于身份的组织形式存在，却不允许另一些时，就有制造意识形态偏袒之嫌的风险。

That is not a sound position for any company, either culturally or from a governance perspective. If Nvidia aspires to be a place of opportunities, religious employees shouldn't be excluded from opportunities to organize in the workplace simply because they're religious. And Nvidia needs to ensure that its policies aren't biased against employees of faith, and allowing faith-based community resource groups is a great place to start. Thank you.

无论文化层面还是治理层面，这都不是任何公司应有的立场。如果英伟达志在成为充满机遇的地方，就不应仅仅因为员工有宗教信仰而将其排除在组织活动的机会之外。英伟达需要确保其政策不会对有信仰的员工构成偏见，而允许建立基于信仰的社区资源小组正是一个很好的起点。谢谢。

Thank you. At this time, we would like to introduce the American Conservative Values ETF, which has provided a prerecorded statement presenting Proposal 6, presented by Ridgeline Research as authorized representative of the proponent. Operator, please begin the statement.

感谢。现在请播放美国保守价值观ETF预先录制的提案六陈述视频，由提案人授权代表Ridgeline Research发言。请播放。

Good morning, shareholders and board members. My name is William Flake, founder and CEO of Ridgeline Research, the investment advisor to the American Conservative Values ETF, ticker symbol ACVF. Speaking on behalf of the shareholders, I'm here to present Proposal No. 6, requesting a report on how Nvidia's hiring and training policies affect employees and applicants across protected classes, particularly religious and political views, and the risk they may create for the company. This proposal is fundamentally about protecting shareholder value.

各位股东、各位董事会成员，大家早上好。我是William Flake，Ridgeline Research的创始人兼CEO，该公司是美国保守价值观ETF（股票代码：ACVF）的投资顾问。我代表股东在此提出第六号提案，要求公司就其招聘和培训政策如何影响各受保护类别的员工和求职者——特别是涉及宗教信仰和政治观点方面——以及这些政策可能给公司带来的风险提交报告。这项提案的核心在于保护股东价值。

Nvidia's workforce policies directly shape how employees are hired, trained and promoted, and whether the company is fostering a high-performance culture grounded in American values or unintentionally introducing internal division and undermining cohesion and performance. Those policies are intended to support as well as creating legal and reputational risk.

英伟达的员工政策直接决定了员工的招聘、培训和晋升方式，也决定了公司是在培育扎根于美国价值观的高绩效文化，还是在不经意间制造内部对立、削弱团队凝聚力和执行力。这些政策在意图支持员工的同时，也可能带来法律和声誉风险。

Today, many companies, Nvidia included, have adopted hiring and training frameworks that emphasize identity-based metrics, commonly referred to as DEI — Diversity, Equity and Inclusion. These approaches divide employees into categories and carry consequences. Research has found these frameworks can increase bias, hostility and discrimination rather than reduce them. And at the same time, surveys show a majority of employees fear reprisal for expressing their religious or political views. These are not abstract concerns. They affect who applies to Nvidia, who stays, and whether employees feel comfortable contributing fully, creating clear operational, reputational and legal risk. And the risk is increasing.

如今，包括英伟达在内的许多公司都采用了强调身份指标的招聘和培训框架，通常被称为DEI——多元、公平与包容。这些方法将员工按类别划分，并带来相应后果。研究发现，这些框架可能加剧偏见、敌意和歧视，而非减少它们。与此同时，调查显示，大多数员工因担心遭到报复而不敢表达自己的宗教或政治观点。这些不是抽象的忧虑，而是切实影响着谁会选择应聘英伟达、谁会留下来、员工是否能够安心地充分贡献才华，从而带来明确的运营风险、声誉风险和法律风险。而且这些风险正在加剧。

Changes in federal enforcement priorities under the Trump administration, including executive action and agency guidance targeting potentially unlawful DEI policies, have heightened scrutiny of employment policies tied to protected characteristics, increasing the likelihood of investigation, litigation and regulatory exposure. Management argues this report is unnecessary because Nvidia complies with the law and maintains strong oversight.

特朗普政府时期联邦执法重点的变化，包括针对可能违法的DEI政策的行政行动和机构指导，加大了对与受保护特征挂钩的就业政策的审查力度，增加了调查、诉讼和监管风险的可能性。管理层辩称这份报告没有必要，因为英伟达遵守法律并维持着强有力的监督。

But their response misses the point. The issue is not whether policies exist — they do. Effective oversight of misguided policies does not eliminate risk. It may actually create risk, because risks that are not fully understood cannot be effectively managed, and risks that are not managed can have real consequences for shareholders. And shareholders currently have no visibility into the impact of those policies.

但他们的回应没有抓住重点。问题不在于政策是否存在——它们确实存在。对错误政策的有效监督并不能消除风险，反而可能制造风险，因为未被充分理解的风险无法得到有效管理，而未被管理的风险可能对股东造成实质性后果。目前股东对这些政策的影响毫无知情权。

This proposal asks the board to conduct a focused evaluation and provide a report on how Nvidia's hiring and training policies function in practice, and to assess the operational and legal risks that may arise. It is a request for transparency. For these reasons, I urge your support for Proposal No. 6.

