极致协同设计：英伟达 AI 主导地位的关键

你带领英伟达进入了 AI 的新时代，从专注于芯片级设计扩展到机架级设计。我认为可以说，长期以来英伟达的成功在于打造最好的 GPU，你仍然在这样做，但现在你已经将其扩展到 GPU、CPU、内存、网络、存储、电源、冷却、软件、机架本身、你宣布的 pod，甚至数据中心的极致协同设计。那么，我们来谈谈极致协同设计。协同设计一个包含这么多复杂组件和设计变量的系统，最难的部分是什么？

You've propelled Nvidia into a new era in AI, moving beyond a focus on chip-scale design to now rack-scale design. And I think it's fair to say that winning for Nvidia for a long time used to be about building the best GPU possible, and you still do, but now you've expanded that to extreme co-design of GPU, CPU, memory, networking, storage, power, cooling, software, the rack itself, the pod that you've announced, and even the data center. So, let's talk about extreme co-design. What is the hardest part of co-designing a system with that many complex components and design variables?

首先，极致协同设计之所以必要，是因为问题不再适合放在一台计算机内由单个 GPU 加速。你要解决的问题是，希望速度超过你增加的计算机数量。所以你增加了一万台计算机，但希望速度快一百万倍。那么突然之间，你必须处理算法，分解它，重构它，对流水线进行分片，对数据进行分片，对模型进行分片。当你这样分布问题时，不仅仅是规模扩大，而是分布问题，那么一切都会成为障碍。这就是阿姆达尔定律问题：你获得的加速取决于它占总工作负载的比例。如果计算占问题的 50%，你无限加速它，总工作负载只加速了两倍。现在你不仅要分布计算，还要对流水线进行分片，解决网络问题，因为所有这些计算机都连接在一起。在我们这种规模下的分布式计算：CPU 是个问题，GPU 是个问题，网络是个问题，交换是个问题，跨所有这些计算机分布工作负载也是个问题。这是一个极其复杂的计算机科学问题，所以我们必须动用所有技术，否则我们只能线性扩展，或者基于摩尔定律扩展，而摩尔定律由于登纳德缩放定律放缓已经大大减速。

First of all, the reason why extreme co-design is necessary is because the problem no longer fits inside one computer to be accelerated by one GPU. The problem you're trying to solve is you would like to go faster than the number of computers that you add. So you add 10,000 computers, but you would like it to go a million times faster. Then all of a sudden you have to take the algorithm, break it up, refactor it, shard the pipeline, shard the data, shard the model. Now when you distribute the problem this way, not just scaling up but distributing, then everything gets in the way. This is Amdahl's law problem: the speedup you get depends on how much of the total workload it is. If computation represents 50% of the problem and you sped it up infinitely, you only sped up the total workload by a factor of two. Now you have to distribute the computation, shard the pipeline, solve the networking problem because all these computers are connected together. Distributed computing at our scale: the CPU is a problem, the GPU is a problem, the networking is a problem, the switching is a problem, and distributing the workload across all these computers is a problem. It's a massively complex computer science problem, so we have to bring every technology to bear, otherwise we scale up linearly or based on Moore's law, which has largely slowed because Dennard scaling has slowed.

我相信这里有取舍。而且你这里有完全不同的学科。我相信你在每个领域都有专家：高带宽内存、网络、NVLink、NVSwitch、光学和铜缆、电源传输、冷却等等。我的意思是，每个领域都有世界级的专家。你怎么让他们聚在一个房间里一起解决？

I'm sure there are trade-offs there. Plus you have completely disparate disciplines here. I'm sure you have specialists in each one of these: high bandwidth memory, networking, NVLink, NVSwitch, optics and copper, power delivery, cooling, all that. I mean, there are world experts in each of those. How do you get them in a room together to figure it out?

这就是为什么我的团队这么大。

That's why my staff is so large.

你能带我了解一下专家和通才的过程吗？比如，当你知道了必须塞进机架的所有东西后，你是怎么组装机架的？一起设计的过程是怎样的？

Can you take me through the process of the specialists and the generalists? Like, how do you put together the rack when you know the set of things you have to shove into a rack together? What does that process look like of designing it all together?

第一个问题：什么是极致协同设计？我们在整个软件栈上进行优化，从架构到芯片到系统到系统软件到算法到应用。这是一个层面。第二件事超越了 CPU、GPU、网络芯片、纵向扩展交换机和横向扩展交换机。然后你还必须包括电源和冷却，因为所有这些计算机都非常耗电。它们做很多工作，能效很高，但总体上仍然消耗大量电力。所以第一个问题是：它是什么？第二个问题是为什么：我们刚刚谈到了原因——你希望分布工作负载，从而超越仅仅增加计算机数量的好处。第三个问题是如何：你怎么做？这就是这家公司的奇迹。当你设计一台计算机时，你必须有一个计算机的操作系统。当你设计一家公司时，你应该首先考虑你希望公司生产什么。我看到很多公司的组织架构图，它们看起来都一样——汉堡包组织架构图、软件组织架构图、汽车公司组织架构图。它们看起来都一样，这对我来说毫无意义。公司的目标是成为产生输出的机器、机制、系统，而这个输出就是我们想要创造的产品。公司的架构应该反映它存在的环境。它几乎直接告诉你应该怎么组织。我的直接下属有 60 人。我不和他们进行一对一的会议，因为那是不可能的。如果你要完成工作，你不可能有 60 个下属。

There's the first question: what is extreme co-design? We're optimizing across the entire stack of software, from architectures to chips to systems to system software to algorithms to applications. That's one layer. The second thing goes beyond CPUs and GPUs and networking chips and scale-up switches and scale-out switches. And then you have to include power and cooling because all these computers are extremely power-hungry. They do a lot of work and are very energy efficient, but in aggregate they still consume a lot of power. So that's the first question: what is it? The second question is why: we just spoke about the reason—you want to distribute the workload so you can exceed the benefit of just increasing the number of computers. And the third question is how: how do you do it? And that's the miracle of this company. When you're designing a computer, you have to have an operating system of computers. When you're designing a company, you should first think about what you want the company to produce. I see a lot of company organization charts and they all look the same—hamburger organization charts, software organization charts, car company organization charts. They all look the same, and it doesn't make any sense to me. The goal of a company is to be the machinery, the mechanism, the system that produces the output, and that output is the product we like to create. The architecture of the company should reflect the environment in which it exists. It almost directly says what you should do with the organization. My direct staff is 60 people. I don't have one-on-ones with them because it's impossible. You can't have 60 people on your staff if you're going to get work done.

所以你仍然有 60 个直接汇报。你还有更多。

So you still have 60 reports. You still have more.

是的。而且他们大多数至少涉足工程。几乎所有人都是。有内存专家、CPU 专家、光学专家、GPU 专家、架构专家、算法专家、设计专家。

Yeah. And most of them at least have a foot in engineering. Almost all of them. There are experts in memory, CPUs, optical, GPUs, architecture, algorithms, design.

所以你一直关注整个技术栈，并且必须进行关于整个技术栈设计的激烈讨论。

So you constantly have an eye on the entire stack and you're having to do intense discussions about the design of the entire stack.

而且没有一次对话是只有一个人的。这就是为什么我不做一对一的会议。我们提出一个问题，然后所有人一起攻克它，因为我们在做极致协同设计，而且实际上这家公司一直在做极致协同设计。

And no conversation is ever one person. That's why I don't do one-on-ones. We present a problem and all of us attack it, because we're doing extreme co-design and literally the company is doing extreme co-design all the time.

所以即使你在谈论一个特定的组件，比如冷却或网络，每个人都在听。

So even if you're talking about a particular component like cooling or networking, everybody's listening in.

是的。

Yeah.

然后他们可以贡献意见：‘嗯，这对配电不行。这对内存不行。这对这个不行。’

And they can contribute: 'Well, this doesn't work for the power distribution. This doesn't work for the memory. This doesn't work for this.'

完全正确。谁想走神就走神。你明白我的意思吗？原因是团队里的人知道什么时候该注意。他们应该——有些他们本可以贡献的东西，他们没有贡献。

Exactly. And whoever wants to tune out, tunes out. You know what I'm saying? The reason is because the people on the staff know when to pay attention. They're supposed to—something they could have contributed to, they didn't contribute to.

那么，正如你提到的，英伟达是一家适应环境的公司。在哪个节点上你可以说环境发生了变化？你们从早期为游戏设计的 GPU，到早期的深度学习革命，再到现在我们开始把它视为 AI 工厂，英伟达做的就是生产 AI。让我们建一个制造 AI 的工厂。

So, as you mentioned, Nvidia is this company that's adapting to the environment. At which point can you say the environment changed? You began adapting sort of secretly in the early days from GPU for gaming to the early deep learning revolution, to now we're going to start thinking of it as an AI factory. What Nvidia does is produce AI. Let's build a factory that makes AI.

我可以系统地推理一下。我们最初是一家加速器公司，但加速器的问题是应用领域太窄。它的好处是能为特定任务进行极致优化，任何专家都有这个优势。但高度专业化的问题在于市场范围更窄，不过这还好。真正的问题是市场规模决定了你的研发能力，而研发能力最终决定了你在计算领域可能拥有的影响力和影响力。所以当我们最初作为加速器起步时，是一个非常具体的加速器，我们一直知道那是我们的第一步。我们必须找到一种方式成为加速计算公司。但问题是，当你成为一家计算公司时，它太通用，会削弱你的专业化。我把两个实际上存在根本矛盾的字连在一起：我们越成为好的计算公司，就越成为差劲的专家；越成为专家，做整体计算的能力就越弱。我故意把这两个词连在一起：公司必须一步一步地找到那条非常狭窄的道路，扩大我们的计算视野，但又不放弃我们最重要的专业化。我们超越加速的第一步是发明了可编程像素着色器。那是迈向可编程性的第一步，是我们进入计算世界的第一段旅程。我们做的第二件事是把 FP32 放入着色器。那个 FP32 步骤，即 IE 兼容的 FP32，是迈向计算的一大步。正是因为这个原因，所有研究流处理器和其他类型数据流处理器的人发现了我们，他们说：‘嘿，突然之间我们也许可以用这个计算能力极强的 GPU，而且它现在兼容了；我可以把以前在 CPU 上写的软件拿来试试用 GPU。’这促使我们在 FP32 之上创建了 CG，我们称之为 CG。那条 CG 路径最终一步步把我们带到了 CUDA。把 CUDA 放在 GeForce 上是一个非常非常艰难的战略决策，因为它消耗了公司巨额利润，当时我们负担不起，但我们还是做了，因为我们想成为一家计算公司。一家计算公司必须有一个计算架构。计算架构必须在我们制造的所有芯片上兼容。

I could reason through it systematically. We started out as an accelerator company, but the problem with accelerators is that the application domain is too narrow. It has the benefit of being incredibly optimized for the job. Any specialist has that benefit. The problem with intense specialization is that your market reach is narrower, but that's even fine. The problem is the market size also dictates your R&D capacity. And your R&D capacity ultimately dictates the influence and impact that you can possibly have in computing. So when we first started out as an accelerator, a very specific accelerator, we always knew that was going to be our first step. We had to find a way to become accelerated computing. But the problem is when you become a computing company, it's too general purpose and it takes away from your specialization. I connected two words that actually have fundamental tension. The better computing company we become, the worse we become as a specialist. The more of a specialist, the less capacity we have to do overall computing. I connected those two words together on purpose: the company has to find that really narrow path step by step to expand our aperture of computing but not give up on the most important specialization we had. The first step we took beyond acceleration was we invented the programmable pixel shader. That was the first step towards programmability, our first journey towards moving into the world of computing. The second thing we did was we put FP32 into our shaders. That FP32 step, IE-compatible FP32, was a huge step in the direction of computing. It was the reason why all the people working on stream processors and other types of data flow processors discovered us and said, 'Hey, all of a sudden we might be able to use this GPU that's incredibly computationally intensive and it's now compliant; I can take my software that I was writing previously on CPUs and see about using the GPU for that.' That led us to create CG on top of FP32, which we called CG. That CG path took us eventually to CUDA, step by step. Putting CUDA on GeForce was a strategic decision that was very, very hard to do because it cost the company enormous amounts of our profits and we couldn't afford it at the time, but we did it anyway because we wanted to be a computing company. A computing company has a computing architecture. A computing architecture has to be compatible across all the chips we build.

你能带我回顾一下那个决策吗？把 CUDA 放在 GeForce 上，你们负担不起。你能解释一下那个决策吗？为什么还是大胆地选择了这么做？

Can you take me through that decision? Putting CUDA on GeForce, you could not afford to do. Can you explain that decision? Why boldly choose to do that anyway?

问得好。那是第一个近乎生死存亡的战略决策。对于不了解的人来说，结果证明——剧透警告——这是公司有史以来最出色的决策之一。CUDA 最终成为 AI 基础设施世界中一个不可思议的计算基础。所以你只是设定了背景。结果证明这是一个好决策。

Excellent. That was the first strategic decision that was as close to an existential threat. For people who don't know, it turned out to be, spoiler alert, one of the most incredibly brilliant decisions ever made by a company. So CUDA turned out to be an incredible foundation for computation in this AI infrastructure world. So you're just setting the context. It turned out to be a good decision.

所以你只是在设定背景。结果证明这是一个好决策。

So you're just setting the context. It turned out to be a good decision.

是的，结果证明这是一个好决策。我想事情是这样的。我们发明了 CUDA 这个东西，它扩大了我们可以用加速器加速的应用范围。问题是如何吸引开发者使用 CUDA？因为计算平台全靠开发者。开发者不会仅仅因为一个计算平台能执行一些有趣的任务就投奔它。他们来是因为安装基数大。因为开发者和其他人一样，希望开发的软件能触及很多人。所以安装基数实际上是一个架构最重要的部分。架构可能会招致大量批评。例如，没有哪个架构比 x86 受到更多批评了。作为一个不那么优雅的架构，它却是当今定义性的架构。这给你一个例子：实际上许多 RISC 架构，由一些世界上最聪明的计算机科学家精心设计、架构优美，却大多失败了。所以我给了你两个例子：一个优雅，另一个几乎谈不上美学，但 x86 生存了下来。安装基数就是一切。安装基数定义了一个架构；其他都是次要的。当时还有其他架构。CUDA 出现了，OpenCL 也在。还有几个竞争架构。但我们做的好决策是：‘嘿，看，归根结底是安装基数。我们怎样才能把一个新的计算架构推向世界？’到那时，GeForce 已经成功了。我们每年已经卖出数百万个 GeForce GPU。我们说应该把 CUDA 放在 GeForce 上，放进每一台 PC，不管客户用不用，把它作为培养安装基数的起点。同时，我们去吸引开发者。我们去大学，写书，讲课，把 CUDA 放到各处。最终人们发现了它。当时，PC 是主要的计算载体。没有云。我们可以把一台超级计算机交到学校里每一个研究员、科学家、工程学院、学生手中，最终会有神奇的事情发生。但问题是，CUDA 极大地增加了那个 GPU 的成本，它是一个消费产品，以至于完全消耗了公司的所有毛利润。所以当时，公司大概值，我不知道，当时是 80 亿美元还是 60-70 亿美元？

Yeah, it turned out to have been a good decision. I think here's the way it went. We invented this thing called CUDA and it expanded the aperture of applications that we can accelerate with our accelerator. The question is how do we attract developers to CUDA? Because a computing platform is all about developers. And developers don't come to a computing platform just because it could perform something interesting. They come to a computing platform because the install base is large. Because a developer, like anybody else, wants to develop software that reaches a lot of people. So the install base is in fact the single most important part of an architecture. The architecture could attract enormous amounts of criticism. For example, no architecture has ever attracted more criticism than the x86. As a less than elegant architecture, yet it is the defining architecture of today. It gives you an example that in fact many RISC architectures, which were beautifully architected, incredibly well-designed by some of the brightest computer scientists in the world, largely failed. So I've given you two examples where one is elegant, the other is barely aesthetic, and yet x86 survived. Install base is everything. Install base defines an architecture; everything else is secondary. There were other architectures at the time. CUDA came out, OpenCL was here. There were several other competing architectures. But the decision we made that was good was we said, 'Hey, look, ultimately it's about installed base. What is the best way we could get a new computing architecture into the world?' By that time, GeForce had become successful. We were already selling millions and millions of GeForce GPUs a year. And we said we ought to put CUDA on GeForce and put it into every single PC, whether customers use it or not, and use it as a starting point of cultivating our installed base. Meanwhile, we'll go and attract developers. We went to universities, wrote books, taught classes, and put CUDA everywhere. Eventually people discover it. At the time, the PC was the primary computing vehicle. There was no cloud. And we could put a supercomputer in the hands of every researcher in school, every scientist, every engineering school, every student, and eventually something amazing will happen. Well, the problem was CUDA increased our cost of that GPU, which is a consumer product, so tremendously that it completely consumed all of the company's gross profit dollars. So at the time, the company was probably worth, I don't know, at the time, eight, was it like $8 billion or something like $6-7 billion?

我们推出 CUDA 后，我意识到它会大幅增加成本，但我们坚信它。你知道，我们的市值一度跌到大约 15 亿美元。我们低迷了一段时间，然后慢慢爬了回来，但我们把 CUDA 搭载在了 GeForce 上。我常说 Nvidia 是 GeForce 建起的大厦，因为正是 GeForce 把 CUDA 带给了所有人。研究人员、科学家，他们是在 GeForce 上发现 CUDA 的，因为他们很多人都是游戏玩家。很多人在大学实验室里自己组装 PC，很多人用 PC 组件自己搭建集群，我们就是这样起步的。

After we launched CUDA, I recognized that it was going to add so much cost, but it was something we believed in. You know, our market cap went down to like $1.5 billion. And so, we were down there for a while and we clawed our way back slowly, but we carried CUDA on GeForce. I always say that Nvidia is the house that GeForce built because it was GeForce that took CUDA out to everybody. Researchers, scientists, they discovered CUDA on GeForce because they were all, you know, many of them were gamers. Many of them built their own PCs anyways in a university lab. Many of them built clusters themselves using PC components and so that's kind of how we got going.

然后这成了平台，深度学习革命的基础。这也是一个很棒的洞察，没错。

And then that became the platform, the foundation for the deep learning revolution. That was also another great observation, yeah.

那个生死攸关的时刻，你还记得那些会议是什么样的吗？那些讨论是什么样的？作为一家公司，决定冒一切风险？

That existential moment, do you remember like what were those meetings like, what were those discussions like deciding as a company risking everything?

嗯，我必须向董事会明确我们想做什么，管理团队也知道我们的毛利率会受到重创。所以你可以想象这样一个世界：GeForce 要承担 CUDA 的负担，而游戏玩家既不会欣赏它，也不会为此买单。你知道，他们只愿意付某个价格，不管你的成本是多少。所以我们把成本提高了 50%，而我们当时是一家毛利率 35% 的公司。所以这是一个非常艰难的决定，但你可以想象有一天它能进入工作站和超级计算机，在这些领域我们或许能获得更高的利润率。所以你可以通过推理让自己相信能够承担得起。但这仍然花了十年时间。

Well, I had to make it clear to the board what we were trying to do and the management team knew our gross margins were going to get crushed. So you could imagine a world where GeForce would carry the burden of CUDA and none of the gamers would appreciate it and none of the gamers would pay for it. You know, they only pay a certain price and it doesn't matter what your cost is. And so we increased our cost by 50% and we were a 35% gross margin company. So it was quite a difficult decision to make, but you could imagine that someday this could go into workstations and it would go into supercomputers and in those segments maybe we can capture more margin. So you could reason your way into being able to afford this. But it still took a decade.

但这更像是和董事会对话，说服他们。但心理上，因为 Nvidia 一直在做出大胆的押注，预测未来，尤其是现在定义未来。所以我几乎是在寻找智慧，关于你如何能够做出那些决定，让公司实现这样的飞跃。

But that's more like conversation with the board convincing them. But you psychologically, because Nvidia has continued to make bold bets that predict the future and in part especially now define the future. So I'm almost looking for wisdom about how you were able to make those decisions to make leaps like that as a company.

嗯，首先，我受好奇心驱动。在某个时刻，有一个推理系统让我如此确信这个结果会发生，以至于它一定会发生。所以我在心里相信它。当你心里相信它时，你知道那种感觉。你显化了一个未来。那个未来如此令人信服，它不可能不发生。中间会有很多痛苦，但你必须相信你所相信的。

Well, first of all, I'm informed by a lot of curiosity. At some point there's a reasoning system that convinces me so clearly this outcome will happen that this will happen. And so I believe it in my mind. And when I believe it in my mind, you know how it is. You manifest a future. And that future is so convincing, there's no way it won't happen. There's a lot of suffering in between, but you've got to believe what you believe.

所以你预见了未来。

So you envision the future.

是的。

Yeah.

而且你基本上是从工程的角度把它显化出来。

And you essentially from a sort of engineering perspective manifest it.

