章节 1：Google的逆袭心态与Gemini 3的里程碑

📝 本节摘要：

本节对话作为开场，聚焦于Google最新发布的模型 Gemini 3。嘉宾Amin Vahdat分享了Google在过去三年AI竞赛中的心路历程，坦言Google曾一度被视为“处于劣势的一方”（Underdog）。他描述了内部团队如何在外界质疑声中坚持研发，最终通过Gemini 3在各项基准测试中重回巅峰，成功证明了唱衰者（Naysayers）的错误。

[原文] [Jeetu Patel]: how's life life is exciting i mean it's actually never been more exciting i think I know you feel it i know you all feel it in the audience uh I'm I'm just thrilled actually Gemini 3 is kind of uh crazy times at Google huh

[译文] [Jeetu Patel]: 生活怎么样？生活很令人兴奋，我是说，实际上从未像现在这样令人兴奋。我想我知道你有这种感觉，我知道观众席上的各位都有这种感觉。呃，我真的非常激动。Gemini 3（的发布）... Google 现在的日子有点疯狂，是吧？

[原文] [Amin Vahdat]: gemini 3 has been awesome i mean I think it's been uh you all know it's uh uh state-of-the-art across essentially all the benchmarks we're really proud of that but I think for us you know we've been on this three plus year journey at Google gemini one came out twoish years ago uh I mean I think that for me I'll just speak for myself i'm I'm always rooting for the underdog and it was great to be a place where Google was the underdog you're always rooting for the underdog rooting for the underdog it was great to be at a place where Google was the underdog all of us believed internally actually all of us believed internally but for us to uh come to this journey the race is not I mean the race is is like any one right but uh we're proud of where we've come over the past few years

[译文] [Amin Vahdat]: Gemini 3 非常棒。我的意思是，我想大家都知道，它在几乎所有的基准测试（benchmarks）中都处于最先进水平（state-of-the-art），我们对此感到非常自豪。但在我看来，对于我们来说，你知道我们在 Google 已经走过了三年多的旅程。Gemini 1 大约是两年前发布的。呃，我是说，就我个人而言，我只代表我自己说，我总是支持处于劣势的一方（underdog）。能在 Google 处于劣势的时候身在其中，感觉很棒。你总是支持劣势方，支持劣势方。能在 Google 被视为劣势方的时候待在这里感觉很棒。我们在内部都坚信，实际上我们所有人在内部都坚信这一点。但对于我们要走完这段旅程... 这场竞赛并不是... 我的意思是这场竞赛就像任何其他竞赛一样，对吧？但呃，我们对过去几年取得的成就感到自豪。

[原文] [Jeetu Patel]: you should you should be very proud it was actually um it's it by the way you proved the naysayers wrong in some pretty profound ways because people had almost given up on Google in certain areas and then all of a sudden they're they're back and so it's a wild success and congratulations to you couldn't have better people kind of group of people

[译文] [Jeetu Patel]: 你们应该，你们应该感到非常自豪。实际上，呃... 顺便说一句，你们以一种非常深刻的方式证明了唱反调的人（naysayers）是错的。因为在某些领域，人们几乎已经放弃了 Google，然后突然之间，他们（Google）又回来了。所以这是一个巨大的成功（wild success），祝贺你们。不可能有比这更好的人员... 这样的一群人了。

章节 2：全栈协同——Google的“秘密武器”

📝 本节摘要：

主持人 Jeetu 提出 Google 最大的优势在于其作为“全栈公司”的独特性——从底层的 TPU 芯片、分布式系统架构，一直到服务数十亿用户的顶层应用。Amin 确认这是 Google 成功的最大单一因素，称其为“秘密武器”。他详细解释了 TPU 并非孤立制造，而是与 DeepMind 及搜索、YouTube 等业务部门“协同设计”（Co-design）的产物。面对硬件和软件长达 2-3 年的研发周期，这种跨部门的深度协作使他们能够预判未来的技术走向，从而优化设计。

[原文] [Jeetu Patel]: so talk to us about your um if if I think about one of Google's superpowers it is probably the most full stack company that exists in the market all the way from the TPUs that you build to distributed systems architecture to the um you know and every layer of the stack all the way to the applications that touch the user and the consumer at scale to billions of users how much of that is is actually a massive compounding contributor uh of your success

[译文] [Jeetu Patel]: 所以跟我们要谈谈，如果让我思考 Google 的一个超能力，那可能就是市场上现存的最全面的全栈（full stack）公司，从你们构建的 TPU 到分布式系统架构，再到——你知道的，栈的每一层——一直到接触数十亿用户的应用程序。这对你们的成功来说，究竟是一个多么巨大的复合贡献因素？