该提案要求董事会进行专项评估，并就英伟达招聘和培训政策在实际中的运作情况提交报告，同时评估可能产生的运营和法律风险。这是一项对透明度的请求。基于以上原因，我恳请各位支持第六号提案。

Thank you. At this time, we would like to introduce Green Century Capital Management and Mercy Investment Services, which have provided a prerecorded statement presenting Proposal 7. Operator, please begin the statement.

感谢。现在请播放Green Century Capital Management和Mercy Investment Services预先录制的提案七陈述视频。请播放。

My name is Giovanna Eichner. On behalf of Green Century Capital Management and Mercy Investment Services, I am presenting Proposal No. 7, asking Nvidia to disclose greenhouse gas emissions from the use of its sold products.

我是Giovanna Eichner。我代表Green Century Capital Management和Mercy Investment Services，提出第七号提案，要求英伟达披露其售出产品在使用过程中产生的温室气体排放。

A warming climate has direct financial consequences for the broader economy and individual semiconductor companies. U.S. emissions have caused over $10 trillion in global economic damages since 1990. One third of global semiconductor production will be reliant on materials threatened by climate disruption by 2035. At the same time, data centers' energy consumption is projected to grow nearly 15% per year from 2024 to 2030 and could emit up to an additional 44 million metric tons of carbon dioxide.

气候变暖对整体经济和半导体企业都带来直接的财务影响。自1990年以来，美国的碳排放已造成超过10万亿美元的全球经济损害。到2035年，全球三分之一的半导体生产将依赖于受气候扰动威胁的原材料。与此同时，数据中心的能耗预计从2024年到2030年将以每年近15%的速度增长，可能额外排放多达4400万吨二氧化碳。

Nvidia recognizes the need to reduce emissions in its 2030 target to decrease the emissions intensity from use of its sold products. Nevertheless, Nvidia does not report data on its use-of-sold-product emissions. Investors are thus unable to assess if Nvidia has plans in place to progress towards its 2030 target and adequately respond to climate-related risk. By continuing to lag peer performance and investor expectations, the company exposes itself to competitive and reputational risk. Nvidia faces competitive risk by failing to keep pace with peers, as disclosures from Advanced Micro Devices, Intel and Hewlett Packard Enterprise have long disclosed emissions from the use of sold products.

英伟达在其2030年目标中承认了减少排放的必要性，目标是降低售出产品使用过程中的排放强度。然而，英伟达并未报告其售出产品使用排放的数据。因此，投资者无法评估英伟达是否已制定计划推进其2030年目标，并充分应对气候相关风险。由于持续落后于同行表现和投资者预期，公司面临竞争和声誉风险。英伟达面临竞争风险，因为未能跟上同行步伐——AMD（超威半导体）、英特尔和慧与科技（HPE）早已披露了售出产品使用过程中的排放数据。

Nvidia's major customers, including Amazon, Google, Meta and Oracle, have climate commitments yet report increasing emissions. A 2025 survey from the Massachusetts Institute of Technology found that 70% of companies include sustainability criteria in supplier selection, and over half request emissions data. Without data on a significant emissions source, Nvidia cannot demonstrate progress and may lose market share to competitors that better report and address their greenhouse gas emissions. Even more, Nvidia's lack of action may push customers to develop proprietary technology that gives greater control and visibility into a company's energy use and emissions — a risk Nvidia acknowledges by citing Alphabet, Amazon and Microsoft as competitors in its 10-K.

英伟达的主要客户，包括亚马逊、谷歌、Meta和甲骨文，均有气候承诺，但报告的排放量却在上升。麻省理工学院2025年的一项调查发现，70%的企业在供应商选择中纳入了可持续性标准，超过一半的企业要求提供排放数据。缺乏对这一重要排放源的数据，英伟达无法展示其进展，可能将市场份额输给在温室气体排放报告和治理方面做得更好的竞争对手。更有甚者，英伟达的不作为可能推动客户开发专有技术，以更好地掌控和洞察企业的能源使用和排放——英伟达在10-K报告中已将Alphabet、亚马逊和微软列为竞争对手，这本身就承认了这一风险。

Failure to report a substantial portion of its emissions also heightens reputational risk. In 2026, the Pew Research Center reported that more Americans say data centers have a negative effect on the environment than a positive one. And local pushback canceled $41.7 billion in data center projects in the first three months of 2026. As scrutiny intensifies and projects are interrupted, comprehensive emissions disclosures would demonstrate accountability and help investors assess and manage risks to Nvidia's profitability.

未能报告其大部分排放数据也加剧了声誉风险。2026年，皮尤研究中心报告称，越来越多的美国人认为数据中心对环境的影响是负面的而非正面的。2026年前三个月，因当地反对而取消的数据中心项目金额高达417亿美元。随着审查力度加大和项目中断，全面的排放披露将体现问责意识，帮助投资者评估和管理英伟达盈利能力面临的风险。

Projected long-term efficiency gains from AI neglect the consequences of rapidly increasing emissions on Nvidia's business and reputation in the near term and its current contribution to long-term systemic risk. By continuing to delay disclosure of material emissions, the company cannot demonstrate progress toward managing the financial risks of climate change and thus falls short of investor expectations. We therefore urge a vote for Proposal No. 7.