是的。你推理如何到达那里。你推理为什么它必须存在。我推理，我们这里所有人都推理。管理团队会推理。我花了很多时间和所有人一起推理。接下来的部分可能是一种技巧，你知道在领导力中，领导者常常保持沉默，或者他们了解了一些东西，然后搞个宣言，新的一年，不知怎么地到了年底明年我们会有全新的计划，大规模裁员，大规模组织变革，新的使命宣言，全新的标志，你知道那种东西。我们从来没有，我从不那样做。当我了解到某件事，它开始影响我的思考时，我会非常清楚地向身边的每个人表明，这很有趣。这会产生影响。这会影响到那个。我一步一步地推理。很多时候我已经下定了决心，但我会抓住每一个可能的机会，外部信息、新见解、新发现、新的工程启示、新的里程碑，我会利用这些机会来塑造其他人的信念体系。我每天都在这样做。我对董事会这样做，对管理团队这样做，对员工这样做。我试图塑造他们的信念体系，这样当有一天我说，‘嘿，我们买下 Mellanox 吧’，每个人都完全明白我们绝对应该这么做。在我那天说，‘嘿，伙计们，让我们全力投入深度学习’的时候，让我告诉你为什么。我已经在为公司内部的不同组织铺路。每个组织，很多人可能都听过所有内容，公司大部分人都听过片段。在我宣布的那天，每个人都已经接受了其中的很多部分。在很多方面，我喜欢宣布这些事情，我想象员工们会说，‘Jensen，你怎么才说？’事实上，我已经在塑造他们的信念体系一段时间了。因此，领导力有时看起来像是你在后面引领。

Yeah. And you reason about how to get there. You reason about why it must exist. And I reason, we all reason here. The management team will reason about it. All the people that I spend a lot of time reasoning about it. The next part of it is probably a skill thing, which is you know oftentimes in leadership, the leadership stays quiet or they learn about something and then they do some manifesto and it's a brand new year and somehow at the end of the year next year we're going to have a brand new plan, big huge layoff this way, big huge organization change this way, new mission statement, brand new logos, you know that kind of stuff. We've just never, I never do things that way. When I learn about something and it's starting to influence how I think, I'll make it very clear to everybody near me that this is interesting. This is going to make a difference. This is going to impact that. And I reason about things step by step by step. Often times I've already made up my mind but I'll take every possible opportunity, external information, new insights, new discoveries, new engineering revelations, new milestones developed, I'll take those opportunities and I'll use it to shape everybody else's belief system. And I'm doing that literally every single day. I'm doing that with my board, I'm doing that with my management team, I'm doing that with my employees. I'm trying to shape their belief system such that when I come the day I say, 'Hey, let's buy Mellanox.' It's completely obvious to everybody that we absolutely should. On the day that I said, 'Hey guys, let's go all in on deep learning.' And let me tell you why. I've already been laying down the bricks to different organizations inside the company. Every organization and many of the people might have heard everything, most of the company heard pieces of it. And on the day that I announce it, everybody's kind of bought into many pieces of it. And in a lot of ways I like to announce these things and I imagine that the employees are kind of saying, 'Jensen, what took you so long?' And in fact I've been shaping their belief system for some time. And therefore leadership sometimes looks like you're leading from behind.

但你一直在塑造他们的信念，到了我宣布的那天，100% 认同。但这就是你想要的，你想让每个人都跟上。否则我们宣布一些关于深度学习的事情，每个人都会说‘你在说什么？’你宣布让我们全力投入这件事，而你的管理团队、董事会、员工、客户，他们会想，‘这是从哪来的？这太疯狂了。’

But you've been shaping their, you know, to the point where on the day that I declared it, 100% buy in. But that's what you want, you want to bring everybody along. Otherwise we announce something about deep learning and everybody goes 'What are you talking about?' You announce something about let's go all in on this thing and your management team, your board, your employees, your customers, they're kind of like, 'Where's this coming from? This is insane.'

所以，GTC 实际上，如果你回顾过去，看看主题演讲，我也在塑造合作伙伴和整个行业的信念体系，并用它来塑造我自己的员工的信念体系。所以当我宣布某件事时，比如，我们刚刚宣布了 Grock，我们已经晚了。我已经谈论了两年半的垫脚石。你们回去看看，哦天哪，他们已经谈论了两年半。所以我一步一步地打下基础。所以当时机成熟你宣布时，每个人都会说，‘你怎么才说？’

And so, GTC, in fact, if you go back in time, you look at the keynotes, I'm also shaping the belief system of my partners and the industry and I'm using that to shape the belief system of my own employees. And so by the time that I announce something, like, for example, we just now we announced Grock, we've been late. I've been talking about the stepping stones for two and a half years. You guys just go back and oh my gosh, they've been talking about it for two and a half years. And so I've been laying the foundation step by step by step. So when the time comes you announce it, everybody's, you know, 'What took you so long?'

但这不仅仅是在公司内部。你在塑造整个格局，更广泛的全球创新格局。就像把这些想法抛出去，你真的在显化现实。

But it's not just inside the company. You're shaping the landscape, the broader global landscape of innovation. Like putting those ideas out there, you really are manifesting reality.

我们不制造计算机。我们实际上不构建云。结果发现，我们是一家计算平台公司，所以没人能从我们这里买到任何东西。这就是奇怪的地方。

We don't build computers. We actually don't build clouds. We don't, as it turns out, we're a computing platform company and so nobody can buy anything from us. That's the weird thing.

你知道，我们垂直设计、垂直整合以进行设计和优化，但随后我们在每一层都开放整个平台，以便集成到其他公司的产品、服务、云、超级计算机和 OEM 计算机中。所以神奇的是，如果我没有先说服他们，我就无法做我所做的事情。因此，GTC 的大部分内容都是关于展现一个未来，以至于当我的产品准备好时，他们会说：‘怎么花了这么长时间？’

You know, we vertically design and vertically integrate to design and optimize, but then we open up the entire platform at every single layer to be integrated into other companies' products and services and clouds and supercomputers and OEM computers. And so the amazing thing is I can't do what I do without having convinced them first. And so most of GTC is about manifesting a future that by the time my product is ready, they're going, 'What took you so long?'

那么，你长期以来一直相信的一件事就是广义上的缩放定律。那么你仍然相信缩放定律吗？

So one of the things you've been a believer for a long time is scaling laws broadly defined. So are you still a believer in the scaling laws?

是的，我们现在有更多的缩放定律。

Yeah, we have more scaling laws now.

所以我认为你概述了其中的四个：预训练、后训练、测试时间和智能体式缩放。当你思考未来，遥远的未来和近期未来时，你最担心哪些障碍，让你夜不能寐，必须克服才能继续扩展？

So I think you've outlined four of them with pre-training, post-training, test time, and agentic scaling. What do you think when you think about the future, deep future and the near-term future, what are the blockers that you're most concerned about that keep you up at night that you have to overcome in order to keep scaling?

嗯，我们可以回顾一下人们曾经认为的障碍。一开始，我们是第一个。预训练缩放定律，你知道，人们认为，理所当然地，我们拥有的高质量数据量将限制我们达到的智能水平。那个缩放定律是一个重要的、非常重要的缩放定律：模型越大，相应地更多的数据会带来更好、更智能的 AI。那就是预训练。伊利亚·苏茨克弗说‘我们没有数据了’之类的话。‘预训练结束了’之类的话。整个行业恐慌了，你知道，认为这是 AI 的终结。当然，这显然不是真的。我们将继续扩展用于训练的数据量。其中很多数据可能是合成的。这也让人们感到困惑。人们没有意识到的是，他们有点忘记了我们用来训练、互相教学、互相告知的大部分数据都是合成的。它是合成的，因为它不是来自自然。你创造了它。我在消费它。我修改它、增强它、重新生成它，别人消费它。所以我们现在已经达到了一个水平，AI 能够获取真实数据，增强它、改进它，合成生成海量数据。后训练的那一部分继续扩展。因此，我们可以使用的人类生成的数据量将越来越小。我们用来训练模型的数据量将继续扩展，直到我们不再受数据限制；训练现在受算力限制。原因是大部分数据是合成的。下一阶段是测试时间。我还记得有人告诉我推理很容易，预训练很难。人们谈论的这些庞大系统，推理一定很容易。所以推理芯片将是微小的芯片。它们不像英伟达的芯片；那些会很复杂和昂贵。未来，推理将是最大的市场，而且会很容易，我们会将其商品化，每个人都可以制造自己的芯片。这对我来说一直不合逻辑，因为推理就是思考，我认为思考很难。思考比阅读难得多。预训练只是记忆和泛化，寻找模式和关系。你读啊读，而思考、推理、解决问题、处理未探索的经历、新经历，并将其分解成可解决的片段，然后我们通过第一性原理推理或通过先前的例子、先前的经验，或者只是探索、搜索和尝试不同的事情。测试时间扩展的整个过程——推理实际上是关于思考、推理、规划、搜索。那么这怎么可能被计算呢？我们对此完全正确。所以测试时间扩展是极其计算密集型的。那么问题来了，现在我们处于推理和测试时间扩展。之后呢？嗯，显然我们现在已经创造了一个智能体式的人，那个智能体式的人拥有我们开发的大型语言模型。但在测试时间，那个智能体式系统会去做研究，敲打数据库，使用工具，它做的最重要的事情之一是衍生出大量子智能体，这意味着我们现在正在创建大型团队。通过雇佣更多员工来扩展英伟达比扩展我自己容易得多。所以下一个缩放定律是智能体式缩放定律。这有点像倍增 AI。我们可以随心所欲地快速衍生智能体。所以，你知道，我有四个缩放定律。当我们使用智能体式系统时，它们会产生更多的数据，更多的经验。其中一些我们会说，‘哇，这真的很好。我们应该记住这个。’那个数据集然后一路回到预训练。我们记忆并泛化它。然后我们改进它并微调回后训练。然后我们通过测试时间进一步增强它，在智能体式系统中，将其投入行业。所以这个循环，这个周期，将一直持续下去。基本上归结为智能将通过一件事来扩展，那就是算力。但有一个棘手的问题你必须预测和预见，那就是其中一些组件需要不同类型的硬件才能真正优化。所以你必须预见 AI 创新将走向何方。例如，带有稀疏性的混合专家模型。对于硬件，你不能在一周的通知时间内转向。你必须预见那会是什么样子。这非常可怕且难以做到，对吧？例如，这些 AI 模型架构大约每六个月发明一次。而系统架构和硬件架构大约每三年一次。所以你需要预见两三年后可能发生的事情。有几种方法可以做到这一点。首先，我们可以在内部自己做研究。这就是我们拥有基础研究的原因之一。我们有应用研究。

Well, we can go back and reflect on what people thought were blockers. In the beginning, we were the first. The pre-training scaling law, you know, people thought, well, rightfully so, that the amount of high-quality data we have will limit the intelligence we achieve. And that scaling law was an important, very important scaling law: the larger the model, the correspondingly more data results in a better, smarter AI. And so that was pre-training. And Ilya Sutskever said, 'We're out of data,' or something like that. 'Pre-training is over,' or something like that. The industry panicked, you know, that this is the end of AI. And of course, that's obviously not true. We're going to keep on scaling the amount of data we have to train with. A lot of that data is probably going to be synthetic. And that also confused people. What people don't realize is they've kind of forgotten that most of the data we train with, that we teach each other with, inform each other with, is synthetic. It's synthetic because it didn't come out of nature. You created it. I'm consuming it. I modify it, augment it, I regenerate it, somebody else consumes it. And so we've now reached a level where AI is able to take ground truth, augment it, enhance it, synthetically generate an enormous amount of data. And that part of post-training continues to scale. And so the amount of data we could use that is human-generated will be smaller and smaller. The amount of data we use to train models is going to continue to scale to the point where we're no longer limited by data; training is now limited by compute. And the reason for that is most of the data is synthetic. Then the next phase is test time. And I still remember people telling me that inference is easy, pre-training is hard. These are giant systems that people are talking about; inference must be easy. And so inference chips are going to be little tiny chips. And they're not like Nvidia's chips; those are going to be complicated and expensive. And in the future, inference is going to be the biggest market, and it's going to be easy, and we're going to commoditize it, and everybody can build their own chips. And that was always illogical to me because inference is thinking, and I think thinking is hard. Thinking is way harder than reading. Pre-training is just memorization and generalization, looking for patterns and relationships. You're reading and reading versus thinking, reasoning, solving problems, taking unexplored experiences, new experiences, and breaking it down into solvable pieces that we then go off either through first-principle reasoning or through previous examples, prior experiences, or just exploration and search and trying different things. And that whole process of test time scaling—inference is really about thinking, reasoning, planning, search. So how could that possibly be computed? And we were absolutely right about that. So test time scaling is intensely compute-intensive. Then the question is, okay, now we're at inference and test time scaling. What's beyond that? Well, obviously we have now created one agentic person, and that one agentic person has a large language model that we've developed. But during test time, that agentic system goes off and does research, bangs on databases, uses tools, and one of the most important things it does is spin off and spawn a whole bunch of sub-agents, which means we're now creating large teams. It's so much easier to scale Nvidia by hiring more employees than it is to scale myself. And so the next scaling law is the agentic scaling law. It's kind of like multiplying AI. We could spin off agents as fast as you want. And so, you know, I have four scaling laws. And as we use the agentic systems, they're going to create a lot more data, a lot of experiences. Some of it we're going to say, 'Wow, this is really good. We ought to memorize this.' That dataset then comes all the way back to pre-training. We memorize and generalize it. We then refine it and fine-tune it back into post-training. Then we enhance it even more with test time, in the agentic systems, put it onto the industry. And so this loop, the cycle, is going to go on and on and on. It kind of comes down to basically intelligence is going to scale by one thing, and it's compute. But there's a tricky thing there that you have to anticipate and predict, which is some of these components require different kinds of hardware to really do it optimally. So you have to anticipate where the AI innovation is going to lead. For example, mixture of experts with sparsity. With hardware, you can't just pivot on a week's notice. You have to anticipate what that's going to look like. That's so scary and difficult to do, right? For example, these AI model architectures are being invented about once every six months. And system architectures and hardware architectures kind of every three years. And so you need to anticipate what likely is going to happen two or three years from now. And there are a couple ways you could do that. First of all, we could do research internally ourselves. And that's one of the reasons why we have basic research. We have applied research.

我们创建自己的模型。所以我们有亲身实践的经验。这就是我所说的代码设计的一部分。

We create our own models. And so we have hands-on life experience right here. This is part of the code design that I'm talking about.

我们也是世界上唯一一家与全球几乎所有 AI 公司都有合作的 AI 公司。我们尽可能去了解人们正在面临的挑战。

We're also the only AI company in the world that works with literally every AI company in the world. And to the extent that we can, we try to get a sense of what are the challenges that people are experiencing.

所以你是在倾听整个行业的低语和脉动。

So you're listening to the whispers across the industry, the adabs.

没错。你必须倾听并向所有人学习，拥有一个灵活、能随风而动的架构。CUDA 的优势之一在于，它一方面是一个不可思议的加速器，另一方面又非常灵活。这种专业化（否则我们无法加速 CPU）与通用化（以便适应不断变化的算法）之间的平衡非常重要。这就是 CUDA 如此有韧性的原因，而我们还在不断改进它。我们现在是 CUDA 13.2，架构演进非常快，能够跟上现代算法。例如，当混合专家模型出现时，这就是为什么我们有了 NVLink 72 而不是 NVLink 8。我们现在可以将整个 4 万亿或 10 万亿参数的模型放入一个计算域中，就像在单个 GPU 上运行一样。大家可能没注意到我说的话，但如果你看看 Grace Blackwell 机架的架构，它完全专注于做一件事：处理大语言模型。突然之间，一年后，你看到了 Vera Rubin 机架。它有存储加速器、一个名为 Vera 的新 CPU、Vera Rubin 和 NVLink72 来运行大语言模型，还有一个名为 Gro 的新增机架。整个机架系统与之前的完全不同，包含所有这些新组件。原因是上一个机架是为运行大语言模型推理而设计的，而这个是为运行智能体而设计的。智能体大量使用工具。显然，这个系统的设计必须在 Claude Code、Codex、OpenClaw 之前就完成了。所以你基本上是在预见未来。这来自于那些低语，来自于对最先进技术的理解。

That's right. You got to listen and learn from everybody and have a flexible architecture that can adapt and move with the wind. One of the benefits of CUDA is that it's an incredible accelerator on the one hand, and on the other hand it's really flexible. That balance between specialization, otherwise we can't accelerate the CPU, versus generalization, so that we can adapt with changing algorithms, is really important. That's the reason why CUDA has been so resilient, and yet we continue to enhance it. We're at CUDA 13.2, evolving the architecture so fast that we can stay with modern algorithms. For example, when mixture of experts came out, that's the reason why we had NVLink 72 instead of NVLink 8. We could now take an entire 4 trillion or 10 trillion parameter model and put it in one computing domain as if it's running on one GPU. People probably didn't notice I said it, but if you look at the architecture of the Grace Blackwell racks, it was completely focused on doing one thing: processing the LLM. All of a sudden, one year later, you're looking at a Vera Rubin rack. It has storage accelerators, a new CPU called Vera, Vera Rubin and NVLink72 to run the LLMs, and also a new additional rack called Gro. This entire rack system is completely different from the previous one, with all these new components. The reason is that the last one was designed to run large language model inference, and this one is to run agents. Agents bang on tools. Obviously, the design of the system had to have been done before Claude Code, Codex, OpenClaw. So you were anticipating the future essentially. That comes from the whispers, from understanding what all the state of the art is.

不，这比那更简单。你只需要推理。首先，推理：无论发生什么，在某个时刻，为了让那个大语言模型成为数字工作者——我们就用这个比喻——它需要做什么？它必须访问真实数据，也就是我们的文件系统。它必须能够做研究。它并非无所不知。我不想等到这个 AI 对过去、现在和未来的一切都变得普遍聪明之后才让它有用。所以，我不如让它去做研究。显然，如果它想帮助我，它必须使用我的工具。很多人会说 AI 将彻底摧毁软件。我们不再需要软件了。我们甚至不再需要工具了。这很荒谬。让我们做一个思想实验。你可以坐在那里，享受一杯威士忌，思考所有这些事情，然后一切都会变得非常明显。如果我要创造未来 10 年我们能想象到的最惊人的智能体，比如说一个人形机器人。如果那个人形机器人被创造出来，它更可能进入我的房子并使用我现有的工具来完成工作吗？还是它的手一会儿变成 10 磅重的锤子，一会儿变成手术刀，为了烧水就从手指射出微波？还是它更可能只是使用微波炉？它第一次走到微波炉前，可能不知道怎么用。但没关系。它连接着互联网。它读取这个微波炉的说明书，瞬间读完，成为专家，然后使用它。

No, it's easier than that. You just reason about it. First of all, just reason: no matter what happens, at some point, in order for that large language model to be a digital worker—let's just use that metaphor—what does it have to do? It has to access ground truth, that's our file system. It has to be able to do research. It doesn't know everything. I don't want to wait until this AI becomes universally smart about everything past, present, and future before I make it useful. So therefore, I might as well let it go do research. Obviously, if it wants to help me, it's got to use my tools. A lot of people would say AI is going to completely destroy software. We don't need software anymore. We don't even need tools anymore. That's ridiculous. Let's use a thought experiment. You could just sit there, enjoy a glass of whiskey, and think about all these things, and it would become completely obvious. If I were to create the most amazing agent we can imagine in the next 10 years, let's say a humanoid robot. If that humanoid robot were to be created, is it more likely that the robot comes into my house and uses the tools that I have to do the work it needs to do? Or does its hand turn into a 10-pound hammer in one instance, a scalpel in another, and to boil water it beams microwaves out of its fingers? Or is it more likely just to use the microwave? The first time it goes up to the microwave, it probably doesn't know how to use it. But that's okay. It's connected to the internet. It reads the manual of this microwave, reads it instantly, becomes an expert, and uses it.

所以我认为我刚才描述的，实际上几乎就是 Open Claw 的所有特性。

And so I think I just described, in fact, almost all of the properties of Open Claw.

嗯。

Mhm.

它会使用工具、访问文件、能够做研究，拥有 IO 子系统。当你以这种方式推理完之后，你会说：‘哦天哪，这对未来计算的影响极其深远。’原因是我们刚刚重新发明了计算机。然后你会说：‘好吧，我们是什么时候推理出这个的？我们是什么时候推理出 Open Claw 的？’如果你看看我在 GTC 上使用的 Open Claw 示意图，你会发现那是两年前。确切地说，两年前在 GTC 上，我就在谈论与今天的 Open Claw 完全一致的智能体系统。当然，许多事情必须汇聚在一起。首先，我们需要 Claude、GPT 以及所有这些模型达到一定的能力水平。所以它们的创新、突破和持续进步非常重要。然后，当然，必须有人创建一个足够健壮和完整的开源项目，让我们都能投入使用。我认为 OpenClaw 为智能体系统所做的，就像 ChatGPT 为生成式系统所做的一样。我认为这是一件大事。

It's going to use tools, access files, be able to do research, has an IO subsystem. And when you're done reasoning through it in that way, then you say, 'Oh my gosh, the impact to the future computing is deeply profound.' The reason for that is I think we've just reinvented the computer. And then now you say, 'Okay, when did we reason about that? When did we reason about Open Claw?' If you take the Open Claw schematic that I used at GTC, you will find it two years ago. Literally two years ago at GTC, I was talking about agentic systems that exactly reflect Open Claw today. And of course, the confluence of many things had to happen. First of all, we needed Claude and GPT and all these models to reach a level of capability. So their innovation and breakthroughs and continual advances were really important. And then of course, somebody had to create an open-source project that was sufficiently robust and complete, and that we can all put to work. I think OpenClaw did for agentic systems what ChatGPT did for generative systems. I just think it's a very big deal.