人工智能带来的长期效率提升预期，忽视了排放量快速增长对英伟达近期业务和声誉的影响，以及其对长期系统性风险的当前贡献。由于持续推迟披露重大排放数据，公司无法展示其在管理气候变化财务风险方面的进展，从而辜负了投资者的期望。因此，我们恳请投票支持第七号提案。

Thank you. As the company has not received notice from any of its stockholders of any other matter to be considered at today's meeting, no other proposals will be addressed. If you have not voted and intend to vote, or if you want to change your vote, please do so online now. Proxies, votes, or any changes or revocations submitted after the closing of the polls will not be accepted. I'll turn it over to Jensen for a business update, following which we will announce the preliminary results of the meeting, if available, and then answer stockholder questions.

感谢。由于公司未收到任何股东关于本次会议需审议的其他事项的通知，不再处理其他提案。如果您尚未投票但有意投票，或者希望更改投票，请立即在线操作。投票通道关闭后提交的委托书、选票或任何变更和撤销将不予受理。现在交由黄仁勋进行业务更新，之后我们将公布初步投票结果（如已统计），并回答股东提问。

This has been an extraordinary year for Nvidia and for computing. Every ten to fifteen years, the computer industry resets: mainframe to PC, the internet to cloud, to mobile cloud. This reset is bigger. For 60 years, humans wrote software and computers executed instructions. That paradigm has changed. With AI, computers can understand, reason, plan, use tools and do useful work. A computer is no longer just a tool in the age of AI. It is an assistant that uses tools. And by extension, the data center is no longer a tool shed. It is an AI factory of digital assistants and infrastructure for producing digital intelligence. Nvidia is building the computing infrastructure for this new era.

这对英伟达和整个计算行业而言，都是非凡的一年。每隔十到十五年，计算产业就会经历一次重塑：从大型机到个人电脑，从互联网到云计算，再到移动云。而这一次的重塑更为深远。60年来，人类编写软件，计算机执行指令。这一范式已经改变。有了人工智能，计算机能够理解、推理、规划、使用工具并完成有意义的工作。在AI时代，计算机不再仅仅是工具，而是会使用工具的助手。由此延伸，数据中心也不再是存放工具的仓库，而是生产数字智能的AI工厂，是数字助手的基础设施。英伟达正在为这个新时代构建计算基础设施。

Two years ago, generative AI captured the world's attention. ChatGPT could write, draw, summarize and answer questions. Then, reasoning AI learned to think through problems. Agentic AI has now arrived. Agents can use tools, access memory, write code, call other agents, test results, and keep working until the job is done. Software coding is the first major enterprise breakout use case. This is important. AI is now useful. When AI does useful work, tokens become valuable. And when tokens become profitable, demand for compute accelerates.

两年前，生成式AI吸引了全世界的目光。ChatGPT能够写作、绘画、总结和回答问题。随后，推理式AI学会了深入思考问题。如今，智能体式AI（Agentic AI）已经到来。智能体能够使用工具、调用记忆、编写代码、调用其他智能体、检验结果，并持续工作直至任务完成。软件编程是第一个重大的企业级突破性应用场景。这一点至关重要——AI现在真正有用了。当AI能做有用的工作时，Token就变得有价值。当Token变得有利可图时，对算力的需求就会加速增长。

Look at the evidence. GitHub developers merged 300 million pull requests in 2023, 400 million in 2024, and 500 million in 2025 — a clear, steady climb. But in the first few months of 2026, the pace nearly tripled.

请看数据。GitHub上的开发者在2023年合并了3亿次Pull Request，2024年达到4亿次，2025年达到5亿次——呈现出清晰而稳定的增长趋势。但在2026年的头几个月，这一增速几乎翻了三倍。

Thirty million software developers earn roughly $3 trillion in salary each year, and their work underpins $100 trillion of global economic activity. With agents, that same workforce is now generating nearly $9 trillion of output — $6 trillion more. Coding is only one of the demand drivers. Useful AI has arrived. The ROI question of AI has been answered, and every industry is racing to adopt agentic AI. Nvidia revenue grew 65% to $216 billion. Operating income rose 60% to $130 billion. Diluted EPS increased 67% to four dollars and ninety cents. We generated $103 billion in operating cash flow and returned $41 billion to shareholders. Data center revenue of $194 billion increased 68%.

全球3000万名软件开发者每年薪酬总计约3万亿美元，他们的工作支撑着100万亿美元的全球经济活动。有了AI智能体，同样的劳动力现在创造了近9万亿美元的产出——增加了6万亿美元。编程只是需求驱动力之一。有用的AI已经到来。AI的投资回报率问题已经有了答案，每个行业都在争相采用智能体式AI。英伟达营收增长65%，达到2160亿美元；营业利润增长60%，达到1300亿美元；摊薄每股收益增长67%，达到4.90美元。我们产生了1030亿美元的经营性现金流，并向股东返还了410亿美元。数据中心营收达1940亿美元，增长68%。

Blackwell has significantly broadened Nvidia's infrastructure footprint across a diverse set of hyperscalers, cloud AI labs, industrial enterprise and sovereign customers. Model builders and hyperscalers have each cumulatively deployed hundreds of thousands of Blackwell GPUs. AI factory buildouts continue to rapidly expand across major industries. Companies like Capital One, Hyundai Motor Group, Jane Street and Eli Lilly are scaling Nvidia infrastructure to deploy AI. International revenue more than tripled to over $30 billion. Nearly 40 countries representing $50 trillion in GDP are building AI factories powered by Nvidia infrastructure. AI infrastructure is no longer experimental — it is in production.

Blackwell架构显著扩大了英伟达在各类超大规模云服务商、云AI实验室、工业企业和主权客户中的基础设施覆盖范围。模型构建商和超大规模云服务商各自累计部署了数十万块Blackwell GPU。AI工厂的建设在各大行业中持续快速扩展。Capital One、现代汽车集团、Jane Street和礼来制药（Eli Lilly）等企业正在扩大英伟达基础设施的部署以落地AI应用。国际营收增长超过两倍，突破300亿美元。近40个国家，GDP合计达50万亿美元，正在建设由英伟达基础设施驱动的AI工厂。AI基础设施已不再是实验性质——它已进入大规模生产部署。

AI is more than a model. It's a new industry. You can think of it as a five-layer cake: energy, chips and systems, infrastructure, models, and applications. Traditional data centers store and serve files. AI factories manufacture tokens — becoming code, answers, designs, actions, and services. Useful AI is profitable. Every token is a unit of profit. And that is why demand for compute is extraordinary. Customers are not buying computers; they're building revenue-generating AI factories. The architecture of the factory matters. The question is how much revenue the factory can produce and at what cost. Inference is the process of generating tokens, and Nvidia Blackwell sets the standard in SemiAnalysis inference benchmarks.

AI不仅仅是一个模型，它是一个全新的产业。你可以把它想象成一个五层蛋糕：能源、芯片与系统、基础设施、模型和应用。传统数据中心存储和传输文件，而AI工厂制造Token——转化为代码、答案、设计、行动和服务。有用的AI是有利可图的。每一个Token都是一个利润单元。这就是为什么算力需求如此旺盛。客户购买的不是计算机，而是正在建造能创造收入的AI工厂。工厂的架构至关重要。核心问题是：工厂能创造多少收入，成本又是多少。推理是生成Token的过程，而英伟达Blackwell在SemiAnalysis推理基准测试中树立了行业标准。

Blackwell is recognized as the inference king, delivering the lowest cost per token and 30 times higher token throughput than the next best platform. This is why architecture is critical. Nvidia systems may not be the cheapest to produce or to purchase, but Nvidia generates the lowest cost tokens, the highest token throughput, and the most revenue.

Blackwell被公认为"推理之王"，实现了最低的每Token成本和比第二名高出30倍的Token吞吐量。这就是架构为何如此关键。英伟达系统的生产和采购成本未必最低，但英伟达能生成成本最低的Token、最高吞吐量的Token，以及最多的收入。

Vera Rubin is the next step. Hopper was built for pre-training. Blackwell brought inference to rack scale. Vera Rubin is built for agents. Agentic AI changes the computing pattern. Agents think, use tools, access databases, retrieve memory, execute code and call applications, repeatedly, until the task is done. Large language models on GPUs drive the thinking, and the CPUs must keep up. If the CPU becomes a bottleneck, the GPUs sit idle. And in an AI factory, idle GPUs are lost revenue. That is why Vera matters.

Vera Rubin是下一步。Hopper为预训练而生，Blackwell将推理扩展到机架级，而Vera Rubin专为智能体打造。智能体式AI改变了计算模式。智能体思考、使用工具、访问数据库、调取记忆、执行代码、调用应用程序，反复循环直至任务完成。GPU上的大语言模型驱动思考，CPU必须跟上节奏。如果CPU成为瓶颈，GPU就会闲置。而在AI工厂中，闲置的GPU意味着收入的损失。这就是Vera的意义所在。

Vera is the CPU for agents. Rubin is the GPU for thinking. NVLink, Spectrum X storage and security on BlueField, and software bring the systems together. Nvidia, in fact, is the only company with three networking businesses. NVLink scale-up connects the GPUs in a rack into one giant computer. Spectrum X is the Ethernet built for AI. It scales out and across AI factories. And it is now larger than all other Ethernet networking peers combined. InfiniBand provides the lowest latency network for the world's largest AI and scientific systems. Together, they enable optimization of the AI factory from the GPU to the rack, across the data center, and across data centers.

Vera是面向智能体的CPU，Rubin是用于思考的GPU。NVLink、Spectrum X、BlueField上的存储与安全，以及软件，将这些系统整合在一起。事实上，英伟达是唯一拥有三大网络业务的公司。NVLink纵向扩展将机架内的GPU连接成一台巨型计算机；Spectrum X是为AI而生的以太网，横向扩展贯穿AI工厂，其规模现已超过所有其他以太网同行的总和；InfiniBand为全球最大的AI和科学计算系统提供最低延迟的网络。三者协同，实现从GPU到机架、从数据中心内部到跨数据中心的AI工厂全面优化。

Vera Rubin is not a chip. It is an AI factory platform, and the ecosystem is already moving. Vera Rubin is in full production, and every major model builder, public cloud, AI cloud and hyperscaler is preparing to build on it. Vera opens an entirely new market. Until now, every CPU was built for people. We live in a world measured in seconds. Agents live in a world measured in nanoseconds.