是的，这是一个非常特殊的时刻。我不太确定它为什么吸引了全世界如此多的关注，但它比 Claude Code、Codex 等更受关注，因为消费者能够接触到它。

Yeah, it's a really special moment. I'm not exactly sure why it captured so much of the world's attention, but it did more than Claude Code and Codex and so on because consumers could reach it.

当然。是的。但其中也有很多是氛围。我和彼得做过一期播客。他是个很棒的人。所以部分原因也在于代表这件事的人。部分原因在于迷因。

Sure. Yeah. But there's also so much of this is vibes. Peter, I had a podcast with him. He's a wonderful human being. So part of it is also the humans that represent the thing. Part of it is memes.

因为我们都在试图弄清楚。当你拥有如此强大的技术时，如何交出你的数据以便它们能做有用的事情，但同时又伴随着可怕的事情，这确实存在非常严重和复杂的安全问题。我们作为一个文明，作为个体，正在努力找到那个正确的平衡点。

Because we're all trying to figure it out. There are really serious and complicated security concerns about when you have such powerful technology, how do you hand over your data so they can do useful stuff, but then there are scary things associated with that. And we as a civilization, as individual people, are figuring out how to find that right balance.

是的，我们立刻行动了，派了一群安全专家过来。

Yeah, we jumped on it right away and we sent a bunch of security experts this way.

我们做了这个叫 Open Shell 的东西。它已经集成到 Open Claw 中了。

And we did this thing called Open Shell. It's already been integrated into Open Claw.

而且 Nvidia 推出了 Nemo Claw。

And Nvidia put forward Nemo Claw.

是的。没错。

Yep. Exactly.

安装非常简单。它确保了安全性。

The install is super easy. It makes sure that it's secure.

我们给你三分之二的权利。智能体系统可以访问敏感信息。

We give you two out of three rights. Agentic systems can access sensitive information.

它可以执行代码，也可以与外部通信。嗯。如果我们只给你三种能力中的两种，我们可以确保安全，但不能同时给三种。在这两种能力中，我们还根据企业授予的权限进行访问控制。然后我们连接到一个所有企业都已经拥有的策略引擎。所以我们会尽力帮助 Open Claw 变得更好。你雄辩地解释了历史上我们曾认为会是障碍的障碍，但我们克服了它们。但现在展望未来，既然智能体将无处不在，你认为可能有哪些障碍？

It can execute code and it can communicate externally. Mhm. We could keep things safe if we gave you two out of those three capabilities at any time, but not all three. And out of those two out of three capabilities, we also give you access control based on whatever rights that you're given by enterprise. And then we connected to a policy engine that all these enterprises already have. And so we're going to try to do our best to help Open Claw become a better claw. So you eloquently explained how we have a long history of blockers that we thought were going to be blockers and we overcame them. But now looking into the future, what do you think might be the blockers now that it's clear that agents will be everywhere?

显然我们需要算力。那么 Scaling 的障碍是什么？电力是一个问题，但不是唯一的问题。这就是为什么我们如此努力地推进极致架构设计，以便每年将每瓦每秒 token 数提高几个数量级。在过去 10 年里，摩尔定律会让计算能力进步约 100 倍。而我们通过扩展将计算能力提升了 100 万倍。我们将继续通过极致架构设计做到这一点。每瓦能效直接影响公司的收入，影响工厂的收入。我们会将其推向极限，以便尽可能快地降低 token 成本。我们的计算机价格在上涨，但 token 生成效率提升得更快，因此 token 成本在下降。每年下降一个数量级。

So it's obviously we're going to need compute. So what is going to be the blocker for that scaling? Power is a concern, but it's not the only concern. But that's the reason why we're pushing so hard on extreme code design so that we can improve the tokens per second per watt orders of magnitude every single year. And so in the last 10 years, Moore's law would have progressed computing about a 100 times in the last 10 years. We progressed and scaled up computing by a million times in the last 10 years. And so we're going to keep on doing that through extreme code design. So energy efficiency per watt completely affects the revenues of a company. It affects the revenues of a factory and we're just going to push that to the limit so that we can keep on driving token cost down as fast as we can. You know, our computer price is going up, but our token generation effectiveness is going up so much faster that token cost is coming down. It's coming down an order of magnitude every year.

所以电力是个有趣的问题。绕过电力障碍的方法是提高每瓦每秒 token 数的效率。当然，还有一个问题：我们如何获得更多电力？

So power that's an interesting one. So the way to try to get around the power blocker is to try to with the tokens per second per watt try to make it more and more efficient. Of course, there's the question, how do we get more power?

我们也应该获得更多电力。

We should also get more power.

这确实很复杂。你提到过小型模块化核电站。关于能源有各种想法。这让你夜不能寐的程度有多大？AI 供应链的瓶颈，比如 ASML 的 EUV 光刻机、台积电的先进封装（如 CoWoS）和 SK 海力士的高带宽内存，我们一直在努力。历史上没有一家公司能在我们这样的规模下同时加速增长。这太不可思议了。

That's a really complicated one. And you've talked about small module nuclear power plants. There's all kinds of ideas for energy. How much does it keep you up at night? The bottlenecks in the supply chain of AI like ASML with EUV lithography machines, TSMC with advanced packaging like CoWoS, and SK Hynix with high bandwidth memory all the time and we're working on all the time. No company in history has ever grown at a scale that we're growing while accelerating that growth. It's incredible.

是的。在 AI 计算的整个世界中，人们甚至很难理解这一点。我们的份额在增加。因此，供应链上游和下游对我们非常重要。我花了很多时间告诉我合作的所有 CEO，是什么动态因素会导致增长持续甚至加速。这就是为什么我右边坐着的 CEO 几乎涵盖了整个 IT 行业上游和整个基础设施行业下游。嗯。那里有几百位 CEO，我不认为有哪次主题演讲有几百位 CEO 出席。部分原因是我在告诉他们我们当前的业务状况，告诉他们近期增长驱动力和正在发生的事情。我也在描述我们下一步的方向，这样他们可以利用所有这些信息和动态来指导投资决策。我像告知自己员工一样告知他们。当然，我也会去拜访他们，确保他们知道这个季度、明年、后年会发生什么。如果你看看 DRAM 行业的 CEO，世界上排名第一的 DRAM 是用于数据中心 CPU 的 DDR 内存。大约三年前，我说服了几位 CEO，尽管当时 HBM 内存使用很少，几乎只用于超级计算机，但它未来会成为数据中心的主流内存。起初这听起来很荒谬，但几位 CEO 相信了我，并决定投资生产 HBM 内存。另一种放在数据中心里相当奇怪的内存是我们用于手机的低功耗内存。我们希望他们将其改造用于数据中心的超级计算机。他们说，手机内存用于超级计算机？我向他们解释了原因。看看这两种内存，LPDDR5 和 HBM4，它们的产量大得惊人。这三家公司都创下了历史纪录。而这些都是有 45 年历史的公司。所以，这是我工作的一部分：告知、塑造和激励。

Yeah. And it's hard for people to even understand this in the overall world of AI computing. We're increasing share. And so supply chain upstream and downstream are really important to us. I spent a lot of time informing all the CEOs that I work with what are the dynamics that's going to cause the growth to continue or even accelerate. It's part of the reasons why to the entire right hand side of me were CEOs of practically the entire IT industry upstream and practically the entire infrastructure industry downstream. Mhm. And they were all there were several hundred CEOs and I don't think there's ever been keynotes where several hundred CEOs show up. And part of it is I'm telling them about our business condition now. I'm telling them about the growth drivers in the very near future and what's happening. And I'm also describing where are we going to go next so that they could use all of this information and all of the dynamics that are here to inform how they want to invest. And so I inform them that way like I inform my own employees. And then of course then I make trips out to them and make sure that hey listen I want you to know this quarter, this coming year, this next year these things are going to happen and if you look at the CEOs of the DRAM industry the number one DRAM in the world was DDR memory for CPUs in data centers. About three years ago, I was able to convince several of the CEOs that even though at the time HBM memory was used quite scarcely, you know, and barely by supercomputers, that this was going to be a mainstream memory for data centers in the future. And at first it sounded ridiculous, but several of the CEOs believed me and decided to invest in building HBM memories. Another memory was rather odd to put into a data center is the low power memories that we use for cell phones. And we wanted them to adapt them for supercomputers in the data center. And they go, cell phone memory for supercomputers. And I explained to them why. Well, look at these two memories, LPDDR5, HBM4. The volumes are so incredible. All three of them had record years in history. And these are 45 year old companies. And so, you know, that's part of my job is to inform and shape, inspire, you know.

所以你不仅是在塑造未来，激励英伟达的工程师们。你还在塑造未来的供应链。所以你在与台积电、ASML 对话，上游、下游，上游、下游。就是这样。

So, you're not just manifesting the future and maybe inspiring Nvidia, the different engineers of the company. You're manifesting the supply chain of the future. So you're having conversations with TSMC, with ASML, upstream, downstream, upstream, downstream. So that's the thing.

GEV、卡特彼勒。是的，那是我们的下游。没错。

GEV, Caterpillar. Yeah, that's downstream from us. Yeah. There you go.

是的，整个体系。但整个半导体行业有那么多极其困难的工程。供应链的复杂程度、组件的数量令人感到可怕，但它 somehow 运转着。

Yeah. The whole thing. I mean, but that's so there's so much incredibly difficult engineering that happens in the entire semiconductor industry. And it's just feels scary how intricate the supply chain is, how many components there are, but it works somehow.

正是如此。深厚的科学、深厚的工程、不可思议的制造，而且很多制造已经实现机器人化，但我们有几百家供应商为我们的 130 万个组件的机架提供技术。嗯。每个机架有 130 万到 150 万个组件。Vera Rubin 机架有 200 家供应商。

Exactly. The deep science, the deep engineering, the incredible manufacturing, and so much of the manufacturing is already robotics, but we have a couple of hundred suppliers that contribute the technology that goes into our 1.3 million component rack. Mhm. Each rack is 1.3 one and a half million components. There are 200 suppliers across the Vera Rubin rack.

有趣的是，你没有把那个列为让你夜不能寐的障碍。

So, it's interesting that you don't list that as the thing that keeps you up at night in the list of blockers.

但我正在做所有必要的事情

But I'm doing all the things necessary to

好的。

Okay.

你看，我可以安心睡觉，因为我已经处理好了。我说：‘好吧，我可以睡觉了，然后我想，让我们推理一下。什么对我们重要？’因为让我们推理一下，因为我们把系统架构从你记得的原始 DGX1 改成了 MVLink 72 机架级计算。嗯。这意味着什么？对软件意味着什么？对工程意味着什么？对我们设计和测试的方式意味着什么？对供应链意味着什么？其中一个含义是，我们将数据中心的超级计算机集成转移到了供应链中的超级计算机制造。嗯。

See, I can go to sleep because I checked it off. I said, 'Okay, you know, I can go to sleep and I go, well, let's see what reason about this. What's important for us?' Because okay let's reason about this because we changed the system architecture from the original DGX1 that you remembered to MVLink 72 rack scale computing. Mhm. What does that mean? What does that mean to software? What does that mean to engineering? What does that mean to how we design and test and what does that mean to the supply chain? Well, one of the things that it meant was we moved supercomputer integration at the data center into supercomputer manufacturing in the supply chain. Mhm.

如果你这样做，你还必须认识到，假设你想同时运行 50 吉瓦的超级计算机，而制造这些超级计算机需要一周时间，那么供应链中每周的超级计算机就需要 1 吉瓦的电力。因此，我们需要供应链增加电力，以便在发货前建造和测试这些超级计算机。

If you're doing that, you also have to recognize you're going to move one and if the total footprint of whatever data center you're going to build, let's say you would like to have 50 gigawatts of supercomputers running simultaneously and it takes one week to manufacture that 50 gigawatts of supercomputers. Then each week in the supply chain, the supercomputers are going to need a gigawatt of power. So we're going to need the supply chain to increase the amount of power it has to build and test the supercomputers before I ship it.

嗯，MVLink72 实际上是在供应链中构建超级计算机，然后每个机架重达两三吨地发货。以前它们是分部件运来，我们在数据中心内部组装。但现在不可能了，因为 MVLink72 密度太高。这就是一个例子。我会飞到供应链中，会见我的合作伙伴，然后说，猜猜看？我们要这样做。这是我们以前建造 DGX 的方式。我们现在要这样建造。这样会好得多，因为我们需要它们用于推理。推理市场即将到来。推理的转折点即将到来。这将是一个巨大的市场。所以我先向他们解释发生了什么，为什么会发生，然后我要求他们各自进行数十亿美元的资本投资。因为他们信任我，我也非常尊重他们。我给他们充分的机会质疑我，我花时间向人们解释事情，并用基本原理推理。我给他们画图，从第一性原理推理。等我说完，他们就知道该做什么了。

Well, MVLink72 literally builds supercomputers in the supply chain and ships them two to three tons at a time per rack. It used to come in parts and we used to assemble them inside the data center. But that's impossible now because MVLink72 is so dense. And that's an example. I would have to go into the supply chain, meet my partners, and say, guess what? Here's what we're going to do. This is the way we used to build our DGXs. We're going to build them this way. This is going to be so much better because we're going to need them for inference. The market for inference is coming. The inflection point for inference is coming. It's going to be a big market. So I first explain to them what's going on, why it's going to happen, and then I ask them to make several billion dollars of capital investments each. Because they trust me, and I'm very respectful of them. I give them every opportunity to question me, and I spend time to explain things to people and reason about it. I draw them pictures and reason about it from first principles. By the time I'm done with them, there's no question what to do.

所以这很大程度上关乎关系，以及建立对未来的共同看法。

So it's a lot about relationships and building a shared view of the future.

是的。

Yeah.

但你担心某些瓶颈吗？我的意思是，供应链中最大的瓶颈是什么？你担心 ASML 的 V 型工具吗？你担心台积电的封装，以及它扩展的速度吗？就像你说的，你不仅增长得非常快，而且还在加速增长。所以感觉供应链中的每个人，那些肯定是瓶颈，都必须扩大规模。

But do you worry about certain bottlenecks? I mean what are the biggest bottlenecks in the supply chain? Are you worried about ASML V tooling? Are you worried about the packaging of TSMC, how fast it could scale? Like you said, you're not only growing incredibly fast, you're accelerating growth. So it feels like everybody in the supply chain and those are certainly bottlenecks would have to scale up.

你跟他们谈过如何更快地扩大规模吗？你担心吗？

Are you having conversations with them like how can you scale up faster? Do you worry about it?

不担心。

No.

好的。

Okay.

因为我告诉了他们我需要什么，他们理解了我的需求。他们告诉了我他们打算做什么，我相信他们会做到。

Because I told them what I needed, they understood what I need. They told me what they're going to do and I believe in what they're going to do.

有意思。听到这个很好。那么，也许我们可以再谈谈电力问题。你对解决能源问题有什么期望？我想讨论并传达的一个领域是：我们的电网是为最坏情况设计的，并留有余量。但 99%的时间我们远未达到最坏情况，因为最坏情况是冬季的几天、夏季的几天和极端天气。大多数时候我们远未达到最坏情况，可能只运行在峰值的 60%左右。所以 99%的时间电网都有过剩电力，闲置在那里。但必须闲置，以防万一，医院需要供电，基础设施需要供电，机场需要运行，等等。所以我的问题是，我们能否帮助他们理解，并创建合同协议，设计计算机架构系统和数据中心，使得当社会基础设施需要最大电力时，数据中心能减少用电。

Interesting. That's great to hear. So maybe if we can just linger on the power for a little bit. What are your hopes for how to solve the energy problem? One of the areas that I would love us to talk about and just get the message out. Our power grid is designed for the worst case condition with some margin. Well, 99% of the time we're nowhere near the worst case condition because the worst case condition is a few days in the winter, a few days in the summer and extreme weather. Most of the time we're nowhere near the worst case condition and we're probably running around 60% of peak. So 99% of the time our power grid has excess power and it's just sitting idle. But it has to be there sitting idle just in case when the time comes hospitals have to be powered, infrastructure has to be powered, airports have to run, and so on. So the question I have is whether we could go and help them understand and create contractual agreements and design computer architecture systems, data centers such that when they need the maximum power for infrastructure in society, the data centers would get less.

但无论如何，这是非常罕见的情况。在那段时间里，我们要么有备用发电机来应对那一点需求，要么让计算机将工作负载转移到别处，要么让计算机运行得更慢。我们可以降低性能，减少功耗，并在有人请求答案时提供稍长的延迟响应。所以我认为，这种使用计算机、建设数据中心的方式，而不是期望 100%的正常运行时间和那些非常严格的合同，给电网带来了很大压力，迫使它们从最大值增加。我只想利用它们的过剩电力。那些电力就闲置在那里。

But that's a very rare instance anyway. And during that time, we either have our backup generator for that little part of it, or we just have our computers shift the workload somewhere else, or we have the computers just run slower. We could degrade our performance, reduce our power consumption and provide for slightly longer latency response when somebody asks for an answer. So I think that way of using computers, of building data centers instead of expecting 100% uptime and these contracts that are really quite rigorous, it's putting a lot of pressure on the grid to be able to increase from their maximum. I just want to use their excess. It's just sitting there.

是的。这一点讨论得不够。那么是什么阻碍了这一点？是监管吗？是官僚主义吗？

Yeah. That's not talked about enough. So what's stopping this? Is it regulation? Is it bureaucracy?

我认为这是一个贯穿始终的问题。它始于最终客户。最终客户对数据中心提出要求，要求它们永远不能不可用。所以最终客户期望完美。为了提供这种完美，你需要备用发电机和电网电力供应商共同实现完美。所以每个人都必须达到六个九的可靠性。

I think it's a throughway problem. It starts with the end customer. The end customer puts requirements on the data centers that they can never not be available. So the end customer expects perfection. Now in order to deliver that perfection, you need a combination of backup generators and your grid power supplier to deliver on perfection. So everybody's got to have 69s.

嗯，首先，我认为现在我们应该让每个人都明白，当客户提出这些要求时，你的数据中心运营团队中有人与 CEO 脱节。我打赌 CEO 不知道这一点。我要和所有 CEO 谈谈。CEO 们可能根本没有注意正在签署的合同。所以每个人都想签最好的合同，他们去找云服务提供商，两个合同谈判者——我现在就能看到他们——在谈判这些多年合同。双方都想要最好的合同。结果，云服务提供商不得不去找公用事业公司，他们期望六个九的可靠性。所以我认为第一件事就是确保所有客户，客户的 CEO 们意识到他们在要求什么。第二件事是，我们必须建设能够优雅降级的数据中心。所以如果电网的公用事业公司告诉我们，听着，我们得把你的用电量降到 80%左右，我们会说完全没问题。

Well, I think first of all, right now, we ought to have everybody understand that when the customer asks for these things, you have somebody in your data center operations team disconnected from the CEO. I bet the CEO doesn't know this. I'm going to talk to all the CEOs. The CEOs are probably not paying any attention to the contracts that are being signed. So everybody wants to sign the best contract of course and they go down to the cloud service providers and the two contract negotiators that are you I could just see them now negotiating these multi-year contracts. Both sides want the best contract. As a result, the CSPs then have to go down to the utilities and they expect the nine 69s. So I think the first thing is just make sure that all the customers, the CEOs of the customers realize what they're asking for. Now, the second thing is we have to build data centers that gracefully degrade. So if the utility of the grid tells us, listen, we're going to have to back you down to about 80%, we're going to say that's no problem at all.

嗯。

Mhm.

我们只需移动工作负载。我们会确保数据永不丢失，但我们可以降低计算速率，使用更少的能源。

We're just going to move our workload around. We're going to make sure that data is never lost, but we can reduce the computing rate and use less energy.

关键工作负载的服务质量会略有下降，我立即将其转移到别处，所以我没有这个问题。因此，无论哪个数据中心仍能保持 100%的正常运行时间。那么，数据中心的智能动态电力分配在工程上有多难？

The quality of service degrades a little bit for the critical workloads I shift that somewhere else right away so I don't have that problem and so you know whoever whichever data center still has 100% uptime and so how difficult of an engineering problem is that the smart dynamic allocation of power in the data center?

只要你能明确规格，就能完美地设计出来。只要它遵循物理定律的第一性原理，我认为就没问题。

As soon as you could specify, you could engineer it beautifully. Put so long as it obeys the laws of physics on first principles, I think we're good.

你提到的第三件事是什么？嗯，第二件事是数据中心……

What was the third thing you were mentioning? Um, so the second thing is the data centers...