Vera Rubin不是一颗芯片，它是一个AI工厂平台，生态系统已经在运转。Vera Rubin已全面量产，每一个主要的模型构建商、公有云、AI云和超大规模云服务商都在准备基于它进行建设。Vera开辟了一个全新的市场。迄今为止，每一款CPU都是为人设计的。我们生活在一个以秒计时的世界里，而智能体生活在一个以纳秒计时的世界里。

Every moment a CPU keeps an agent waiting is a moment the most expensive thing in the building — the GPUs — sits idle. So we built a brand-new CPU from the ground up for agents. This is a new market. CPUs for humans were sliced and rented per core. Agents don't rent cores; they demand ultra-fast responses. There will be billions of agents, and they will need a CPU built for them. We believe Vera will be one of the most significant product launches in our company's history, and the orders are already coming in.

CPU让智能体多等一刻，就意味着数据中心里最昂贵的设备——GPU——多闲置一刻。因此，我们从零开始为智能体打造了一款全新的CPU。这是一个全新的市场。面向人类的CPU按核心数切分和租用，而智能体不租用核心，它们要求超快响应。未来将有数十亿个智能体，它们需要一款专为它们而生的CPU。我们相信Vera将成为我们公司历史上最重要的产品发布之一，订单已经在涌入。

CUDA is one of the most important investments we have ever made. For 20 years, we've been dedicated to one architecture for accelerated computing. The install base attracts developers, developers create breakthrough applications, breakthrough applications create new markets, new markets expand the install base. That flywheel is accelerating. CUDA-X is the library stack built on top of CUDA. These libraries are the crown jewels of Nvidia. They solve some of the hardest problems in science and industry: computational lithography, optimization, genomics, physics, data processing, robotics, AI, radio access networks and more. Now these libraries are becoming tools for agents.

CUDA是我们有史以来最重要的投资之一。20年来，我们一直专注于加速计算的统一架构。装机量吸引开发者，开发者创造突破性应用，突破性应用开辟新市场，新市场扩大装机量。这个飞轮正在加速转动。CUDA-X是构建在CUDA之上的库集合。这些库是英伟达的皇冠明珠，它们解决了科学和工业领域一些最棘手的问题：计算光刻、优化、基因组学、物理学、数据处理、机器人、AI、无线接入网络等等。现在，这些库正在成为智能体的工具。

This week, we announced BioNeMo, a suite of digital biology and drug discovery tools for agents. Nvidia has vertically integrated and horizontally opened. We build the full stack so that we can optimize the system end to end, then we open it so the whole industry can build on it.

本周，我们发布了BioNeMo——一套面向智能体的数字生物学和药物发现工具。英伟达的做法是纵向整合、横向开放。我们构建完整的技术栈，以便端到端地优化系统，然后将其开放，让整个行业都能在此基础上进行开发。

Our business is broad and becoming more diverse. To make it easier to understand our business, we now describe Nvidia through two market platforms: data center and edge computing. In data center, we serve two markets: hyperscale and what we call AIC — AI Cloud — and industrial and enterprise. These two markets. Our customer base is diverse and growing. Edge computing includes PCs, workstations, gaming, AI base stations, robotics and automotive. The reason we can serve all these markets is because we have one architecture, one software stack and a rich ecosystem.

我们的业务广泛且日趋多元化。为了便于理解我们的业务，我们现在将英伟达划分为两大市场平台：数据中心和边缘计算。在数据中心领域，我们服务两大市场：超大规模和AIC（AI云），以及工业和企业市场。我们的客户群体多元且持续增长。边缘计算涵盖PC、工作站、游戏、AI基站、机器人和汽车。我们能够服务所有这些市场的原因在于，我们拥有统一的架构、统一的软件栈和丰富的生态系统。

Physical AI is the next wave of growth for Nvidia. Physical AI is agentic AI in the real world. Robots, cars and factories will perceive, reason, plan and operate in dynamic environments. Nvidia pioneers this field and builds the full loop: AI factories train models, Omniverse simulates them in virtual worlds, Nvidia Jetson computers operate in the robot, and Cosmos is the world foundation model that powers it all. Robots and robotic systems are being built across industries, from transportation, manufacturing, surgical robotics to hospitality service.

物理AI是英伟达的下一个增长浪潮。物理AI就是现实世界中的智能体AI。机器人、汽车和工厂将在动态环境中感知、推理、规划和运作。英伟达是这一领域的先驱，构建了完整的闭环：AI工厂训练模型，Omniverse在虚拟世界中进行仿真，英伟达Jetson计算机在机器人中运行，Cosmos是驱动这一切的世界基础模型。机器人和机器人系统正在各行各业中被构建，涵盖交通运输、制造业、手术机器人到酒店服务等领域。

The last several months have seen tremendous acceleration in AI. Let me leave you with this: Useful AI has arrived and it is profitable. And so compute is revenue-barrel. Vera Rubin is in full production. Vera is opening a brand-new CPU market built for agents, and every enterprise is becoming an agent company, and they run on Nvidia. The age of AI is moving at full steam, and Nvidia is building the infrastructure to power it.

过去几个月，AI的发展经历了巨大的加速。最后我想说的是：有用的AI已经到来，而且它是有利可图的。因此，算力就是收入的引擎。Vera Rubin已全面量产。Vera正在开辟一个专为智能体打造的全新的CPU市场，每一家企业都在成为智能体企业，而它们都运行在英伟达之上。AI时代正全速推进，英伟达正在构建驱动这一切的基础设施。

As we grow, we will continue to scale R&D, invest in our ecosystem and return capital to shareholders. We recently announced a significant increase in our dividend and stock buyback program. We plan to return 50% or more of our free cash flow to shareholders this year, next year and beyond. Thank you.

随着公司成长，我们将继续扩大研发投入、投资生态系统并向股东返还资本。我们近期宣布了大幅增加股息和股票回购计划。我们计划今年、明年及未来每年将50%以上的自由现金流返还给股东。谢谢。

Thanks, Jensen. The time is 9:00 a.m. Pacific Time and the polls are now closed. Based on preliminary voting results, the inspector of elections has confirmed that all ten nominees have received the requisite number of votes to be reelected as directors of the company, and Proposals 2 and 3 have been approved. Proposal 4, to replace supermajority voting provisions in our charter and bylaws with a simple majority standard, has been approved. Proposals 5, 6 and 7 have not been approved. The final vote results are not yet available to us but will be published in a Form 8-K, which we expect to file with the SEC within four business days. That concludes the formal portion of today's annual meeting, and I now declare the official portion of the meeting adjourned. I'll turn it over to Toshi for stockholder questions.

感谢黄仁勋。现在是太平洋时间上午9点整，投票通道已关闭。根据初步投票结果，选举监察人确认全部十名候选人均已获得连任公司董事所需的票数，提案二和提案三获得通过。提案四——以简单多数表决标准取代公司章程和细则中的超级多数表决条款——获得通过。提案五、六和七未获通过。最终投票结果尚未出炉，将以8-K表格形式发布，预计将在四个工作日内向美国证券交易委员会提交。今天年度股东大会的正式议程到此结束，我现在宣布会议正式休会。下面交给Toshi主持股东问答环节。

Thanks, Tim. We'll now move to Q&A. We group together duplicative questions that we received to streamline the discussion. Question 1: How sustainable is the current AI infrastructure buildout? At what point could the principal drivers of your business mature and growth slow?

感谢Tim。现在我们进入问答环节。我们对收到的重复问题进行了合并，以提高讨论效率。第一个问题：当前AI基础设施建设的可持续性如何？贵公司的主要业务驱动力何时可能趋于成熟、增速放缓？

As I mentioned during my presentation, AI is more than a model. It represents a fundamental shift in computing. For the first time in 60 years, computing is being reinvented — from largely retrieving, storing and sending information to generating intelligence with AI.

正如我在演讲中提到的，AI不仅仅是一个模型，它代表着计算范式的根本性转变。60年来首次，计算正在被重新定义——从以检索、存储和传输信息为主，转向以AI生成智能。

Tokens are the fundamental unit of intelligence. They are manufactured in a new type of data center called an AI factory and monetized to generate revenue. The more compute, the more tokens; the more tokens, the more revenue. This buildout will be measured in decades, similar to other essential infrastructure initiatives, including the electric grid, transportation systems and the internet. We believe this will become the largest infrastructure buildout in human history. Agentic AI is accelerating infrastructure investments as, for the first time, AI is doing real work and creating real economic value.

Token是智能的基本单元。它们在一种被称为AI工厂的新型数据中心中被制造出来，并转化为收入。算力越多，Token越多；Token越多，收入越多。这一建设周期将以十年为单位衡量，类似于电网、交通系统和互联网等重大基础设施工程。我们相信这将成为人类历史上规模最大的基础设施建设。智能体式AI正在加速基础设施投资，因为AI第一次在做真正的工作、创造真正的经济价值。

As organizations seek to manufacture intelligence at scale, demand for AI factories will extend far beyond today's clouds to enterprises, sovereign nations and regional AI clouds. Nvidia is uniquely enabled to support AI factory buildouts with our full stack, end-to-end co-design infrastructure and a large ecosystem of partners supporting these initiatives.

随着各类机构寻求大规模制造智能，对AI工厂的需求将远远超出现有的云服务商，延伸至企业、主权国家和区域性AI云。英伟达凭借全栈端到端协同设计的基础设施以及庞大的合作伙伴生态系统，具备独特的能力来支持AI工厂的建设。

The next major leg of growth is physical AI, which is only beginning. Over time, AI will move beyond the digital world into robots, taxis, humanoid robots and industrial systems, enabling these machines to perceive, reason and act with agency in the physical world. Harnessing physical AI will require a new wave of infrastructure investment. Here, Nvidia provides AI infrastructure, Omniverse simulation and digital twin platforms, open models, Jetson and Thor embedded compute, and software stacks to develop and deploy physical AI at scale. We have a long runway of growth ahead of us.