第三件事是，我们需要公用事业公司也认识到这是一个机会。与其说‘看，我需要 5 年才能提升电网容量’，如果你愿意接受这种保证级别的电力，我下个月就能以这个价格提供给你。因此，如果公用事业公司也提供更多分段的电力交付承诺，那么我认为每个人都会知道该怎么利用它。是的，但目前电网中的浪费太多了。我们应该解决这个问题。

And the third thing is we need the utilities to also recognize that this is an opportunity. And instead of saying, 'Look, it's going to take me 5 years to increase my grid capability,' if you are willing to take power of this level of guarantee, I can make them available for you next month at this price. And so if utilities also offered more segments of power delivery promises, then I think everybody will figure out what to do with it. Yeah. But there's just way too much waste in the grid right now. We should go after it.

你高度赞扬了埃隆和肖在孟菲斯建造巨像超级计算机的成就，可能创纪录地只用了 4 个月。现在它拥有 20 万块 GPU，并且增长非常迅速。你能谈谈他的方法中有什么值得所有数据中心创建者借鉴的吗？他的工程方法、整个建设管理方法，等等。

You've highly lauded Elon and Xi's accomplishment in Memphis in building Colossus Supercomputer probably in record time in just 4 months. It's now at 200,000 GPUs and growing very quickly. Is there something that you could speak to understand about his approach that's instructive broadly to all the data center creators that enabled that kind of accomplishment? His approach to engineering, his approach to the whole management of construction, everything.

首先，埃隆在众多不同领域都有很深的造诣，同时他也是一个非常出色的系统思考者。因此，他能够跨学科思考。他显然会推动事情，质疑一切：第一，这是否必要；第二，是否必须这样做；第三，是否必须花这么长时间。他有能力质疑一切，直到每个部分都精简到必要的最小量。你无法再移除任何东西，但产品的必要功能仍然保留。他是你能想象到的最极简主义者，而且是在系统规模上做到的。我还喜欢他亲临现场这一点。他会直接去那里，如果有问题，他就会去那里，让我看问题所在。当你把所有这些结合起来，就能克服很多‘我们一直这样做’的惯性。我在等他们，我的意思是，每个人都有很多借口。最后一点是，当你个人以如此紧迫感行动时，会促使其他人也紧迫起来。每个供应商都有很多客户和项目在进行。他让自己的项目成为别人项目的最高优先级。他通过实际行动来做到这一点。

First of all, Elon is deep in so many different topics, yet he's also a really good systems thinker. And so he's able to think through multiple disciplines. He obviously pushes things, questions everything: whether number one, is it necessary; number two, does it have to be done this way; and number three, does it have to take this long. And so he has the ability to question everything to the point where everything is down to its minimal amount that's necessary. You can't take anything else out, and yet the necessary capabilities of the product retain. He is as minimalist as you could possibly imagine, and he does it at a system scale. I also love the fact that he is present at the point of action. He'll just go there and if there's a problem, he'll just go there and show me the problem. When you do all of this in combination, you overcome a lot of 'this is just the way we do it.' I'm waiting for them. I mean, just everybody has a lot of excuses. And then the last thing is, when you act personally with so much urgency, it causes everybody else to act with urgency. Every supplier has a lot of customers and projects going on. He makes it his business that he's the top priority of everybody else's projects. And he does that by demonstrating it.

是的。我参加过很多这样的会议。看着很有趣，因为真的没有足够多的人会问这样的问题：‘好吧，那这件事能不能快很多？怎么做到？为什么必须花这么长时间？’

Yeah. I've been in a bunch of those meetings. It's fun to watch because really not enough people ask the question like, 'Okay, so can this be done a lot faster and how? Why does it have to take this long?'

这往往就变成了一个工程问题。是的，我认为当你获得真实情况时……我记得有一次我和他在一起，他实际上在仔细检查将电缆插入机架的整个过程，他和一名正在执行该任务的现场工程师一起工作，试图理解这个过程是什么样的，以便减少出错的可能性。通过从构建数据中心的每一个任务中积累这种直觉，你立即就能在细节层面和整体系统层面感受到效率低下的地方，从而可以不断提高效率。此外，你还有一个大锤子，可以说：‘我们完全换一种方式来做。’

And then that becomes an engineering question often. And yes, I think when you get the ground truth of actually... I remember one of the times I was hanging out with him, he literally is going through the entire process of how to plug in cables into a rack, and he was working with an engineer on the ground that's doing that task, and he's just trying to understand what that process looks like so it can be less error-prone. And just building up that intuition from every single task involved in putting together the data center, you start to immediately get a sense at the detailed scale and at the broad system scale of where the inefficiencies are, and so you can make it more and more efficient. Plus, you have the big hammer of being able to say, 'Let's do it totally different.'

并移除所有可能的障碍。

And remove all possible blockers.

没错。

That's right.

在埃隆处理系统工程的方式中，你是否看到了与英伟达极端系统代码设计方法相似之处？

Is there parallels in the NVIDIA extreme systems code design approach that you see in the way Elon approaches systems engineering?

嗯，首先，代码设计是一个终极系统工程问题。因此，我们从第一性原理出发来处理我们的工作。我们做的另一件事，是一种哲学、一种心态、一种我 30 年前开始的方法，叫做‘光速’。光速不仅仅关乎速度。光速是我对‘物理定律的极限是什么？’的简称。因此，我们所做的每一件事都与光速进行比较：内存速度、数学速度、功耗成本、时间、精力、人数、制造周期。当你考虑延迟与吞吐量、成本与吞吐量、成本与容量等所有这些因素时，你分别针对光速进行测试，以满足这些不同的约束。然后当你综合考虑时，你必须做出权衡，因为实现极低延迟的系统与实现极高吞吐量的系统在架构上是根本不同的。但你想知道一个实现高吞吐量的系统的光速是多少？一个实现低延迟的系统的光速是多少？然后当你考虑整个系统时，你可以做出权衡。

Well, first of all, the code design is an ultimate systems engineering problem. And so, we approach the work that we do from that first principle. The other thing that we do, and this is a philosophy, a state of mind, a method that I started 30 years ago, and it's called the speed of light. The speed of light is not just about the speed. Speed of light is my shorthand for 'what's the limit of what physics can do?' And so every single thing that we do is compared against the speed of light: memory speed, math speed, power cost, time, effort, number of people, manufacturing cycle time. And when you think about latency versus throughput, when you think about cost versus throughput, cost versus capacity, all of these things, you test against the speed of light to achieve all of these different constraints separately. And then when you consider it together, you have to make compromises because a system that achieves extremely low latency versus a system that achieves very high throughput are architected fundamentally differently. But you want to know what's the speed of light of a system that achieves high throughput? What's the speed of light of a system that achieves low latency? And then when you think about the total system, you can make trade-offs.

所以我强迫每个人在做事之前先思考第一性原理，思考一切事物的物理极限，并用这个标准去检验一切。这是一种很好的思维方式。我不喜欢另一种方法，也就是持续改进。持续改进的问题在于，首先你应该从第一性原理出发，以光速思考的速度进行工程设计，只受物理极限和物理学限制。在那之后，当然你可以随着时间的推移不断改进。但我不喜欢遇到一个问题时，有人说：‘嘿，现在做这个需要 74 天，我们可以帮你缩短到 72 天。’我宁愿把它完全归零。首先，向我解释为什么一开始是 74 天，然后想想今天可能做到什么。如果我从头开始构建，需要多长时间？很多时候你会惊讶地发现，可能只需要 6 天。从 6 天到 74 天的差距可能是合理的妥协、成本削减等各种因素造成的，但至少你知道了它们是什么。而当你知道了 6 天是可能的，从 74 天到 6 天的对话就会出奇地有效。

And so I force everybody to think about what's the first principles, the physical limits for everything before we do anything, and we test everything against that. So that's a good frame of mind. I don't love the other methods, which is continuous improvement. The problem with continuous improvement is, first of all, you should engineer something from first principles at the speed of light thinking, limited only by physical limits and physics limits. After that, of course, you would improve it over time. But I don't like going into a problem and somebody says, 'Hey, you know, it takes 74 days to do this today, and we can do it for you in 72 days.' I rather strip it all back to zero. First of all, explain to me why it's 74 in the first place, and let's think about what's possible today. If I were to build it completely from scratch, how long would it take? Often times you'd be surprised and might come to 6 days. Now the rest of the 6 days to 74 could be very well-reasoned compromises, cost reductions, and all kinds of different things, but at least you know what they are. And now that you know that 6 days is possible, the conversation from 74 to 6 is surprisingly much more effective.

在你处理的这些极其复杂的系统中，简单性有时是一个好的启发式方法吗？我的意思是，你发布的 Vera Rubin pod 简直不可思议。我们说的是 7 个芯片、7 种芯片类型、5 种专用机架类型、40 个机架、1.2 千万亿个晶体管、近 2 万个 Nvidia 芯片、超过 1100 个 Rubin GPU、60 exaflops 算力、每秒 10 PB 的规模带宽。这还只是一个 pod。就一个 pod。而单单 NVL72 机架就有 130 万个组件、1300 个芯片、4000 磅重，塞进一个 19 英寸宽的机架里。Lex，我们大概每周会生产大约 200 个这样的 pod，让你对这么多不同组件有个概念。我想简单是不可能的，但这是你在设计时追求的一个指标吗？

In such incredibly complex systems that you're working with, is simplicity sometimes a good heuristic to reach for? I mean, the Vera Rubin pod that you announced is just incredible. We're talking about seven chips, seven chip types, five purpose-built rack types, 40 racks, 1.2 quadrillion transistors, nearly 20,000 Nvidia dies, over 1,100 Rubin GPUs, 60 exaflops, 10 petabytes per second of scale bandwidth. That's all just one pod. That's just one pod. And even the NVL72 rack alone is 1.3 million components, 1,300 chips, 4,000 pounds crammed into a single 19-inch wide rack. And Lex, we'll probably crank out about 200 of these pods a week just to put in perspective the amount of different components. I suppose simplicity is impossible, but is that a metric that you kind of reach for in trying to design things?

我最常用的一句话是：我们需要事物在必要的情况下尽可能复杂，但又要尽可能简单。问题是：所有这些复杂性都是必要的吗？我们应该检验这一点，挑战这一点。除此之外，多余的东西都是不必要的。

The phrase that I use most often is: we need things to be as complex as necessary but as simple as possible. The question is: is all that complexity there necessary? We ought to test for that and challenge that. And after that, everything else above it is gratuitous.

但广义上来说，这是半导体行业最令人惊叹的成就之一，而 Nvidia 正在做的事情是历史上最伟大的工程之一。这些系统确实是工程学的奇迹。

But it's some of the most incredible semiconductor industry broadly, but what Nvidia is doing is some of the greatest engineering in history. These systems are truly marvels of engineering.

这是世界有史以来制造的最复杂的计算机。

It is the most complex computer the world has ever made.

是啊，工程团队。我的意思是，我不知道这是不是一场竞赛，但如果这是工程团队的奥运会，台积电做了不可思议的工程，ASML 在各个尺度上都很出色，但 Nvidia 会让他们面临激烈竞争。真是不可思议的团队，每个项目都是金牌得主，全部聚集在这里，必须协同工作，直接向你汇报。这太棒了。

Yeah, the engineering teams. I mean, I don't know if it's a competition, but if it were like an Olympics of engineering teams, TSMC does incredible engineering, ASML at every scale, but Nvidia is going to give them a run for their money. Just incredible teams, gold medalists in every single sport, all assembled right here and have to work together and report directly to you. This is wonderful.

你最近去了中国。所以很有意思问你：中国在建设科技产业方面取得了巨大成功。你如何理解中国在过去 10 年里能够建立这么多令人难以置信的世界级公司、世界级工程团队，以及这个产生如此多惊人产品的技术生态系统？

You've recently traveled to China. So it's interesting to ask you: China's been incredibly successful in building up its technology sector. What do you understand about how China is able to, over the past 10 years, build so many incredible world-class companies, world-class engineering teams, and just this technology ecosystem that produces so many incredible products?

原因有很多。首先，从一些事实开始：全世界 50%的 AI 研究人员是中国人，上下浮动，而且他们大多仍在中国。我们这里有很多，但中国仍有出色的研究人员。他们的科技产业出现在一个非常恰当的时机，即移动云时代。他们的贡献方式是软件。这是一个拥有惊人科学和数学水平的国家，孩子们教育非常好。他们的科技产业是在软件时代创建的，他们对现代软件非常熟悉。中国不是一个单一的经济大国，它有许多省份和城市，市长们相互竞争。这就是为什么有这么多电动汽车公司、这么多 AI 公司、这么多你能想到的各种公司。它们都创造了一些公司，结果内部竞争极其激烈，最终存活下来的是了不起的公司。他们还有一种社会文化：家庭第一，朋友第二，公司第三。他们之间的交流量本质上一直是开源的。他们对开源贡献更多是非常合理的，因为他们可能在想：‘我们在保护什么？’我的工程师，他们的兄弟在那家公司，他们的朋友在那家公司，他们都是校友。校友的概念：一个校友就是你一生的兄弟。他们分享知识非常快，所以没有隐藏技术的必要，不如把它开源。开源社区随后放大并加速了创新过程。因此，你得到了快速、极其优秀的人才，由于开源、朋友之间的本质以及公司间的激烈竞争，创新迅速。最终涌现出令人惊叹的东西。这是当今世界上创新最快的国家。我刚才所说的一切都基于孩子们的成长方式：优秀的教育、父母希望他们学业有成、他们的文化。这些都是他们国家的特点，而他们恰好出现在技术呈指数级发展的时刻。

A whole bunch of reasons. First of all, let's start with some facts: 50% of the world's AI researchers are Chinese, plus or minus, and they're mostly in China still. We have many of them here, but there are amazing researchers still in China. Their tech industry showed up at precisely the right time, at the time of the mobile cloud era. Their way of contributing was software. This is a country with incredible science and math, really well-educated kids. Their tech industry was created during the era of software. They're very comfortable with modern software. China is not one giant economic country; it's got many provinces and cities with mayors all competing with each other. That's the reason why there are so many EV companies, so many AI companies, so many of every company you could imagine. They all create some of them, and as a result, they have insane competition internally, and what remains is an incredible company. They also have a social culture where it's family first, friends second, and company third. The amount of conversation that goes back and forth between them is essentially open source all the time. The fact that they contribute more to open source is so sensible because they're probably thinking, 'What are we protecting?' My engineers, their brothers are in that company, their friends are in that company, and they're all schoolmates. The schoolmate concept: one schoolmate is your brother for life. They share knowledge very quickly, so there's no sense keeping technology hidden; you might as well put it on open source. The open source community then amplifies and accelerates the innovation process. So you get this rapid, incredibly great talent, rapid innovation because of open source and the nature of friends and insane competition among companies. What emerges is incredible stuff. This is the fastest innovating country in the world today. Everything that I've just said is fundamental to how the kids were grown: excellent education, parents wanting them to do well in school, their culture. These are just the things about their country, and they showed up at precisely the time when technology is going through that exponential.

另外，在文化上，当工程师很酷。这与你提到的所有因素都相关。

Plus, culturally, it's pretty cool to be an engineer. It connects to all the components that you're mentioning.

这是一个建设者的国家。

It's a builder nation.

这是一个建设者的国家。

It's a builder nation.

是的，这是一个建设者的国家。我们国家的领导人很了不起，但大多是律师。他们努力保护我们的安全，维护法治和治理。而他们的国家是从贫困中建设起来的，所以大多数领导人是杰出的工程师，一些最聪明的人。

Yeah, it's a builder nation. Our country's leaders, incredible, but they're mostly lawyers. They're trying to keep us safe, rule of law, governing. Their country was built out of poverty, and so most of their leaders are incredible engineers, some of the brightest minds.

稍微岔开一下话题，因为你提到了开源，我必须提到 Perplexity，你长期以来一直是它的粉丝。

To take a small tangent because you mentioned open source, I have to go to Perplexity here, who you have been a fan of a long time.

我很喜欢。是的。

I love it. Yeah.

感谢你开源了 Nemotron 3 Super，也可以在 Perplexity 内部用来查找资料。

And thank you for releasing open source Nemotron 3 Super, which you can also use inside Perplexity to look stuff up.

是的。

Yeah.

这是一个 1200 亿参数的开源权重模型。你对开源有什么愿景？你提到了中国的 DeepSeek、MiniMax 等公司都在推动开源 AI 运动，而英伟达在接近最先进水平的开源大语言模型方面确实领先。你的愿景是什么？

Which is a 120 billion parameter open weight model. What's your vision with open source? So you mentioned China with DeepSeek, with MiniMax, all these companies really pushing forward the open-source AI movement, and Nvidia is really leading the way in close to state-of-the-art open source LLMs. What's your vision there?

首先，如果我们想成为一家伟大的 AI 计算公司，就必须了解 AI 模型是如何演进的。我喜欢 Nemotron 3 的一点是，它不仅仅是一个纯 Transformer 模型，而是 Transformer 和 SSM 的结合。我们早期开发了条件 GAN，它逐步引向了扩散模型。因此，我们在模型架构和不同领域进行基础研究，这让我们能够洞察什么样的计算系统适合未来的模型，这是我们极致协同设计策略的一部分。其次，我认为我们正确地认识到，一方面我们希望拥有世界级的模型作为产品，它们应该是专有的；另一方面，我们也希望 AI 能渗透到每个行业、每个国家、每个研究人员和每个学生。如果一切都是专有的，就很难进行研究，也很难在此基础上创新。因此，开源对于许多行业加入 AI 革命来说是根本必要的。英伟达拥有规模、技能和动机，只要我们在，就会持续构建这些 AI 模型。所以我们应该这样做。我们可以开放，可以激活每个行业、每个研究人员、每个国家，让他们加入 AI 革命。还有第三个原因：认识到 AI 不仅仅是语言。这些 AI 可能会使用在其他信息模态上训练的工具、模型和子智能体。可能是生物学、化学、物理定律、流体和热力学，这些并不都是语言结构。所以必须有人确保天气预报、生物学 AI、物理 AI 等都能被推向极限和前沿。我们不造车，但希望每个汽车公司都能获得优秀的模型。我们不发现药物，但希望礼来拥有世界上最好的生物学 AI 系统，用于发现药物。所以这三个根本原因：认识到 AI 不仅仅是语言，AI 非常广泛，我们希望让每个人都参与到 AI 世界中，以及 AI 的协同设计。

First, if we're going to be a great AI computing company, we have to understand how AI models are evolving. One of the things that I love about Nemotron 3 is it's not just a pure Transformer model. It's Transformer and SSM. And we were early in developing the conditional GANs, which led step by step to diffusion. So the fact that we're doing basic research in model architecture and in different domains gives us visibility into what kind of computing systems would do a good job for future models, and so it is part of our extreme co-design strategy. Second, I think we rightfully recognize that on the one hand we want world-class models as products and they should be proprietary. On the other hand, we also want AI to diffuse into every industry and every country, every researcher, every student. And if everything is proprietary, it's hard to do research and it's hard to innovate on top of it. So open source is fundamentally necessary for many industries to join the AI revolution. Nvidia has the scale and we have the motives to not only skills, scale and motivation to build and continue to build these AI models for as long as we shall live. So therefore, we ought to do that. We can open up, we can activate every industry, every researcher, every country to be able to join the AI revolution. There's a third reason: recognizing that AI is not just language. These AIs will likely use tools and models and sub-agents that were trained on other modalities of information. Maybe it's biology or chemistry or laws of physics or fluids and thermodynamics, and not all of it is in language structure. So somebody has to go make sure that weather prediction, biology AI, physical AI, all of that stuff can be pushed to the limits and pushed to the frontier. We don't build cars, but we want to make sure every car company has access to great models. We don't discover drugs, but I want to make sure that Lilly has the world's best biology AI systems so that they can use it for discovering drugs. So these three fundamental reasons: recognizing that AI is not just language, that AI is really broad, that we want to engage everybody into the world of AI, and also co-design of AI.

好吧，我必须再次感谢你真正开源了 Nemotron 3。

Well, I have to say once again, thank you for open sourcing really truly open sourcing Nemotron 3.

是的，感谢你提到我们开源了模型、权重、数据以及创建方法。

Yeah, I appreciate you saying that we open source the models, we open source the weights, we open source the data, we open source how we created it.

是的，这非常了不起，真的令人难以置信。你来自台湾，与台积电关系密切。所以我必须问：台积电在工程团队和令人难以置信的工程工作方面是一家传奇公司。你对台积电的文化和方法的理解是什么，能解释他们如何在半导体领域取得如此独一无二的成功？

Yeah, it's pretty amazing. It's really incredible. You're originally from Taiwan and have a close relationship with TSMC. So I have to ask: TSMC is a legendary company in terms of the engineering teams and the incredible engineering work they do. What do you understand about TSMC culture and their approach that explains how they achieve this singular unmatched success in everything they're doing with semiconductors?

首先，对台积电最大的误解是认为他们只有技术，好像他们有一个非常好的晶体管，如果有人拿出另一个晶体管，游戏就结束了。实际上是技术，当然我指的不只是晶体管，还有金属化系统、封装、3D 封装、硅光子学等等他们拥有的所有技术。这些技术确实让这家公司与众不同。但他们协调全球数百家公司动态需求的能力——当这些公司需求上升、下降、推进、推迟、客户间变化、晶圆启动、停止、紧急启动，所有这些随着世界不断变化的复杂性——他们却能以高吞吐量、高良率、极好的成本和出色的客户服务运营工厂。他们认真对待工作，认真对待承诺。当承诺晶圆交付时，晶圆就会到，这样你就能正常运营公司。他们的制造系统堪称奇迹。第二点是他们的文化。这种文化一方面专注于技术，推动技术进步，另一方面同时以客户服务为导向。很多公司非常注重客户服务，但技术并不卓越，不在技术前沿；或者很多公司处于技术前沿，但客户服务不是最好的。而他们 somehow 平衡了这两点，两者都做到了世界级。第三点可能是我最看重的：他们创造了一种叫做信任的无形资产。我信任他们，把我的公司放在他们之上。这非常重要。你们之间建立了非常紧密的关系，这种信任基于多年的表现，但也涉及人际关系。三十年，我不知道通过他们做了多少成百上千亿美元的生意，而且我们没有合同。这非常了不起。

First of all, the deepest misunderstanding about TSMC is that their technology is all they have, that somehow they have a really great transistor and if somebody shows up another transistor, game over. It's the technology, and of course I don't mean just the transistor, the metallization systems, the packaging, the 3D packaging, the silicon photonics, all of the technology they have. That technology is really what makes the company special. But their ability to orchestrate the dynamic demands of hundreds of companies in the world as they're moving up, shifting out, increasing, decreasing, pushing out, pulling in, changing from customer to customer, wafer starting, wafer stopping, emergency wafer starts, all of this dynamics of the world's complexity as the world is shapeshifting all the time, and somehow they're running a factory with high throughput, high yields, really great costs, excellent customer service. They take their work seriously, they take their promises seriously. When your wafers were promised to show up, the wafers show up, so that you could run your company appropriately. Their manufacturing system is completely miraculous. Then the second thing is their culture. This culture is simultaneously technology-focused on one hand, advancing technology, and simultaneously customer service-oriented on the other hand. A lot of companies are very customer service-oriented, but they're not very technology excellent. They're not at the bleeding edge of technology. Or a lot of companies who are at the bleeding edge of technology, but they're not the best customer service-oriented company. Somehow they've balanced these two and they're world class at both. Then probably the third thing is the technology that I most value in them: they created this intangible called trust. I trust them to put my company on top of them. That's a very big deal. There's a really close relationship there that you've established, and that trust is established based on many years of performance. But there are human relationships involved there as well. Three decades. I don't know how many tens, hundreds of billions of dollars of business we've done through them, and we don't have a contract. That's pretty great.

太棒了。好吧。有一个故事说，2013 年台积电创始人张忠谋曾邀请你担任台积电首席执行官。你说你已经有工作了。这个故事是真的吗？

Amazing. Okay. There's a story that in 2013 the founder of TSMC, Morris Chang, offered you the chance to become TSMC's chief executive. And you said you already had a job. Is this story true?

故事是真的。我没有拒绝。是的。但我深感荣幸，当然我当时就知道，现在也知道，台积电是历史上最重要的公司之一。

Story is true. I didn't dismiss it. Yeah. But I was deeply honored, and of course I knew then as I know now TSMC is one of the most consequential companies in history.

是的。

Yeah.

Morris 是我一生中最受尊敬的高管之一，也是我的商业和个人朋友。他邀请我，我感到谦卑和荣幸。但我在这里的工作非常重要。我在脑海中看到了 Nvidia 的未来以及我们能产生的影响。这是非常重要的工作，我唯一的责任就是实现它。所以我拒绝了，不是因为这不是一个绝佳的邀请——这确实是一个难以置信的邀请——而是我实在无法接受。

Morris is one of the highest regarded executives and a business and personal friend that I've had in my life. For him to ask, I was humbled and really honored. But the work that I'm doing here is really important. I've seen in my mind's eye what Nvidia was going to be and the impact that we could have. It was really important work, and it's my sole responsibility to make this happen. So I declined it, not because it wasn't an incredible offer—it's an unbelievable offer—but I simply couldn't take it.

我认为 Nvidia 和台积电是人类文明史上最伟大的两家公司。经营其中任何一家都极其复杂，你必须全身心投入。

I think Nvidia and TSMC are two of the greatest companies in the history of human civilization. Running either one is incredibly complicated; you have to be truly all-in.

是的。

Yeah.

每个层级的人，不仅仅是 CEO，都全身心投入以应对这种复杂性。

Everybody at every scale, not just the CEO level, is truly all-in to accomplish this kind of complexity.

是的。你看，现在我可以帮助两家公司。

Yeah. See, now I can help both companies.

没错。那么 Nvidia 现在是全球市值最高的公司。我必须问一下，Nvidia 最大的护城河是什么？也就是保护你免受竞争的优势。

Exactly. So Nvidia is now the most valuable company in the world. I have to ask, what is Nvidia's biggest moat, the edge that protects you from competition?

我们公司最重要的资产是我们计算平台的安装基数。今天最重要的就是 CUDA 的安装基数。二十年前当然没有安装基数。但为什么它如此特别？如果有人搞出一个 CUDA 或 TUDA，那根本无关紧要。原因在于这从来不仅仅是技术问题。技术当然令人难以置信且富有远见，但关键在于公司对它的投入、坚持和扩展。不是三个人让 CUDA 成功，而是四万三千人让 CUDA 成功，还有数百万开发者相信我们，信任我们会持续推出 CUDA 1、2、3、13，并决定将他们的软件——成堆的软件——移植并构建在 CUDA 之上。所以安装基数是最重要的优势。当这个安装基数与我们执行的速度相结合，达到我们所说的规模——历史上没有公司建造过如此复杂的系统，绝无仅有。而且每年建造一次是不可能的。这种速度加上开发者心中的安装基数，就会占据开发者的心智。从开发者的角度看，如果我明天支持 CUDA，它会好上 10 倍。我平均只需等待 6 个月。不仅如此，如果我在 CUDA 上开发，我能触及数亿人的电脑。我存在于每一个云、每一家电脑公司、每一个行业、每一个国家。所以如果我创建一个开源包并首先放在 CUDA 上，我同时获得这两个属性。不仅如此，我 100%相信 Nvidia 会一直保留 CUDA，维护它、改进它，并持续优化库，直到永远。这一点你可以放心。最后一部分——信任——你把所有这些放在一起。如果我是今天的开发者，我会首先瞄准 CUDA。我会最优先考虑 CUDA。这就是为什么归根结底，这是我们的第一个核心优势。

Our single most important property as a company is the installed base of our computing platform. Our single most important thing today is the installed base of CUDA. Twenty years ago, of course, there was no installed base. But what makes it special? If somebody came up with a CUDA or a TUDA, it wouldn't make any difference at all. The reason is that it's never been just about the technology. The technology was incredible and visionary, but it's the fact that the company was dedicated to it, stuck with it, and expanded its reach. It wasn't three people that made CUDA successful; it was 43,000 people that made CUDA successful, and the several million developers who believed in us and trusted that we would continue to make CUDA 1, 2, 3, 13, and decided to port and dedicate their software on top of it—their mountain of software. So the installed base is the number one most important advantage. That installed base, when amplified with the velocity of our execution at the scale we're talking about—no company in history had ever built systems of this complexity, period. And then to build it once a year is impossible. That velocity combined with the installed base in the developer's mind is just going to take the developer's mind. From the developer's perspective, if I support CUDA tomorrow, it will be 10 times better. I just have to wait 6 months on average. Not only that, if I develop on CUDA, I reach a few hundred million people's computers. I'm in every cloud, every computer company, every single industry, every single country. So if I created an open source package and put it on CUDA first, I get both attributes simultaneously. And not only that, I trust 100% that Nvidia is going to keep CUDA around, maintain it, improve it, and keep optimizing the libraries for as long as they shall live. You could take that to the bank. That last part—trust—you put all that stuff together. If I were a developer today, I would target CUDA first. I would target CUDA most. And that's the reason that in the final analysis is our first core advantage.

我们的第二个优势是我们的生态系统。我们垂直整合了这个极其复杂的系统，但又水平整合到每一家公司的计算机中。我们在 Google Cloud、Amazon、Azure 中。我们现在正疯狂地扩展 AWS。我们在 CoreWeave 和 Nscale 这样的新公司中。我们在 Lilly 的超级计算机中。我们在企业计算机中。我们在边缘的无线基站中。这太疯狂了。一个架构存在于所有这些不同的系统中。我们在汽车、机器人、卫星、太空中。所以这个单一的架构和如此广泛的生态系统，基本上覆盖了世界上每一个行业。

Our second one is our ecosystem. The fact that we vertically integrated this incredibly complex system, but we integrated horizontally into every single company's computers. We're in Google Cloud, Amazon, Azure. We're ramping up AWS like crazy right now. We're in new companies like CoreWeave and Nscale. We're in supercomputers at Lilly. We're in enterprise computers. We're at the edge in radio base stations. It's just crazy. One architecture is in all these different systems. We're in cars, robots, satellites, out in space. So the fact that you have this one architecture and the ecosystem is so broad, it basically covers every single industry in the world.

CUDA 的安装基数未来如何随着 AI 工厂作为护城河而演变？你认为未来的 Nvidia 可能完全围绕 AI 工厂吗？

How does the CUDA installed base evolve into the future with AI factories as a moat? Do you think it's possible that Nvidia of the future is all about the AI factory?

嗯，对我们来说，计算单元曾经是 GPU，然后变成了一台计算机，再变成了一个集群。现在是一个完整的 AI 工厂。当我看到一台计算机，看到 Nvidia 过去建造的东西时，我会想象芯片。当我宣布新产品，比如‘女士们先生们，今天我们宣布 Ampere’，我会拿起芯片。那是我对自己所建造东西的心智模型。今天，拿起芯片仍然很可爱，但它不是我对自己所做事情的心智模型。我的心智模型是这个巨大的千兆瓦级的东西，有发电能力。它连接到电网。它有冷却系统和极其庞大的网络。一万人在里面试图安装它。数百名网络工程师在里面。背后还有数千名工程师试图给它通电。你知道，启动其中一个工厂不是某人说‘现在开了’那么简单。需要数千人才能启动它。所以，当我在思考单个计算单元时，晚上睡觉时，我在想一组机架。是 Pod，而不是单个芯片——整个基础设施。我希望我的下一个飞跃是当我思考建造行星规模的计算时。那将是下一个飞跃。

Well, the unit of computing used to be a GPU to us, then it became a computer, then it became a cluster. Now it's an entire AI factory. When I see a computer, when I see what Nvidia builds in the old days, I would visualize the chip. When I announced a new product, like 'Ladies and gentlemen, we're announcing Ampere today,' I picked up the chip. That was my mental model of what I was building. Today, picking up the chip is still adorable, but it's not my mental model of what I'm doing. My mental model is this giant gigawatt thing that has power generation. It's connected to the grid. It's got cooling systems and networking of incredible monstrosity. 10,000 people are in there trying to install it. Hundreds of networking engineers are in there. Thousands of engineers behind it are trying to power it up. Powering up one of those factories, as you know, is not somebody going, 'It's on now.' It takes thousands of people to bring it up. So mentally, when I'm thinking about a single unit of compute, when I go to bed at night, I'm thinking about a collection of racks. Pods, not individual chips—entire infrastructure. And I'm hoping my next click is when I'm thinking about building computers at planetary scale. That would be the next click.

你怎么看待 Elon 提到的太空角度——在太空中进行计算以解决扩展中的一些能源问题？冷却问题并不容易。

What do you think about the space angle that Elon has talked about—doing compute in space to solve some of the energy issues in terms of scaling? Cooling issues are not easy.

冷却——嗯，这涉及大量的工程复杂性。

Cooling—well, there are a large number of engineering complexities involved with that.

所以 Nvidia 也宣布你们已经在考虑这一点了。

So Nvidia has also announced that you're already thinking about that.

是的，我们已经做到了。Nvidia GPU 是太空中的第一批 GPU。我没意识到这这么有趣。我可能会宣布——也许我们在太空中，我们的一个 GPU 上穿着小宇航服。但我们已经在太空了。这是进行大量成像的正确地方。

Yeah, we're already there. Nvidia GPUs are the first GPUs in space. I didn't realize it was so interesting. I would have declared it—maybe we're in space, with a little astronaut suit on one of our GPUs. But we've been in space. It's the right place to do a lot of imaging.

你知道，因为这些卫星拥有非常高分辨率的成像系统，它们现在正在持续扫描地球。你想要全球范围内连续进行的厘米级成像，这样你就能基本上实时获取一切遥测数据。你不想把这些数据传回地球。那是 PB 级别的数据。你必须在边缘直接进行 AI 处理。扔掉所有不需要的东西。你以前见过的、没有变化的东西，只保留你需要的。所以 AI 应该在边缘进行。显然，如果我们把设备放在极地，可以 24 小时利用太阳能，但没有传导、没有对流，所以基本上只能靠辐射。但太空很大，我想我们会在那里放置巨大的散热器。

You know, because those satellites have really high resolution imaging systems and they're sweeping the Earth continuously now. And you want centimeter scale imaging that is done continuously for the world so that you'll basically have real time telemetry of everything. You don't want to beam that back down to Earth. It's petabytes and petabytes of data. You have to do AI right there at the edge. Throw away everything you don't need. You've seen before, didn't change, and then just keep the stuff that you need. So AI ought to be done at the edge. Obviously we have 24/7 solar if we put it at the poles, but there's no conduction, no convection, so you're pretty much just radiation. But space is big, I guess. We're just going to put big giant radiators out there.

你觉得这个想法有多疯狂？是 5 年后、10 年后还是 20 年后？

How crazy of an idea do you think it is? Is this 5 years out, 10 years out, 20 years out?

所以，我们在讨论 AI Scaling 的阻碍因素。我其实非常务实。我首先寻找下一波机会在哪里。与此同时，我在培育太空领域。我派工程师去研究这个问题。我们开始学到很多。如何处理辐射？如何处理性能下降？如何处理持续的测试和缺陷验证？如何处理冗余和优雅降级？软件方面呢？你怎么看待太空中的软件、冗余和性能？让计算机永不损坏，只是变慢。所以我们可以先进行大量的工程探索，但与此同时，我最喜欢的答案是消除浪费。

So, we're talking about blockers for AI scaling. I'm just so much more practical. I look for where the next bucket of opportunities are first. Meanwhile, I'm cultivating space. I send engineers to go work on the problem. We're starting to learn a lot about it. How do we deal with radiation? How do we deal with degrading performance? How do we deal with continuous testing and attestation of defects? How do we deal with redundancy and degrade gracefully? And what about software? How do you think about software and redundancy and performance out in space? Make it so that the computer never breaks. It just gets slower. So we could start doing a lot of engineering exploration up front, but in the meantime, my favorite answer is eliminate waste.

我们有那么多闲置的算力。我想尽快把它利用起来。

We've got all that idle power. I want to evacuate it as fast as possible.

是的，地球上有很多容易实现的目标，我们可以利用它们进行 AI Scaling。

Yeah, there's a lot of low hanging fruit here on Earth that we can utilize for AI scaling.

你认为英伟达有朝一日可能价值 10 万亿美元吗？我们换个方式问。如果这是真的，世界的未来会是什么样子？

Do you think Nvidia may be worth 10 trillion at some point? Let's ask it this way. What does the future of the world look like where that's true?

我认为英伟达的增长极有可能，在我看来是不可避免的。让我解释原因。我们是历史上最大的计算机公司。仅这一点就应该引发疑问：为什么？原因有两个基础技术原因。第一，计算从基于检索的文件检索系统转变而来。几乎所有东西都是文件。我们预先编写、预先录制、绘制东西，放到网上，放入文件，然后使用推荐系统、一些智能过滤器来决定为你检索什么。所以我们是一个预先录制、人类预先录制和文件检索的系统。这基本上就是计算机的样子。现在 AI 计算机具有上下文感知能力，这意味着它必须实时处理和生成 token。所以我们从基于检索的计算系统转变为基于生成的计算系统。在这个新世界中，我们需要比旧世界多得多的处理能力。在旧世界，我们需要大量存储。在这个新世界，我们需要大量计算。这是第一部分。我们从根本上改变了计算以及计算的方式。唯一会让它倒退的是，如果这种计算方式——生成上下文相关、情境感知、基于新洞察的信息——这种计算密集型的方式如果无效的话。如果在过去 10-15 年研究深度学习的任何时候，我得出了结论说这行不通，是死胡同，或者无法扩展，无法解决这种模态，无法用于这种应用，那么我当然会有不同的感觉。但我认为过去五年给我的信心比之前十年更多。第二个想法：计算机，因为它是一个存储系统，很大程度上是一个仓库。我们现在正在建造工厂。仓库赚不了多少钱。工厂直接与公司的收入相关。所以计算机做了两件事。它不仅改变了做事的方式，它在世界上的目的也改变了。它不再是一台计算机，而是一个工厂。一个用于产生收入的工厂。我们现在看到，这个工厂不仅生产产品、人们想要消费的商品，而且这些商品如此有趣、对这么多不同的受众如此有价值，以至于 token 开始像 iPhone 一样细分。你有免费 token、高级 token，以及中间的各种 token。事实证明，智能是一种可扩展的产品。有极高智能的产品，用于专门领域的 token。人们愿意付费。有人愿意为每百万 token 支付 1000 美元的想法即将成为现实。不是会不会，而是何时。所以我们现在看到，这个工厂生产的商品实际上是有价值的，能够产生收入和利润。现在的问题是，世界需要多少这样的工厂？世界需要多少 token？社会愿意为这些 token 支付多少？如果生产力如此大幅提升，世界经济会发生什么？我们会发现新药、新产品、新服务吗？当你把这些因素结合起来，我绝对确信世界 GDP 将加速增长。我绝对确信，用于计算的 GDP 占比将比过去高出 100 倍，因为它不再是一个存储单元，而是一个产品生成单元。

I think that Nvidia's growth is extremely likely and in my mind inevitable. Let me explain why. We're the largest computer company in history. That alone should beg the question why. The reason is two foundational technical reasons. First, computing went from being a retrieval-based file retrieval system. Almost everything is a file. We pre-write something, pre-record something, draw something, put it on the web, put it in a file, and we use a recommender system, some smart filter to figure out what to retrieve for you. So we were a pre-recording, human pre-recording and file retrieving system. That's what a computer is largely. Now AI computers are contextually aware, which means it has to process and generate tokens in real time. So we went from a retrieval-based computing system to a generative-based computing system. We're going to need a lot more processing in this new world than in the old world. We need a lot of storage in the old world. We need a lot of computation in this new world. That's the first part. We fundamentally changed computing and the way computing is done. The only thing that would cause it to go back is if this way of computing—generating information that's contextually relevant, situationally aware, grounded on new insight before it generates information—this computation-intensive way of doing computing would only go back if it's not effective. If for the last 10-15 years while working on deep learning, at any single moment I would have come to the conclusion that this is not going to work out, it's a dead end, or it's not going to scale, not going to solve this modality, not going to be used in this application, then of course I would feel very differently. But I think the last five years has given me more confidence than the previous 10 years. The second idea: computers, because it was a storage system, was largely a warehouse. We're now building factories. Warehouses don't make much money. Factories directly correlate with a company's revenues. So the computer did two things. Not only did it change the way it did it, its purpose in the world changed. It's no longer a computer, it's a factory. It's a factory used for generation of revenues. We're now seeing not only is this factory generating products, commodities that people want to consume, we're seeing that the commodities are so interesting, so valuable to so many different audiences that the tokens are starting to segment like iPhones. You have free tokens, you have premium tokens, and you have several tokens in the middle. Intelligence, as it turns out, is a scalable product. There are extremely high intelligence products, tokens that are used for specialized things. People will be willing to pay. The idea that somebody's willing to pay $1,000 per million tokens is just around the corner. It's not if, it's only when. So now we're seeing that the commodity that this factory makes is actually valuable and is revenue generating and profit generating. Now the question is how many of these factories does the world need? How many tokens does the world need? How much is society willing to pay for these tokens? And what would happen to the world's economy if productivity were to improve so substantially? Are we going to discover new drugs, new products, new services? When you take these things in combination, I am absolutely certain that the world's GDP is going to accelerate in growth. I'm absolutely certain the percentage of that GDP that will be used for computation will be a 100 times more than the past because it's no longer a storage unit. It's a product generation unit.