下一个主要增长引擎是物理AI，而这才刚刚起步。随着时间推移，AI将超越数字世界，进入机器人、出租车、人形机器人和工业系统，使这些机器能够在物理世界中感知、推理并采取有自主性的行动。驾驭物理AI需要新一轮基础设施投资。在这一领域，英伟达提供AI基础设施、Omniverse仿真与数字孪生平台、开放模型、Jetson和Thor嵌入式计算以及软件栈，以大规模开发和部署物理AI。我们面前有着漫长的增长跑道。

Nvidia's GPUs now power the majority of the world's AI training infrastructure. But as inference workloads overtake training, how confident is Nvidia that the GPU architecture remains the preferred platform for inference at scale?

英伟达的GPU目前驱动着全球大部分AI训练基础设施。但随着推理工作负载超越训练，英伟达有多大信心认为GPU架构在大规模推理场景下仍然是首选平台？

Nvidia is indeed excellent at training, and with Blackwell we also established leadership for inference, which is monetizing AI. Because data centers are power-limited, their token throughput and revenue potential are defined by the AI infrastructure's performance per watt. In SemiAnalysis inference benchmarks, Blackwell was declared the inference king, delivering the best performance per watt and the lowest cost per token, and 30 times higher token throughput.

英伟达在训练方面确实表现出色，而凭借Blackwell，我们同样在推理领域确立了领先地位——推理正是AI变现的关键。由于数据中心受功率限制，其Token吞吐量和收入潜力取决于AI基础设施的每瓦性能。在SemiAnalysis推理基准测试中，Blackwell被宣布为"推理之王"，实现了最佳每瓦性能和最低每Token成本，以及30倍的Token吞吐量优势。

In the latest round of MLPerf inference benchmarks, we delivered our seventh consecutive win. And for AI agents, SemiAnalysis new AgentPerf results showed Grace Blackwell 300 NVLink 72 systems run up to 20 times more agents per megawatt compared to Hopper. Its superior performance is the culmination of Nvidia's extreme co-design across chips, systems, algorithms and software. Nvidia AI infrastructure delivers the best performance and therefore the best inference economics.

在最新一轮MLPerf推理基准测试中，我们实现了连续第七次胜出。在AI智能体方面，SemiAnalysis最新的AgentPerf测试结果显示，Grace Blackwell 300 NVLink 72系统每兆瓦可运行的智能体数量是Hopper的20倍。这一卓越性能是英伟达在芯片、系统、算法和软件层面极致协同设计的结晶。英伟达AI基础设施提供了最佳性能，因此也提供了最优的推理经济性。

But our advantage extends beyond our leading performance. Our large install base enables developers to reach the greatest number of AI users. Nvidia's programmable GPU architecture runs over 7,000 applications, giving cloud customers the greatest revenue opportunity and potential for financing and offtake. For enterprises and sovereigns, Nvidia offers unmatched deployment flexibility, the broadest set of leaderboard open models for custom AI, and the industry's deepest IT ecosystem support. Today, the vast majority of Nvidia's compute footprint is used for inference, and we are favorably positioned to expand our share, as we demonstrated with our recent announcements with Anthropic and Apple.

但我们的优势不仅限于领先的性能。庞大的装机量使开发者能够触达最多的AI用户。英伟达可编程GPU架构运行超过7000个应用程序，为云客户提供了最大的收入机会和融资潜力。对于企业和主权客户，英伟达提供无与伦比的部署灵活性、最丰富的排行榜开源模型用于定制AI，以及业界最深的IT生态系统支持。如今，英伟达绝大部分算力用于推理，而我们处于有利地位来扩大份额——我们与Anthropic和Apple的近期合作公告就是例证。

Is supporting frontier systems worth the national security and reputational risk?

支持前沿系统是否值得承担国家安全和声誉风险？

NVDIA has long been a leading American innovator. For over three decades, it is a core part of the United States AI and semiconductor industrial base. For over a decade, Nvidia designed, built and delivered the first AI supercomputer to American AI researchers, and since that time has remained dedicated to keeping American researchers, academics, startups and businesses at the frontier of AI.

英伟达长期以来一直是美国领先的创新企业。30多年来，它始终是美国AI和半导体产业基础的核心组成部分。十多年来，英伟达为美国AI研究人员设计、制造和交付了首批AI超级计算机，此后一直致力于让美国的研究人员、学者、初创企业和企业保持在AI前沿。

America's safety, security and prosperity depend on leadership across all five layers of the AI stack: from energy, chips and infrastructure to models and applications. Supporting frontier systems strengthens that leadership, creates high-skilled American jobs, helps re-industrialize our nation, and reinforces both economic and national security.