所以当你从这个背景来看，然后反推英伟达是什么、英伟达做什么，以及我们需要从这种新经济、新产业中获益多少，我认为我们会变得非常非常大。剩下的问题是，英伟达在不久的将来能否成为一家营收 3 万亿美元的公司？答案当然是肯定的。原因在于它不受任何物理限制。我没有看到任何东西说 3 万亿美元是不可能的。事实上，英伟达的供应链负担由 200 家公司分担，而且我们是在这个生态系统的合作下扩展的。问题是我们是否有足够的能源？我们当然会有。所以所有这些加起来，那个数字只是一个数字。我还记得英伟达第一次突破 10 亿美元的时候。有位 CEO 告诉我：‘黄仁勋，一家优秀的半导体公司理论上不可能超过 10 亿美元。’我不想细说原因，但这当然不合逻辑，而且有很多证据表明我们不是这样。然后有人告诉我：‘黄仁勋，你永远不会超过 250 亿美元，因为某家公司。’有人告诉我你永远不会……所以那些都不是基于第一性原理的思考。简单的思考方式是：我们制造什么，我们能创造多大的机会？英伟达做的不是市场份额生意。我刚才谈到的几乎所有东西都还不存在。

And so when you look at it in that context and then you back into what is Nvidia, what does Nvidia do, and how much of that new economics, new industry would we have to benefit to address, I think we're going to be a lot, lot bigger. And then the rest of it to me is, you go, is it possible for Nvidia to be a $3 trillion revenue company in the near future? The answer is of course yes. And the reason for that is because it's not limited by any physical limits. There's nothing that I see that says $3 trillion is not possible. And as it turns out, Nvidia's supply chain burden is shared by 200 companies, and the fact that we scale out on the backs of, with the partnership of this ecosystem. The question is do we have the energy to do so? And surely we will have the energy to do so. And so all of these things combined, that number is just a number. You know, I still remember Nvidia was the first time we crossed a billion dollars. I was reminded of a CEO who told me, 'Jensen, it's theoretically impossible for a fabulous semiconductor company to exceed a billion dollars.' And I won't bore you with why, but of course it's illogical and there's a lot of evidence we're not. And then somebody told me, 'Jensen, you'll never be more than $25 billion because of some other company.' Somebody told me that you'll never be, you know, because... So those aren't principled first-principle reasoning. The simple way to think about that is what is it that we make and how large is the opportunity that we can create. Now Nvidia is not in the market share business. Almost everything that I just talked about doesn't exist.

嗯。

Mhm.

这就是困难的部分。如果英伟达是一家 100 亿美元的公司，试图抢占英伟达的份额，那么股东很容易看到：哦，如果他们能拿下 10%的份额，就能变得这么大。但人们很难想象我们能有多大，因为我没有可以从谁那里抢份额的对象。

That's the part that's hard. You know, if Nvidia was a $10 billion company trying to take Nvidia's share, then it's easy to see for shareholders that, oh yeah, if they could just take 10% share, they could be this much larger. But it's hard for people to imagine how large we could be because there's nobody I could take share from.

所以我认为这对世界来说是一个挑战——对未来的想象力。但我有足够的时间，我会继续推理，继续谈论它，每一届 GTC 都会变得越来越真实。

You know, and so I think that that's one of the challenges for the world is the imagination of the future. But I got plenty of time and I'll keep reasoning about it and I'll keep talking about it, and every single GTC will become more and more real.

然后越来越多的人会谈论它。总有一天我们会达到的。我百分之百确定我们会达到。

You know, and then more and more people will talk about it. One of these days, we'll get there. But I'm 100% we'll get there.

是的。

Yeah.

这种关于代币工厂的观点，本质上就是每瓦每秒的代币，每个代币都有价值，就像它是一件实际带来价值的东西，它给不同的人带来不同种类、不同数量的价值，但它是价值。这就是实际的产品。可以粗略地理解为代币。所以你有一堆代币工厂，从第一性原理出发很容易想象，考虑到 AI 能解决的所有潜在问题，你将需要指数级更多的代币工厂。

This view of token factories essentially, this token per second per watt, and every token having value, like it's an actual thing that brings value, and it brings different kinds of value, different amounts of value to different people, but it's value. That's the actual product. It could be loosely thought of as the token. And so you have a bunch of token factories, and it's very easy from first principles to imagine a future, given all the potential things that AI can solve, that you're going to need an exponential number more of token factories.

是的。

Yeah.

真正有趣的是，我之所以如此兴奋，是因为代币的 iPhone 已经到来。

And what's really interesting, the reason why I was so excited about it, the iPhone of tokens arrived.

你叫什么？等等，你是说 OpenAI 的 iPhone？

What do you call? Wait, are you saying OpenAI's iPhone?

是的。

Yeah.

这很有趣。呃……

That's interesting. Uh...

智能体。

Agents.

是的，智能体。没错。

Yeah, agents. True.

总的来说是智能体。代币的 iPhone 已经到来。它是历史上增长最快的应用。它直线上升。

Agents in general. The iPhone of tokens arrived. It is the fastest growing application in history. It went straight up.

是的，直线上升。

Yeah, went straight up.

这说明了一些问题。

That says something.

是的。毫无疑问，OpenAI 就是代币的 iPhone。从大约 12 月开始，确实发生了一些非常特别的事情，人们真正意识到了 Claude、代码、OpenAI 的力量。我不好意思承认，在来这里的路上，在机场，我第一次在公共场合这样做——我通过跟笔记本电脑说话来编程，我有点尴尬，因为我假装自己在跟一个人类同事说话。

Yep. There's no question OpenAI is the iPhone of tokens. Yeah, there's something truly special happening from about December where people really woke up to the power of Claude, of code, of OpenAI. I mean, I'm embarrassed to admit that on the way here in the airport, I've, this first time I've done this in public, I was programming, quote unquote, by talking to my laptop, and I was embarrassed because I was pretending like I'm talking to a human colleague.

嗯。

Mhm.

我不确定我对未来人人都在跟 AI 说话的感觉如何，但这是完成事情的高效方式。而且更有可能的是，你的 AI 会一直烦你。原因在于它完成事情太快了。

I'm not sure how I feel about the future where everybody is walking around talking to their AI, but it's such an efficient way to get stuff done. And it's more likely that your AI is bothering you all the time. And the reason for that is because it's getting stuff done so fast.

是的。

Yeah.

它会向你汇报：‘我完成了。你接下来想让我做什么？’我认为这是大多数人没有意识到的。最常跟他们聊天、发短信的人将是他们的 Claude 或助手。

It's reporting back to you, 'I got that done. What do you want me to do next?' That's the part that I think most people don't realize. The person who's going to be chatting with them, texting them most is their Claude or their assistant.

多么不可思议的未来。

What an incredible future.

我读到你把你的成功很大程度上归功于你比任何人都更努力工作和承受更多痛苦的能力。所以，我们可以列出这涉及到的许多事情：应对失败、我们谈到的持续工程问题、人的问题、不确定性、责任、疲惫、尴尬、你提到的公司濒死时刻，还有现在作为这家公司的 CEO 所面临的压力——经济体和国家围绕它制定战略、规划财务分配、规划 AI 基础设施。你如何应对这么大的压力？考虑到有多少国家和人民依赖你，是什么给了你力量？

I read that you attribute a lot of your success to your ability to work harder than anyone and withstand more suffering than anyone. So, we can list many of the things that entails: dealing with failure, the constant engineering problems we've talked about, the human problems, uncertainty, responsibility, exhaustion, embarrassment, the near-death company moments that you've mentioned, but also the pressure now as the CEO of this company that economies and nations strategize around, plan their financial allocations around, plan their AI infrastructure around. How do you deal with this much pressure? What gives you strength given how many nations and peoples depend on you?

我意识到英伟达的成功对美国非常重要。我们创造了巨额税收。我们为国家建立了技术领导地位。技术领导力对国家安全至关重要——不仅是某一个方面，而是所有方面。当我们的国家更加繁荣时，我们可以在国内政策和社会福利方面做得更好，因为我们正在美国创造如此多的再工业化。我们创造了大量的就业机会。我们正在帮助将制造业带回美国，涉及许多不同的工厂：芯片、计算机，当然还有这些 AI 工厂。我完全意识到我拥有这样的好处，这是一份真正的礼物——主流投资者、教师、警察，他们出于某种原因投资了英伟达，或者因为看了 Jim Cramer 买了些股票，现在成了百万富翁。

I'm conscious about the fact that Nvidia's success is very important to the United States. We generate enormous amounts of tax revenues. We establish technology leadership for our nation. Technology leadership is important for national security. National security not just in one aspect, but all aspects. When our country is more prosperous, we can do a better job with domestic policies and helping social benefits because we're generating so much re-industrialization in the United States. We're creating mountains of jobs. We're helping shift how we build things back to the United States in so many different plants: chips, computers, and of course these AI factories. I'm completely aware that I have the benefit, and this is a real gift, with mainstream investors, teachers, policemen who have somehow, for whatever reason, invested in Nvidia, or because they watch Jim Cramer bought some stock and now are millionaires.

嗯。

Mhm.

我完全清楚这种情况。我意识到英伟达是我们身后和下游一个非常庞大的生态系统合作伙伴网络的核心。所以我处理这个问题的方式就是我刚才所做的。

And I am completely aware of that circumstance. I'm aware of the circumstance that Nvidia is central to a very large network of ecosystem partners behind us and downstream from us. And so the way I deal with that is exactly what I just did.

我会思考：我们在做什么？它导致了什么？对他人有什么影响，无论是积极的还是带来巨大负担，比如供应链？那么问题就是，你打算怎么做？几乎所有事情，我都会拆解。我会思考：情况如何？什么变了？难点在哪？我打算怎么做？我把它拆解、分解问题。把情况分解后，就变成了我能处理的可行事项。之后唯一能做的就是：你做了吗？要么你自己做，要么让别人做。如果你没做，你意识到需要做却没做，也没让别人做，那就别抱怨了。所以我对自己相当严格。但我也分解问题，这样我就不会恐慌。我能睡着，因为我列出了需要做的事情。我确保所有可能危及公司、合作伙伴或行业的事情，我都告诉了别人。所有我觉得可能危及任何人的事情，我都告诉了别人。而且我告诉了有能力处理的人。这样我就卸下了负担，或者我正在处理。除此之外，Lex，你还能做什么呢？

I reason about what is it that we're doing? What is it causing? What's the impact that has on other people, positively or even through great burden, for example the supply chain? And the question is therefore what are you going to do about it? Almost everything I feel I break it down. I reason about okay, what's the circumstance? What has changed? What's hard? And what am I going to do about it? I break it down, decompose the problem. The decomposition of these circumstances turns it into manageable things that I can do. The only thing after that I can do is: did you do it? Did you either do it or get somebody else to do it? And if you didn't do it, you reason that you need to do it and you didn't do it and you didn't get anybody else to do it, then stop crying about it. So I'm fairly tough on myself. But I also break things down so that I don't panic. I can go to sleep because I've made the list of things that needed to be done. I've made sure that everything that could put our company in harm's way, could put my partners in harm's way, put our industry in harm's way, I've told somebody. Everything that I feel could put anybody in harm's way, I've told someone. And I've told someone who could do something about it. So I've gotten it off my chest or I'm doing something about it. After that, Lex, what else can you do?

那么，考虑到打造英伟达的旅程中经历了那么多疯狂而强烈的痛苦，你心理上有没有跌入过低谷？

So given all the insane intense amount of suffering on the journey of building up Nvidia, have you hit low points psychologically?

哦，是的。哦，当然。一直都有。一直都有。

Oh yeah. Oh yeah, sure. All the time. All the time.

然后你就把问题分解成小块。

And there you just break down the problem into pieces.

是的。而且部分原因，Lex，是遗忘。你知道，AI 学习最重要的特性之一就是系统性遗忘。你需要知道什么时候该忘记一些事情。你不能记住所有东西。你不能保留所有东西。你不想背负一切。我很快做的一件事就是分解问题。我思考问题，然后分担负担。我说我告诉所有人，本质上就是在分担那个负担。尽可能快。任何让我担心的事，就告诉别人。不要憋着。把问题分解成更小的部分，让别人去做，并激励他们去行动。但一部分就是遗忘。很多时候你必须对自己狠一点，振作起来，别抱怨了，我们开始干吧，然后你就起床了。另一部分是你被下一个闪光点吸引，下一个未来，下一个机会。好了，那已经过去了。接下来是什么？这很重要。你看伟大的运动员就是这样。他们只关心下一个得分点。上一个得分点已经过去了。尴尬、挫折都过去了。而且因为我很多工作都是公开进行的，Lex，你也有不少公开工作。所以我很多工作都是公开的。我说很多当时听起来合理或有趣的话。大多时候只是因为当时我觉得有趣，但事后回想就没那么有趣了。不过……

Yeah. And part of it, you know Lex, part of it is forgetting. One of the most important attributes of AI learning, as you know, is systematic forgetting. You need to know when to forget some things. You can't memorize everything. You can't keep everything. You don't want to carry everything. One of the things I do very quickly is I decompose the problem. I reason about the problem and I share the load. When I say I tell everybody, I'm essentially sharing that burden. As quickly as possible. Whatever worries me, tell somebody else. Don't just keep it. Decompose the problem into smaller parts and get people to do something about it, and inspire them to be able to go do something about it. But part of it is just forgetting. A lot of it is you got to be tough on yourself, just come on, stop crying about it, let's get going, and then you get out of bed. The other part is you're attracted to the next shiny light, the next future, the next opportunity. Okay, that's behind us. What's next? It's a lot. I think you watch this with great athletes. They just worry about the next point. The last point is behind them. The embarrassment, the setback. And because I do so much of my job publicly, you know Lex, you do a fair amount of your job publicly too. So I do a lot of my job publicly. I say a lot of things that seem sensible at the time or funny at the time. Mostly it's just because it's funny to me at the time, and then you reflect on it and it's less funny. But...

是的。不，相信我，我懂。但你基本上让自己被未来的光芒牵引。忘记过去，一直朝着那个方向努力。

Yeah. No, trust me, I know. But you basically allow yourself to be pulled by the light of the future. Forget the past and just keep working towards that.

没错。

That's right.

我是说，你说过一句名言：如果你知道打造英伟达有多难——结果比你预想的难一百万倍——你就不会去做了。但这不是所有值得做的事都这样吗？

I mean, you did say this kind of famous thing that if you knew how hard it would be to build Nvidia, it turned out to be a million times more hard than you anticipated, you wouldn't do it. But isn't that probably true about everything worth doing?

正是。顺便说一句，这正是我想解释的：拥有孩童般的心态是一种不可思议的超能力。我经常对自己说，当我看到某件事，几乎每件事，我的第一反应是‘这能有多难？’你让自己进入那种模式。这能有多难？从来没人做过。看起来巨大无比。要花几千亿美元。要投入所有这些。然后你就想，‘是啊，但能有多难？’你必须让自己进入那种心态。你不想过度预演所有的挫折、考验、磨难和失望。你不想提前模拟所有那些。你不想知道那些。你想带着‘这会很完美、会很棒、会非常有趣’的想法进入新体验。然后当你身处其中时，你需要有耐力。你需要有毅力，这样当挫折真的发生时——那些挫折会让你惊讶，失望不会让你惊讶，尴尬会让你惊讶，羞辱会让你惊讶——你不能让它们阻止你。你得开启另一面：忘记它，继续前进，保持前进。只要我对未来的假设以及未来为何会实现的原因没有改变或没有实质性改变，那么我就应该预期结果不会改变。所以我模拟的未来结果仍然会发生。如果它仍然会发生，我就仍然会去追求它。我相信它会实现。所以这里结合了两三种人类特质：以全新心态进入体验的能力、忘记挫折的能力、相信自己、相信自己所信并坚持信念的能力。但你也在不断重新评估。这种三、四、五种特质的组合，我认为对韧性非常重要。我很幸运，无论是什么人生经历造就了这些，我拥有了那四五种特质。我总是充满好奇，总是在学习。我总是向每个人学习。

Exactly. That is by the way what I was trying to explain: there's an incredible superpower of having the mind of a child. I say to myself often when I look at something, almost everything, my first thought is 'How hard can it be?' You get yourself into that mode. How hard could it be? Nobody's ever done it. It looks gigantic. It's going to cost hundreds of billions of dollars. It's going to take all this. And you just go, 'Yeah, but how hard could it be?' You got to get yourself into that state of mind. You don't want to overstimulate everything and all the setbacks and all the trials and tribulations and all the disappointments. You don't want to simulate all that in advance. You don't want to know that. You want to go into a new experience thinking it's going to be perfect. It's going to be great. It's going to be incredibly fun. And then while you're there, you need to have endurance. You need to have grit so that when the setbacks actually happen and those setbacks are going to surprise you, the disappointments aren't going to surprise you. The embarrassments are going to surprise you, the humiliations are going to surprise you. You just can't let that stop you. You got to turn on the other bit which is just forget about it, move on, keep moving. To the extent that my assumptions about the future and why the future is going to manifest, so long as those assumptions and that input doesn't change or didn't change materially, then I should expect that the output won't change. So my simulated output of the future is still going to happen. And if it's still going to happen, I'm still going to go after it. I believe it's going to happen. So there's a combination of two or three human characteristics: the ability to go into an experience fresh-minded, the ability to forget the setbacks, the ability to believe in yourself, to believe what you believe and stay true to that belief. But you're constantly re-evaluating. This combination of three, four, five things I think is really important for resilience. I'm fortunate that whatever life experience has led to this, I've got kind of those four or five things. I'm always curious, always learning. I'm always learning from everybody.

你知道，我总是在问，因为我对一切都保持谦逊，我总是在想：‘天哪，他们做得真好。他们做得太棒了。我想知道他们在想什么。他们是怎么做到的……’所以我在很多方面都在模拟每个人。我几乎在模仿我观察到的每一个人。你对你观察和尊重的人所做的一切都抱有同理心，因此你不断在学习。

You know, I'm always asking, and because I'm humble about everything, I'm always thinking, 'Gosh, they did that so nicely. They did that so wonderfully. I wonder what they're thinking through. How do they...' So I'm simulating everybody in a lot of ways. I'm emulating almost everybody I watch. You're empathetic towards everything they do that you're observing and respect, and so you're constantly learning.

你现在是地球上最富有的人之一，也是最成功的人之一。保持谦逊是否变得更难了？金钱、权力和名声的影响是否让你更难在自己内心承认错误，从而倾听别人的不同意见并向他们学习？

You're now one of the wealthiest people on earth, one of the most successful humans on earth. Is it harder to be humble and to feel the effect of money and power and fame in making it harder for you to be wrong in your own head enough to hear out an opinion of somebody else when it disagrees with you and learn from them?

令人惊讶的是，并没有。我反而会走向另一个方向，因为我的很多工作都是公开进行的。当我错了的时候，几乎每个人都能看到。

Surprisingly, no. I would actually go the other way because I do so much of my work publicly. When I'm wrong, pretty much everybody sees it.

你被谦卑了。

You get humbled.

是的。当我错了，或者事情没有按预期发展时，我在外部说的多数事情我都是相当确定的。原因是因为这会影响别人，我对此非常关心和谨慎。对于我在会议中推理的事情，很多结果可能不同。但这从未阻止我推理。我管理和领导的方式是，我经常在人们面前推理。即使我在和你说话，你也能看到我在推理事情。我想确保你理解我说的话，不是因为我告诉了你，而是因为我向你展示了我得出这个结论的步骤。然后你可以决定是否最终相信我说的话。我整天都在和所有员工的会议中这样做。我不断推理。‘让我告诉你我是怎么看的。’然后我推理出来。这给了每个人机会来打断并说：‘我不同意那部分。’推理事情并让人们与之互动的好处是，他们不必不同意你的结论。他们可以不同意你的推理步骤，他们可以把我拉向不同的方向，然后我们一起向前推理。我们有点像集体探路，这真的很棒。

Yeah. And when I'm wrong, or it didn't turn out that way, most of the things that I say outside I'm fairly certain about. The reason is because it's going to impact somebody else, and I want to be quite concerned about that and quite circumspect about that. For stuff that I'm reasoning about inside a meeting, a lot of things could turn out differently. But it doesn't ever stop me from reasoning. The way that I manage and lead, I'm constantly reasoning in front of people. Even when I'm talking to you, you can kind of see me reasoning through things. I want to make sure that you understand what I'm saying, not because I told you, but because I show you the steps that I got there. And then you could decide whether you believe what I said in the end. I'm doing that all day long in meetings with all of my employees. I'm constantly reasoning through. 'Let me tell you how I see it.' And I reason through it. It gives everybody the opportunity to intercept and say, 'I disagree with that part.' The nice thing about reasoning through things and letting people interact with it is that they don't have to disagree with your outcome. They can disagree with your reasoning steps and they could pull me in different directions and then we can reason forward. We're kind of collective pathfinding, and it's really fantastic.

是的。你在解释事情时有一种方式，我能感觉到你实际上在现场推理，并且始终保持开放心态，让我觉得我可以引导你的思考。在这么多年的成功和痛苦之后，你还能保持这一点，真的很美。我认为有时候痛苦会让你封闭一点。而保持对尴尬的容忍度，这确实是一件真实的事。

Yeah. You have this way about you of when you're explaining stuff, I can feel you actually reasoning on the spot about it with a constant open-mindedness where I could feel like I could steer your thinking. And that's really beautiful that you've been able to maintain that after so many years of success and pain. I think sometimes pain makes you close down a bit. And I think to maintain tolerance for embarrassment is a real thing.

是的。有很多年都在让自己尴尬。即使在那些会议中，你知道周围有人，你宣布了一个想法，然后被证明是错的，能够承认并从中学到东西，这在人性层面上是非常困难的。

Yeah. There's many years of embarrassing yourself. Even those meetings knowing that there's people around you where you declared one idea and it was shown that that idea was wrong and be able to admit that and to grow from that. That's very difficult on a human level.

是的。嗯，你知道，他们知道我曾经……最近我的第一份工作是打扫厕所。所以我很高兴你保持了在丹尼斯工作的那种精神。那很美。你从丹尼斯开始的整个旅程都很美。让我问问你关于电子游戏的事。我是一个大游戏迷。我必须感谢英伟达多年来令人难以置信的图形技术。

Yeah. Well, you know, they knew I was... recently my first job was cleaning toilets. So I'm glad you maintain that same spirit of Denny's work. That was beautiful. Your whole journey from starting at Denny's is a beautiful one. Let me ask you about video games. So I'm a big gaming fan. I have to say thank you to Nvidia for many years of incredible graphics.