美国的安全、保障和繁荣取决于在AI栈全部五个层面——从能源、芯片、基础设施到模型和应用——的领先地位。支持前沿系统能够巩固这一领导力，创造高技能的美国就业岗位，助力国家再工业化，并增强经济安全和国家安全。

America's technology leadership is one of our national treasures, and NVDIA is proud to be one of its core pillars. We are an American company building American technology, investing in American workers and supporting a broader industrial base that helps ensure that the United States, not strategic competitors, shapes the future of AI. This is a once-in-a-generation opportunity to re-industrialize America with chips and systems manufacturing, bring back advanced manufacturing jobs, upgrade our aging power grid and invest in sustainable energy. Because of AI,

美国的技术领导力是我们国家的宝贵财富之一，英伟达很自豪能成为其核心支柱之一。我们是一家美国公司，建造美国技术，投资美国工人，支持更广泛的产业基础，以确保美国——而非战略竞争对手——塑造AI的未来。这是一代人一次的再工业化机遇，涵盖芯片和系统制造、带回先进制造业就业岗位、升级老旧电网以及投资可持续能源。得益于AI，

we have the profits and scale to invest in partners such as Intel, Micron, TSMC, Amkor, Foxconn, Quanta, Corning, Coherent and Lumentum — all of which are hiring and building factories in the United States to support NVDIA's business and the broader AI ecosystem. This deeply U.S.-anchored ecosystem enhances supply chain resilience and supports long-term value creation for our shareholders.

我们拥有足够的利润和规模来投资英特尔、美光、台积电、Amkor、富士康、广达、康宁、Coherent和Lumentum等合作伙伴——它们都在美国招聘并建厂，以支持英伟达的业务和更广泛的AI生态系统。这个深深扎根于美国的生态系统增强了供应链韧性，支持为股东创造长期价值。

In the energy sector, companies such as Caterpillar, GE, Baker Hughes and Bertie, along with partners in nuclear and renewable energy, we're working with them to enhance America's power generation and grid capabilities. This buildout of resilient, modern infrastructure is essential to enable America to lead this next industrial revolution and to ensure that our AI infrastructure is secure, reliable and domestically anchored.

在能源领域，我们与卡特彼勒、通用电气、Baker Hughes和Bertie等公司以及核能和可再生能源合作伙伴携手，提升美国的发电和电网能力。这一韧性现代化基础设施的建设，对于美国引领下一次工业革命、确保我们的AI基础设施安全、可靠且扎根国内至关重要。

Is NVDIA concerned that its products reach restricted or adversary end users through diversion?

英伟达是否担心其产品通过转售渠道落入受限或对手方最终用户手中？

National security is first and foremost. Where commercial opportunities conflict with U.S. national security, national security comes first. Nvidia complies with U.S. export controls and is prepared to accept commercial restrictions where required by law or policy. We are committed to compliance with all U.S. laws and work in collaboration with the entire AI ecosystem and the U.S. government. The company sells data center GPUs to well-known, large-scale customers such as original equipment manufacturers, hyperscalers and cloud service providers. We work with our partners' legal and compliance teams and coordinate with law enforcement to ensure compliance with all regulations.

国家安全始终是第一位的。当商业机会与美国国家安全发生冲突时，国家安全优先。英伟达遵守美国出口管制法规，并在法律或政策要求时准备接受商业限制。我们致力于遵守所有美国法律，并与整个AI生态系统和美国政府协作。公司向原始设备制造商、超大规模云服务商和云服务商等知名大型客户销售数据中心GPU。我们与合作伙伴的法律和合规团队协同，并与执法部门配合，确保遵守所有法规。

Our compliance efforts have repeatedly stopped would-be smugglers who risk prosecution on both sides of the globe. Moreover, advanced AI data centers are massive, integrated systems that require trusted hardware, software, networking and continuing support. Trying to cobble together data centers with some smuggled products is a dead end. Building them out of contraband would be extremely difficult, and we do not provide any support or repairs to restricted products.

我们的合规工作多次阻止了企图走私的行为，让这些人在全球各地都面临被起诉的风险。此外，先进的AI数据中心是庞大的集成系统，需要可信赖的硬件、软件、网络和持续支持。试图用走私产品拼凑数据中心是死路一条。用违禁品搭建数据中心极其困难，而且我们不对受限产品提供任何支持或维修。

Why hasn't NVDIA offered opinions to the U.S. government regarding export controls? Is there a formal plan in place for selling data center compute to the China market?

英伟达为何没有就出口管制向美国政府表达意见？是否已有向中国销售数据中心算力的正式计划？

As an American company, Nvidia has an obligation to help our government promote national security, economic security and sustain leadership of America's technology industry. We believe these goals can be achieved together through strong, balanced policies that both protect sensitive information and technologies and preserve U.S. leadership. National security is foremost. Our government counts on America's innovators and private corporations to deliver prosperity, create high-paying jobs and provide tax revenue that supports the national defense and supports essential government services.

作为一家美国公司，英伟达有义务帮助政府促进国家安全、经济安全并维持美国技术产业的领先地位。我们相信，通过强有力的、平衡的政策，这些目标可以同时实现——既保护敏感信息和技术，又维护美国的领先地位。国家安全是第一位的。我们的政府依靠美国的创新者和私营企业来创造繁荣、提供高薪就业和税收收入，以支持国防和基本政府服务。

The United States will only maintain its edge in AI if government, industry and researchers work together. Our role is to contribute our expertise so that policymakers can design strong, effective rules that keep the U.S. ahead of strategic competitors. Real-time information regarding advanced technologies, the AI ecosystem and emerging competition is essential to policymakers.

本文篇幅有限，内容有所精简，完整原文可移步我们的公众号查看，点击以下链接即可阅读全文

创始人必读｜英伟达2026股东大会：AI工厂开始赚钱，黄仁勋给出了下一轮产业答案