顺便说一句，GeForce 至今仍是我们第一位的营销策略。人们在青少年时期就了解了英伟达。然后他们上大学，知道英伟达是谁。一开始只是玩《使命召唤》、《堡垒之夜》，后来他们使用 CUDA，再后来他们在 Blender 和 Do 以及 Auto 中使用英伟达。

By the way, GeForce is still to this day our number one marketing strategy. People learn about Nvidia while they're in their teenage years. Then they go to college and they know who Nvidia is. And in the beginning it's just playing Call of Duty, Fortnite, and then later they're using CUDA, and then later they're using Nvidia in Blender and Do and Auto.

我跟一个朋友提到我在和你谈话。他说：‘哦，他们做很棒的游戏 GPU。’

I mentioned to a friend that I'm talking with you. He said, 'Oh, they make great gaming GPUs.'

没错。没错。还有更多，但是的。人们真的很喜欢它。它给很多人带来了很多快乐。硬件真的让这些世界栩栩如生。

Exactly. Exactly. There's more to it, but yeah. People really love it. It brought a lot of joy to a lot of people. The hardware really brings these worlds to life.

关于 DLSS 5 有一些争议。你能给我解释一下这出戏吗？我猜网上的玩家担心它会让游戏看起来像 AI 垃圾。你怎么看这出戏？

There was some controversy around this with DLSS 5. Can you explain to me the drama around this? I guess gamers online were concerned that it makes games look like AI slop. What do you think of this drama?

是的，我认为他们的观点有道理，我能理解他们的想法，因为我自己也不喜欢 AI 垃圾。所有 AI 生成的内容越来越相似，而且都很漂亮。我对他们的想法感同身受。但这并不是 DLSS 5 想要做的。我展示过几个例子，但 DLSS 5 是 3D 条件化的、3D 引导的。它是由真实结构数据引导的。所以艺术家决定几何形状。我们在每一帧中都完全忠实于几何形状。它由纹理、艺术家的艺术性所条件化。每一帧它都会增强，但不会改变任何东西。现在的问题是关于增强。DLSS 5 还允许，因为系统是开放的，你可以训练自己的模型来决定，甚至将来你可以提示它：‘我想要卡通着色器。我希望它看起来像这种……’所以你可以给它一个例子，它会以那种风格生成，全部与艺术家的艺术性、风格和意图一致。所有这些都是为艺术家做的，这样他们可以创造出更美丽但仍然符合他们风格的作品。我认为他们得到的印象是游戏会以原样发布，然后我们进行后处理。这不是 DLSS 的意图。DLSS 与艺术家集成。它是关于给艺术家提供 AI 工具、生成式 AI 工具。他们可以选择不使用它。

Yeah, I think their perspective makes sense and I can see where they're coming from because I don't love AI slop myself. All of the AI generated content increasingly looks similar and they're all beautiful. I'm empathetic towards what they're thinking. That's just not what DLSS 5 is trying to do. I showed several examples of it, but DLSS 5 is 3D conditioned, 3D guided. It's ground truth structure data guided. So the artist determines the geometry. We are completely truthful to the geometry in every single frame. It's conditioned by the textures, the artistry of the artist. Every single frame it enhances but it doesn't change anything. Now the question is about enhancing. DLSS 5 also lets, because the system is open, you could train your own models to determine, and you could even in the future prompt it: 'I want it to be a toon shader. I want it to look like this kind of...' So you can give it an example and it would generate in the style of that, all consistent with the artistry, the style, the intent of the artist. All of that is done for the artist so that they can create something that is more beautiful but still in the style they want. I think they got the impression that the games are going to come out the way they are shipped and then we're going to post-process it. That's not what DLSS is intended to do. DLSS is integrated with the artist. It's about giving the artist the tool of AI, the tool of generative AI. They could decide not to use it.

我认为人们对人脸非常敏感。

I think people are very sensitive to human faces.

是的。

Yeah.

我们现在正处在一个美好的时刻，人们对 AI 垃圾内容变得敏感。

And we're now living in this moment, which I think is a beautiful one, which is people are sensitive to AI slop.

是的。

Yeah.

它像一面镜子，让我们意识到我们追求的不完美，有时并不是完美的图形，它帮助我们理解我们在创造的世界中发现什么引人入胜。

It puts a mirror to ourselves to help us realize that what we seek as imperfections, what we seek is sometimes not perfect graphics, it helps us understand what we find compelling in the worlds we create.

这很美。只要它是帮助我们创造这些世界的工具。

And that's beautiful. And as long as it's tools that help us create those worlds.

这很美。只要它是帮助我们创造这些世界的工具。

And that's beautiful. And as long as it's tools that help us create those worlds.

是的，没错。

Yeah, that's right.

太棒了。

It's wonderful.

没错。这只是又一个工具。他们希望生成模型生成与照片级真实相反的东西。

That's right. It's yet another tool. And they want the generative models to generate the opposite of photoreal.

嗯。

Mhm.

是的。它也能做到。所以这只是又一个工具。我认为玩家们可能也会欣赏，在过去的几年里，我们向游戏开发者介绍了皮肤着色器，许多游戏都有包含次表面散射的皮肤着色器，使皮肤看起来更像皮肤。因此，游戏开发者们正在寻找越来越多的工具来表达他们的艺术。所以这只是他们可以决定使用的又一个工具。

Yeah. It'll do that too. And so it's just yet another tool. I think the gamers might also appreciate that in the last couple years we introduced skin shaders to the game developers and many of those games have skin shaders that include subsurface scattering that make skin look more skin-like. And so the industry's game developers are looking for more and more tools to express their art. And so this is just one more tool they could decide what to use.

一个荒谬的问题。你认为有史以来最伟大或最有影响力的游戏是什么？也许从英伟达的角度来看。

Ridiculous question. What do you think is the greatest or most influential game ever made? Maybe from Nvidia's perspective.

《毁灭战士》。

Doom.

《毁灭战士》。毫无疑问。那是 3D 的开端。我认为《毁灭战士》处于文化影响和行业将 PC 转变为游戏设备的交汇点。那是一个非常重要的时刻。当然，飞行模拟公司在此之前，但它们没有《毁灭战士》那样的普及度，使行业将 PC 从办公自动化工具转变为家庭和游戏玩家的个人电脑。所以《毁灭战士》在那方面影响深远。从实际游戏技术的角度来看，我会说是《VR 战士》。我们和这两家公司都是好朋友，你知道。然后还有更近期的游戏。我的意思是，《赛博朋克 2077》，非常棒的 GPU 加速图形，比如完全光线追踪。另外，我个人是《上古卷轴 5：天际》的超级粉丝，它发布很久了，但人们发布模组，就像是一个不同的游戏，让我一遍又一遍地重玩，让你意识到你可以用一种全新的方式重新体验你已经热爱的世界。所以我经常这样做。我最喜欢的游戏之一就是在《天际》里闲逛。我们创建了一个叫 RTX Mod 的东西。它是一个模组工具。它允许社区将最新技术注入到老游戏中。

Doom. Unquestionably. That was the start of the 3D. I would say Doom from the intersection of the cultural implication as well as the industry turning a PC into a gaming device. That was a very important moment. Now, of course, flight simulation companies were before it, but they just didn't have the popularity that Doom did to have made the industry turn the PC from an office automation tool into a personal computer for families and gamers and things like that. And so Doom was really impactful there. From a from an actual game technology perspective, I would say Virtual Fighter. And so we're great friends with both of them, you know. And then there's games more recently. I mean, Cyberpunk 2077, really nice GPU accelerated graphics, like fully ray traced. Also, I personally am a huge fan of Skyrim, Elder Scrolls, and it's been released a long time ago, but people release mods and it's like a different game and it just allows me to replay the game over and over and it makes you realize you can reexperience in a totally new way the world you already love. So I do that all the time. One of my favorite games just walk around Skyrim. We created this thing called RTX Mod. It's a modding tool. And it allows the community to inject the latest technology into an old game.

太棒了。

Awesome.

当然，一款伟大的电子游戏不仅仅在于图形。还有故事和角色发展。但美丽的图形可以增加沉浸感，让你感觉被传送到另一个地方。

Of course, what makes a great video game is not just graphics. It's also story and character development. But beautiful graphics can add to the immersion, the feeling like it's another place you're transported to.

你说得很准确，AGI 的时间线问题取决于你对 AGI 的定义。那么让我问你一个可能的时间线。让我们采用这个荒谬的 AGI 定义：一个能够基本上做你的工作的 AI 系统。也就是说，运营、不，是创办、发展并运营一家成功的科技公司，价值超过十亿美元。所以你知道做所有这些事情有多难。那么我们还差多远？我们说的是一个智能体，它要做所有极其复杂的事情，包括创新、寻找客户、销售、管理、组建一个由一些智能体和一些人类组成的团队，诸如此类。这是 5 年、10 年、15 年还是 20 年后？

You said I think accurately that the AGI timeline question rests on your definition of AGI. So let me ask you about a possible timeline. Let's take this ridiculous definition of AGI: an AI system that's able to essentially do your job. So run, no start, grow and run a successful technology company that's worth more than a billion dollars. So you know how hard it is to do all those components. So how far are we away from that? We're talking about an agent that does all the incredibly complex stuff required to innovate, find customers, sell to them, manage, build a team of some agents, some humans, all that kind of stuff. Is this 5, 10, 15, 20 years away?

我认为就是现在。我认为我们已经实现了 AGI。

I think it's now. I think we've achieved AGI.

你认为你可以让一个 AI 系统像这样运营一家公司？

You think you can have a company run by an AI system like this?

有可能。原因如下。你说十亿美元，但你没说永远。所以举例来说，一个智能体能够创建一个网络服务，一些有趣的小应用，突然有几十亿人用了一段时间，然后不久就倒闭了，这并非不可能。我们在互联网时代看到了一大堆这样的公司，那些网站中的大多数并不比今天智能体能够生成的东西更复杂。

Possible. And the reason for that is this. You said a billion and you didn't say forever. And so for example, it is not out of the question that an agent was able to create a web service, some interesting little app that all of a sudden a few billion people used for a while and then it went out of business again shortly after. Now, we saw a whole bunch of those type of companies during the internet era and most of those websites were not anything more sophisticated than what an agent could generate today.

有趣。实现病毒式传播并从中获利。

Interesting. Achieve virality and monetize that virality.

是的。只是我不知道它是什么，但当时我也无法预测那些公司中的任何一个。

Yeah. It's just that I don't know what it is, but I couldn't have predicted any of those companies at the time either.

你这句话会让很多人兴奋。

You're going to get a lot of people excited with that statement.

是的。就像，你什么意思？我只要启动一个智能体就能赚很多钱？顺便说一句，这正在发生，对吧？你知道，当你去中国，你会看到很多人教他们的智能体出去找工作、做事情、赚钱。如果出现一些社交事件，或者有人创建了一个超级可爱的数字影响者，或者某个社交应用喂养你的电子宠物之类的东西，然后突然爆红，我一点也不会惊讶。很多人用几个月，然后就消失了。现在，十万个这样的智能体建立英伟达的概率是 0%。而我要做的一部分，也是我希望我们都做的，是认识到人们真的很担心他们的工作，我只想提醒他们，你工作的目的、任务和用来完成工作的工具是相关的，但不是一回事。我已经做了 33 年的工作。我是世界上任期最长的科技 CEO。34 年来，我用来完成工作的工具一直在变化，有时在几年内变化相当剧烈。而我希望每个人都听到的一个故事是，计算机科学家和 AI 研究人员说第一个会消失的工作是放射学，因为计算机视觉将达到超人水平，而它确实做到了。计算机视觉在 2019 年、2020 年或稍晚一点达到了超人水平。所以计算机视觉达到超人水平已经很久了。预测是放射科医生会消失，因为研究放射扫描是过去的事了。AI 会做这个。好吧，他们完全正确。计算机视觉完全超人。今天每一个放射学平台和软件包都是由 AI 驱动的。

Yeah. It's like, what do you mean? I can just launch an agent and make a lot of money? Well, by the way, it's happening right now, right? You know that when you go to China, you're going to see a whole bunch of people teaching their agents to try to go out and look for jobs and do work, make money. And I wouldn't be surprised if some social thing happened or somebody created a digital influencer, super super cute, or some social application that feeds your little Tamagotchi or something like that and it becomes an out of the blue instant success. A lot of people use it for a couple of months and it kind of dies away. Now the odds of 100,000 of those agents building Nvidia is 0%. And the one part that I will do and I want to make sure we all do is to recognize that people are really worried about their jobs and I just want to remind them that the purpose of your job and the tasks and the tools that you use to do your job are related, not the same. I've been doing my job for 33 years. I'm the longest running tech CEO in the world. 34 years and the tools that I've used to do my job has changed continuously in the last 34 years and sometimes quite dramatically over the course of a couple of years. And the one story that I really want to make sure that everybody hears is the story that the first job that computer scientists and AI researchers said was going to go away was radiology because computer vision was going to achieve superhuman levels and it did. Computer vision was superhuman in 2019, 2020 maybe a little bit later. And so it's been a long time since computer vision has been superhuman. And so the prediction was radiologists would go away because studying radiology scans was a thing of the past. AI will do that. Well, they were absolutely right. Computer vision is completely superhuman. Every radiology platform and package today is driven by AI.

然而放射科医生的数量却增长了。那么问题来了，为什么？现在全球都缺放射科医生。那些危言耸听的警告过头了，把人们吓得不敢从事这个对社会如此重要的职业，造成了伤害。那为什么当初的预测是错的呢？因为放射科医生的职责是诊断疾病、帮助患者和医生确诊。现在扫描速度大大加快，可以研究更多影像，诊断更精准，患者入院更快，接诊更多人。医院赚更多钱，住院病人更多，自然需要更多放射科医生。这结果显而易见。

And yet the number of radiologists grew. And so the question is why? And we now have a shortage of radiologists in the world. And so one the alarmist warning went too far and has scared people from doing this profession that is so important to society. And so it did harm. Now why was it wrong? The reason why is because the purpose of a radiologist, the purpose is to diagnose disease and help patients and doctors diagnose disease. And because we're able to study scans so much faster now, you could study more scans. You could diagnose better. You could get patients in faster. We can see more people. The hospitals are making more money. You have more patients in the hospital. You need more radiologists. I mean the amazing thing is it's so obvious this was going to happen.

英伟达的软件工程师数量会增长，不会减少。因为软件工程师的职责和编码任务相关，但并不等同。我希望我的软件工程师解决问题，不在乎他们写多少行代码。他们的工作本质没变：解决问题、团队协作、诊断问题、评估结果、寻找新问题、创新、串联思路——这些都不会消失。

The number of software engineers at NVIDIA is going to grow, not decline. And the reason for that is because the purpose of a software engineer and the task of a software engineer for coding are related, not the same. I wanted my software engineers to solve problems. I didn't care how many lines of code they wrote. But their job, their purpose of their job didn't change. Solving problems, working as a team, diagnosing problems, evaluating the result, looking for new problems to solve, innovation, connecting dots — none of that stuff is going to go away.

所以，你认为即使单看编程，全球程序员数量也可能增加而非减少？

So, you think it's possible that let's even take coding, you think the number of programmers in the world might increase, not decrease?

原因如下。什么是编程？我认为今天的编程就是写规格说明，如果你想更直接，甚至可以给出软件架构。那么有多少人能这么做？描述一个规格，让计算机去构建东西。有多少人？我认为从 3000 万跃升到 10 亿。未来每个木匠都会是程序员，但有了 AI 的木匠同时也是建筑师。他们为客户创造的价值提升了，技艺也大幅提高。我相信每个会计同时也是财务分析师、财务顾问。所有这些职业都升级了。如果我是木匠，看到 AI 我会欣喜若狂——我能为客户提供的服务；如果我是水管工，同样欣喜若狂。而现在的程序员和软件工程师，他们处于前沿，直觉上懂得如何用自然语言与智能体沟通，设计出最好的软件。

And the reason for that is this. What is the definition of coding? I believe that the definition of coding as of today is simply specifying specification and maybe if you want to be rather directive you could even give it an architecture of the software that you wanted to write. So the question is how many people could do that? Describe a specification for a computer to go telling the computer what to go build. How many people? I think we just went from 30 million to probably 1 billion. And so every carpenter in the future will be a coder. Except a carpenter with AI is also an architect. They just increased the value that they could deliver to the customer. Their artistry just elevated tremendously. I believe that every accountant is also your financial analyst, also your financial adviser. So all of these professions have just been elevated and if I were a carpenter, I see AI, I would just go completely berserk. The services I can bring to my clients, if I were a plumber, completely go berserk. And the people that are currently programmers and software engineers, I think they're at the cutting edge of understanding intuitively how to communicate with the agents using natural language in order to design the best kind of software.

没错。所以长期来看它们会融合，但我认为学习编程仍有价值，比如了解编程语言是什么、传统编程、编程语言的最佳实践、大型软件系统的设计原则。

That's right. So over time they'll converge but I think there's still value in learning how to program, like learning what programming languages are, the old kind of programming, what are good practices for programming languages, what are design principles for programming languages for large software systems.

原因在于，Lex，我想对观众说，规格说明的目标和艺术性取决于你要解决什么问题。当我思考公司战略、制定方向时，我会描述得足够具体，让大家理解方向并可以执行，但我会故意留白，让 4.3 万名优秀员工把它做得比我预想的更好。所以和工程师合作时，我会想：我要解决什么问题？和谁合作？规格和架构的定义程度与此相关。每个人都要学会在编程光谱中找到自己的位置。写规格就是编程。你可能选择非常具体，因为目标明确；也可能选择探索性更强，留出空间与 AI 反复互动，甚至突破自己的创造力边界。这种在光谱中定位的艺术，就是编程的未来。

And the reason for that, Lex, and I just say for the audience, I think the goal of specification, the artistry of specification, is going to depend on what problem you're trying to solve. When I'm thinking about giving the company strategies and formulating corporate directions and things that we should do, I describe it at a level that is sufficiently specific that people generally understand the direction and it's actionable, but I underspecify it on purpose so that enable 43,000 amazing people to make it even better than I imagined. And so when I'm working with engineers, I think about what problem am I trying to solve? Who am I working with? And the level of specification, the level of architecture definition relates to that. And so everybody's going to have to learn where in the spectrum of coding they want to be. Writing a specification is coding. And so you might decide to be quite prescriptive because there's a very specific outcome you're looking for. You might decide that this is an area you want to be much more exploratory. And so you might underspecify and enable you to go back and forth with the AI to even push your own boundaries of creativity. And so this artistry of where you are in the spectrum, this is the future of coding.

但我想多谈一点，除了编程，很多人确实担心自己的工作，尤其是白领阶层。自动化与新技术到来时总是动荡不安，我们谁也不知道该怎么办。首先，我们需要同情心，要感受到失业者及其家庭的痛苦。每当像人工智能这样的变革性技术出现，总会带来很多痛苦，我不知道如何应对。希望随着工具进化，能为这些人创造更多机会，让他们的工作更高效、更有趣，就像编程带给我的乐趣一样——我从未如此享受编程。希望它能自动化枯燥的部分，把创造性部分留给人类。但痛苦和苦难依然存在。所以我第一个建议——这也是我应对焦虑的方式——我们之前也聊过。

But just to linger on it, outside of coding, I think a lot of people rightfully so are worried about their jobs, have a lot of anxiety about their jobs, especially in the white collar sector. I don't think any of us know what to do with tumultuous times that always come when automations and new technology arrives. And I just first of all I think we all need to have compassion and the responsibility to feel the burden of what the actual suffering feels like for individual people and families that lose their job. I think whenever you have transformative technology like that's coming with artificial intelligence, there's going to be a lot of pain and I don't know what to do about that pain. Hopefully, it creates much more opportunities for those same people for the same kind of job as the tooling evolves and makes them more productive and makes it more fun hopefully as it does in the programming. I've been having so much fun programming, I have to say. Like I've never had this much fun. So hopefully it makes their job automates the boring parts and makes the creative parts the ones that the human beings are responsible for. But still there's going to be a lot of pain and suffering. So my first recommendation before and this is now how I deal with anxiety. In fact, we just talked about it earlier.

嗯。

Mhm.

对未来的巨大焦虑、对压力的巨大焦虑、对不确定性的巨大焦虑。我首先分解它，然后告诉自己：有些事情你可以改变，有些事情你无能为力。但对于能改变的事，我们来理性分析，然后去做。

Enormous anxiety about the future, enormous anxiety about the pressure, enormous anxiety about uncertainty. I first break it down and then I'm going to tell myself, okay, there are some things you can do something about. There are some things you can't do anything about, but for the stuff that you can do something about, let's reason about it and let's go do it.

如果今天我们要招聘一名应届毕业生，有两个候选人：一个对 AI 一无所知，另一个是使用 AI 的专家。我会雇那个 AI 专家。如果是会计、营销、供应链、客服、销售、商务拓展、律师，我都会雇那个 AI 专家。所以我建议每个大学生、每位老师都应该鼓励学生去使用 AI。

If we were to hire a new college graduate today and I have a choice between two, one that has no clue what AI is and one that is expert in using AI, I would hire the one who's expert in using AI. If I had an accountant, a marketing person, the one that is expert in using AI, supply chain, customer service, a salesperson, business development, a lawyer, I would hire the one who is expert in using AI. And so I would advise that every college student, every teacher should encourage their student to go use AI.

每个大学生毕业时都应该是 AI 专家。每个人，如果你是木匠，如果你是电工，去用 AI。去看看它能如何改变你当前的工作。提升自己。如果我是农民，我绝对会用 AI。如果我是药剂师，我会用 AI。我想看看它能如何提升我的工作，这样我就能成为自己革新这个行业的创新者。

Every college student should graduate and be an expert in AI. And everybody, if you're a carpenter, if you're an electrician, go use AI. Go see what it can do to transform your current job. Elevate yourself. If I were a farmer, I would absolutely use AI. If I were a pharmacist, I would use AI. I want to see what it could do to elevate my job so that I could be the innovator to revolutionize this industry myself.

所以那会是我做的第一件事。然后我也会帮助他们。确实，这项技术会取代并消除许多任务。因为它会自动化。如果你的工作就是任务本身，那么你很可能受到冲击。如果你的工作目的包含某些任务，那么学习用 AI 自动化这些任务就至关重要。中间还有很大的谱系。

And so that would be the first thing that I would do. And then I would also help them. It is the case that the technology will dislocate and will eliminate many tasks. Because it will automate it. If your job is the task, then you're very highly going to be disrupted. If your job's purpose includes certain tasks, then it's vital that you go learn how to use AI to automate those tasks. And there's the world of spectrum in between.

顺便说一句，AI 的美妙之处，聊天机器人版本，在于你可以通过和它交谈来分解你的焦虑和问题。我最近发现，你能通过它梳理生活中的问题，这真的很不可思议。我不是指心理治疗问题。我是说非常实际的：'我担心我的工作。需要什么技能？需要采取哪些步骤？如何提高 AI 水平？'你刚才说的每件事，你都可以直接问，它会给你逐点回答。它就是个很棒的人生教练，句号。

And by the way, the beautiful thing about AI, the chatbot versions, is you can break down your anxiety and break down the problem by talking to it. I've recently found it's really incredible how much you can think through your life's problems. I don't mean therapy problems. I mean very practically: 'I'm worried about my job. What are the skills? What are the steps I need to take? How do I get better at AI?' Everything you just said, you can literally ask and it will give you a point-by-point answer. It's just a great life coach, period.

我不知道怎么用 AI。然后 AI 说：'好吧，让我来教你。'

I don't know how to use AI. And the AI goes, 'Well, let me show you.'

没错。这非常元，但也很不可思议。所以人们绝对应该去用。

Exactly. It's very meta, but it's kind of incredible. So people definitely should.

你不能走到 Excel 面前说'我不知道怎么用 Excel'，那就完了。这正是 AI 在我生活的方方面面所做的：消除了初次使用某样东西时作为初学者的初始摩擦。我可以直接问任何一件事：'我需要采取哪些第一步？'

You can't walk up to Excel and say, 'I don't know how to use Excel.' You're done. That's really what AI has done for me in all walks of life: that initial friction of being a beginner using a thing for the first time. I can literally ask about any single thing: 'What are the first steps I need to take?'

没错。

That's right.

这种手把手的引导，消除了世界上所有体验的摩擦，就像我私下跟你提到的：你说我要去中国和台湾。

And that handholding, removing the friction of all the experiences that the world offers, is like what I mentioned to you offline: you mentioned I'm going to China and Taiwan.

太棒了。我去哪里？做什么？所有这些问题立刻得到回答，太美好了。

So awesome for you. Where do I go? What do I do? All those questions immediately answered, and it's beautiful.

嗯，你去台湾的时候，直接问 AI：'Jensen 在台湾最喜欢的餐厅有哪些？'

Well, when you go to Taiwan, just ask AI, 'What are Jensen's favorite restaurants in Taiwan?'

对。

Yeah.

它真的会……哦，对。准确吗？好吧。

And it'll actually... Oh, yeah. Is it accurate? Okay.

是的。好吧。

Yeah. All right.

遍布台湾。

It's all over Taiwan.

嗯，你在那里是摇滚明星。而且就像我们私下提到的，也许我们会在 Computex GTC Taiwan 相遇，那会非常棒。

Well, you're a rock star over there. And like we also mentioned offline, maybe our paths will cross, which would be really wonderful at Computex GTC Taiwan.

你认为人性、人类意识中是否有某些东西本质上是非计算的？也许有些东西，无论芯片多强大，都无法复制？

Do you think there are some things about human nature, about human consciousness, that is fundamentally non-computational? Maybe something a chip, no matter how powerful, can never replicate?

我不知道芯片是否会紧张。当然，引起焦虑或紧张或任何情绪的条件。我相信 AI 将能够识别和理解这些。我不认为我的芯片会感受到这些。因此，那种焦虑、那种感觉、那种兴奋、所有这些感受如何体现在人类表现中——例如，极其出色的人类表现、运动表现、平均或低于平均——完全相同的环境对不同的人产生不同的结果、不同的表现。我不认为我们正在构建的任何东西会表明，两台不同的计算机在完全相同的上下文下会……当然它会产生统计上不同的结果，但那不是因为它们感受不同。

I don't know if the chip will ever get nervous. And that's the... of course, the conditions by which that causes anxiety or nervousness or whatever emotion. I believe that AI will be able to recognize those and understand those. I don't think my chips will feel those. And therefore, how that anxiety, how that feeling, how that excitement, how all of those feelings manifest in human performance—for example, extremely amazing human performance, athletic performance, average or lesser than average—that entire spectrum of human performance that comes out of exactly the same circumstances for different people, manifesting in different outcomes, different performance. I don't think there's anything about anything we're building that would suggest that two different computers, presented with all of exactly the same context, would... of course it would produce statistically different outcomes, but it's not because it felt different.

是的。主观……我们人类感受到的主观体验确实有某种特别之处。就像我跟你提到的，我跟你说话时很紧张。希望、恐惧、焦虑，以及生活本身，生活的丰富性，一切多么奇妙，我们爱得多深，心碎得多深，我们多么害怕死亡，当亲人去世时我们多么痛苦。所有这一切，全部。我不认为很难想象 AI，一个计算设备，能够做到这些。但关于这一切，我们尚未揭示的谜团太多，我对此持开放态度，愿意被惊喜。

Yeah. The subjective... there's something truly special about the subjective experience that we humans feel. Like I mentioned to you, I was pretty nervous talking to you. The hope, the fear, the anxiety, and just life itself, the richness of life, how amazing everything is, how deeply we fall in love, how deeply our hearts get broken, how afraid we are of death, and how much pain we feel when our loved ones pass away. All of that, the whole thing. I don't think it's very hard to think AI, a computational device, being able to do that. But there are so many mysteries about this whole thing that we're yet to uncover that I am open to be surprised.

过去几个月和几年里，我经常被惊喜。Scaling（规模扩张）可以在智能领域创造一些不可思议的奇迹。看着它真是美妙至极。所以我对惊喜持开放态度。而且，分解什么是智能非常重要。我们一直用的这个词。它不是一个神秘的词。智能有含义。它是一个系统，包括感知、理解、推理和规划能力。那个循环就是智能的根本。智能不是一个完全等同于人性的词。我认为区分两者非常重要。我们有两个词来表达。我不会过度幻想或过度浪漫化智能。我实际上认为智能是一种商品。我身边都是聪明人。而且我身边都是比我更聪明的人，在他们各自的领域里。然而我在那个圈子里有一个角色。这其实挺有趣的。他们比我受教育程度高。他们上的学校比我好。他们在各自领域比我更深。所有人。我有 60 个这样的人。他们对我来说都是超人。而我却坐在中间，指挥着这 60 个人。所以你得问问自己，一个洗碗工有什么特质，能让他坐在超人们中间？这说得通吗？

I've been surprised a lot over the past few months and few years. Scaling can create some incredible miracles in the space of intelligence. It has been truly marvelous to watch. So I'm open to surprise. And it's just really important to break down what is intelligence. The word we use all the time. It's not a mysterious word. Intelligence has a meaning. It's a system that includes perception and understanding and reasoning and the ability to plan. That loop is fundamentally what intelligence is. Intelligence is not one word that is exactly equal to humanity. I think it's really important to separate the two. We have two words for that. I don't over fantasize about and I don't over romanticize about intelligence. I actually think intelligence is a commodity. I'm surrounded by intelligent people. And I'm surrounded by intelligent people more intelligent than I am in each one of the spaces that they're in. And yet I have a role in that circle. It's actually kind of interesting. They're more educated than I am. They went to better schools than I did. They're deeper in whatever field they're in. All of them. I have 60 of them. They're all superhuman to me. And somehow I'm sitting in the middle orchestrating all 60 of them. So you got to ask yourself, what is it about a dishwasher that allows that dishwasher to sit in the middle of superhumans? Does that make sense?

是的。

Yeah.

所以，但这就是我的观点。我的观点是智能是一种功能性的东西。人性不是功能性地定义的。它是一个大得多的词。而我们的生活经验、对痛苦的忍耐、决心，这些是和智能不同的词。

And so, but that's my point. My point is intelligence is a functional thing. Humanity is not specified functionally. It's a much, much bigger word. And our life experience, our tolerance for pain, our determination, those are different words from intelligence.

所以我想让观众理解一件事，如果只能给一个的话，那就是‘智能’这个词被我们过度神化了。真正应该被推崇的是人性、品格、人性、同情心、慷慨——你刚才说的那些。我相信那些才是超人的能力，而智能现在将被商品化，因为我们一直在谈论它。最重要的是你的教育。即使他们说最重要的是你的教育，当你上学时，你获得的不仅仅是知识。但不幸的是，我们的社会把所有东西都塞进了一个词里。而生活远不止一个词。我告诉你，我的经历表明，在智力曲线上低于周围所有人，并不妨碍我是最成功的那个。我希望这能激励其他人：不要因为智能的民主化、商品化而感到焦虑。你应该为此感到振奋。

And so the thing that I want to help the audience understand, if I could give them one thing, is that intelligence is a word that we've elevated to a very high form over time. The word we should really elevate is humanity, character, humanity, all of those things, compassion, generosity, all of the things that you said just now. I believe those are superhuman powers, and that now intelligence is going to be commoditized because we've spoken about it. The most important thing is your education. Even when they said the most important thing is your education, when you went to school, there's more than just knowledge that you gained. But unfortunately, our society had put everything into one single word. And life is more than one word. And I'm just telling you, my life would suggest that being lower on the intelligence curve than everybody around me doesn't change the fact that I'm the most successful. And I think that kind of... I'm trying hopefully to inspire everybody else that don't let this democratization of intelligence, this commoditization of intelligence, cause you anxiety. You should be inspired by that.

是的。我认为 AI 会帮助我们更多地赞美人类。我当然是人性至上、人类至上。我认为让这个世界如此不可思议的是人类，永远都是。而 AI 只是一个不可思议的工具，让我们……

Yeah. I think AI will help us celebrate humans more. And I'm certainly humanity and human first. And I think what makes this world incredible is humans forever will be so. And AI is just this incredible tool that makes us...

完全正确。

That's exactly right.

人类更强大。

Humans more powerful.

完全正确。

That's exactly right.

英伟达的成功，以及我提到的数百万人的生活，很大程度上都依赖于你。但你只是一个凡人，和我们所有人一样终有一死。你会思考自己的死亡吗？你害怕死亡吗？

So much of the success of Nvidia and the lives of millions of people that I mentioned depend on you. But you're just one human, like we mentioned, mortal like all of us. Do you think about your mortality? Are you afraid of death?

我真的不想死。我拥有美好的人生、美满的家庭和非常重要的工作。这不是一次‘一生一次’的经历——那种经历很多人都有过，只是不同的人而已。我正在经历的是一次‘人类历史一次’的经历。英伟达是历史上最重要的科技公司之一。我们在做非常重要的工作，我非常认真对待。所以一些实际的问题，比如我们如何看待继任计划？我以‘不相信继任计划’而闻名。原因不是因为我长生不老，而是因为如果你担心继任计划，如果你为此焦虑，那你应该怎么做？你得把问题拆解到底。今天最重要的事，如果你关心公司在你之后的未来，就是尽可能频繁、持续地传递知识、信息、洞察、技能和经验。这就是为什么我不断在团队面前推理一切。每一次会议都是一次推理会议。我在公司内外的每一刻都在尽可能快地向人们传递知识。我学到的东西从不会在我桌上停留超过一瞬间。我立刻传递出去：‘天哪，这太酷了。我还没完全学会，就已经指给别人了。快来看看，这太酷了，你会想学的。’所以我不断传递知识，赋能他人，提升周围每个人的能力。这样我期望的结果是——我希望——我在工作中死去，最好是瞬间的，没有长期的痛苦。

I really don't want to die. I have a great life. I have a great family. I have really important work. This is not a once-in-a-lifetime experience that suggests it has been experienced by many people, just not one person. This is a once-in-humanity experience what I'm going through. Nvidia is one of the most consequential technology companies in history. We're doing very important work. I take it very seriously. And so some of the practical things, like how do we think about succession planning? I'm famous for saying that I don't believe in succession planning. The reason isn't because I'm immortal. The reason is because if you're worried about succession planning, if you're anxious about it, then what should you do about it? You break it all the way back down. The most important thing you should do today, if you care about the future of your company after you, is to pass on knowledge, information, insight, skills, experience as often and continuously as you can. That's why I continuously reason about everything in front of my team. Every single meeting is a reasoning meeting. Every moment I spend inside the company or outside is about passing on knowledge to people as fast as I can. Nothing I learn ever sits on my desk longer than a fraction of a second. I'm passing that information. 'Oh my gosh, this is cool. Before I even finish learning it all myself, I've already pointed it to somebody else. Get on this. This is so cool. You're going to want to learn this.' So I'm constantly passing knowledge, empowering people, elevating the capability of everybody around me so that the outcome I seek, I hope for, is that I die on the job, and hopefully instantaneously, with no long periods of suffering.

从一个粉丝的角度，鉴于你对文明的巨大积极影响，我当然希望你继续下去，而且看着你所做的一切也很有趣。创新的速度令人难以置信。我是工程的超级粉丝。英伟达持续进行着如此多不可思议的工程。看着就很有趣。这是对人类的庆祝，对伟大建设者的庆祝，对伟大工程的庆祝。所以它代表着某种特别的东西。我希望你和英伟达继续前进。关于人类这一切，什么给了你希望？当你展望未来，思考 10 年、20 年、50 年、100 年后，什么给了你希望？

Well, from a fan perspective, given your enormous positive impact on civilization, of course I hope you keep going, but also it's just fun to watch what you're doing. The rate of innovation is incredible. I'm a huge fan of engineering. So much incredible engineering is continuously being done by Nvidia. It's just fun to watch. It's a celebration of humanity, a celebration of great builders, a celebration of great engineering. So it represents something special. I hope you and Nvidia keep going. What gives you hope about this whole thing we got going on about humanity? About the future of humanity when you look out and think about the future quite a bit, when you look out 10, 20, 50, 100 years from now, what gives you hope?

我一直对人类的善良、慷慨、同情心和能力充满信心。我一直非常确信这一点。有时甚至过于自信，以至于被利用。但这从未让我改变。我总是从‘人们想做好事’开始。人们想帮助他人。而绝大多数情况下，我被证明是对的，不断被证明是对的，而且常常超出预期。所以我对人类的能力完全有信心。给我带来巨大希望的是，我现在看到的事情是可能的，而且根据我们正在做的事情推断，很可能发生。我们想解决的事情太多了，想解决的问题太多了，想建造的东西太多了。有太多美好的事情现在触手可及，而且在我有生之年就能实现。你不可能不为此感到浪漫。你怎么能不浪漫呢？期待疾病终结是合理的。期待污染大幅减少是合理的。期待以光速旅行确实在我们的未来，也是合理的。

I've always had great confidence in the kindness, the generosity, the compassion, the human capacity. I've always been extremely confident of that. Sometimes more so than I should, and I get taken advantage of. But it doesn't ever cause me not to. I start always with that people want to do good. People want to help others. And vastly I am proven right, constantly proven right, and often exceeds my expectations. So I have complete confidence in the human capacity. The things that give me incredible hope are what I see now as possible, and as I extrapolate based on the things we're doing, what will very likely happen. There are so many things we want to solve, so many problems we want to solve, so many things we want to build. There are so many good things we want to do that are now within our reach, and within the reach of my lifetime. You just can't possibly not be romantic about that. How can you not be romantic about that? The fact that it's a reasonable thing to expect the end of disease. It's a reasonable thing to expect that pollution will be drastically reduced. It's a reasonable thing to expect that traveling at the speed of light is actually in our future.

是啊。活在这个时代多么激动人心。

Yeah. What an exciting time to be alive.

是的。

Yeah.

真的，真的。

Like truly, truly.

所以你怎么能不浪漫呢？期待疾病终结是合理的。期待污染大幅减少是合理的。期待以光速旅行确实在我们的未来，也是合理的。

So how can you not be romantic about that? The fact that there is a reasonable thing to expect the end of disease. It's a reasonable thing to expect. It's a reasonable thing to expect that pollution will be drastically reduced. It's a reasonable thing to expect that traveling at the speed of light is actually in our future.

而且，你知道，不是长距离而是短距离。人们问我怎么做。首先，很快我会把一个类人机器人送上宇宙飞船，那是我的机器人，我们会尽快发射它，它会在飞行过程中不断改进和增强。然后到时候，我所有的意识都已经——你知道，我生命中的大部分内容已经上传到互联网上了。把我所有的收件箱，我做过的一切，我说过的一切都拿去。它们被收集起来，成为我的 AI。然后到时候，我们就以光速发送它，追上我的机器人。

And then, you know, not for long distances but short distances. You know, people ask me how. Well, first of all, very soon I'm going to put a humanoid on a spaceship, and it's going to be my humanoid, and we're going to send it out as soon as possible, and it's going to keep improving and enhancing along the flight. And then when it's time, all of my consciousness has already been—you know, so much of my life has been uploaded on the internet. Take all my inbox, take everything that I've done, everything I've said. It's been collected and become my AI. And I'm just, you know, when the time comes, we just send that at the speed of light, catch up with my robot.

哦，太棒了。但对我来说，这有点偏向应用，不过从好奇心和最大化视角来看，所有这些谜团都充满了迷人的科学问题。理解生物机器就在眼前。不是十年，可能五年。

Oh, that's brilliant. I mean, but for me, that's sort of application focused, but also for me the curiosity, the maximizing perspective, I just all of those mysteries. It's so much fascinating scientific questions there. Understanding the biological machine is right around the corner. It's not 10 years. It's 5 years probably.

然后你的生物机器，人类心智，以及破解物理学、理论物理学，这太令人兴奋了。

And then your biological machine, the human mind, and cracking physics, theoretical physics open. It's so exciting.

解释意识，那将非常棒。

Explaining consciousness, that one would be awesome.

而且这一切都在我们的掌握之中。

And it's all within our reach.

是的。

Yeah.

Jensen，非常感谢你多年来所做的一切。感谢你为世界所做的一切。感谢你做你自己。我看得出你是一个伟大的人，祝你今年取得巨大成功。作为一个粉丝，我等不及了。我迫不及待想看你下一步做什么，希望能在台湾见到你。非常感谢你今天接受采访。

Jensen, thank you so much for everything you've done over the years. Thank you for everything you're doing for the world. Thank you for being who you are. I can tell you're a great human being and I wish you incredible success this year. I can't wait as a fan. I can't wait to see what you do next and hopefully I'll see you in Taiwan. And thank you so much for talking today.

谢谢你，Lex。我度过了一段美好的时光。另外，我还想说一件事：感谢你做的所有访谈，你带来的深度、尊重，以及你为揭示那些你多年来采访过的杰出人物所做的研究。我非常享受这些访谈。作为一个创新者，创造出这种长篇形式——难以置信——而且它如此引人入胜。总之，感谢你所做的一切。

Thank you, Lex. I had a great time. And also, if I could just say one more thing: thank you for all the interviews that you do, the depth, the respect that you go through with, and the research that you do to reveal, for all of us, the amazing people that you've interviewed over the years. I've enjoyed them immensely. And as an innovator, to have created this long form—unbelievable—and yet it's just captivating. So anyways, thank you for everything you do.

这意义重大。谢谢你，Jess。

It means the world. Thank you, Jess.

谢谢你，Lex。

Thank you, Lex.

感谢收听本期与黄仁勋的对话。要支持本播客，请查看描述中的赞助商信息，你还可以找到联系我、提问、反馈等链接。现在，让我用艾伦·凯的一句话作为结尾：‘预测未来的最好方法就是创造未来。’感谢收听，期待下次再见。

Thank you for listening to this conversation with Jensen Huang. To support this podcast, please check out our sponsors in the description, where you can also find links to contact me, ask questions, give feedback, and so on. And now let me leave you with some words from Alan Kay: 'The best way to predict the future is to invent it.' Thank you for listening and hope to see you next time.