[原文] [Amin Vahdat]: it's a it's a great question G2 and I think that this is actually one of if not the biggest part in other words if you look at uh our stack of course we're proud of our TPUs we're proud of Gemini we're proud of our distributed systems architecture we're proud of our data center architecture our power delivery mechanisms etc but uh the real I would say secret weapon that we have is that we get to work together across the stack at the company to solve the end problem that's what is actually most gratifying

[译文] [Amin Vahdat]: 这是一个很好的问题，Jeetu（注：原文听译为 G2）。我认为这实际上是其中之一，如果不是最大的部分的话。换句话说，如果你看我们的技术栈，当然我们为我们的 TPU 感到自豪，我们为 Gemini 感到自豪，我们为我们的分布式系统架构感到自豪，我们为数据中心架构和电力输送机制等感到自豪。但我会说，我们真正的秘密武器（secret weapon）是，我们在公司内部能够跨越整个技术栈协同工作，以解决最终的问题，这才是最令人欣慰的。

[原文] [Amin Vahdat]: so if you look at u TPUs they're not designed in isolation they're co-designed with a deep mind but also taking input from all of the different use cases search ads YouTube you work very closely with Dennis on this on a on a regular basis you know one of the best parts of my job is working with them as very closely

[译文] [Amin Vahdat]: 所以如果你看 TPU，它们不是孤立设计的，它们是与 DeepMind 协同设计的（co-designed），同时也吸取了所有不同用例的输入，如搜索（Search）、广告（Ads）、YouTube。你定期与 Demis（注：指 Google DeepMind 创始人 Demis Hassabis，原文误听为 Dennis）非常紧密地合作。你知道，我工作中最棒的部分之一就是与他们非常紧密地合作。

[原文] [Amin Vahdat]: and if you look at it uh the infrastructure and the models I mean we're in this um great challenging but great place where uh we wind up being the limiting factor in terms of what the company can deliver you being your team because of the infrastructure that and how fast of course we're I I hope that we're also appreciated yeah yeah but it is I mean in other words if we if we had more it would be a positive it would be a big positive so yes Demis and I were uh uh speak regularly our teams engage deeply

[译文] [Amin Vahdat]: 既然你看基础设施和模型，我的意思是，我们正处于一个既充满巨大挑战又很棒的位置，我们最终成为了公司交付能力的限制因素——“你”指的是你的团队，因为基础设施以及（部署）速度。当然，我希望我们也受到赞赏（Jeetu: 是的，是的）。但确实如此，我的意思是换句话说，如果我们有更多（算力/资源），那将是积极的，巨大的积极因素。所以是的，Demis 和我定期交谈，我们的团队深度参与。

[原文] [Amin Vahdat]: so it's this uh co-design process that is so important for hardware in particular but also for software i mean the kinds of software that we're building today they have two three-year lead times the hardware has as you well know two three year lead times so the ability to predict the future working handinhand with the research teams that are developing the models and say ask and be able to answer where are things likely to go of course no one can predict the future that far out but if we know what the probability distribution function looks like what the paro of prob outcomes looks like y then we can actually evaluate designs relative to one another looking at to what must might be most likely

[译文] [Amin Vahdat]: 所以正是这种协同设计过程，对于硬件来说尤为重要，对软件也是如此。我是说，我们今天构建的这类软件有两三年的交付周期（lead times），而硬件——正如你所知——也有两三年的交付周期。所以能够预测未来，与开发模型的研究团队携手合作，去问并能够回答“事情可能会往哪里发展”。当然，没有人能预测那么远的未来，但如果我们知道概率分布函数（probability distribution function）是什么样子的，可能的结果分布（Pareto of outcomes）是什么样子的，那么我们实际上就可以相对地评估设计，看看什么是最可能发生的。

章节 3：专用芯片(XPU)时代的算力哲学

📝 本节摘要：

主持人 Jeetu 指出 TPU 是 XPU（专用加速器）领域的首个巨大成功，询问 Google 是否计划将其向公开市场出售。Amin 明确表示 TPU 将完全保留作为 GCP（Google Cloud Platform）的产品服务，但他同时强调了与 Nvidia 的深度合作伙伴关系，指出 GPU 依然是 Google 云服务的核心支柱。Amin 提出了计算领域的“阶段性突变”（Phase Change）：行业正从“一刀切”的通用架构，转向针对特定工作负载进行极致优化的专用架构时代。

[原文] [Jeetu Patel]: and so let's let's talk about TPUs because I think it it was um it's the first kind of massive success of XPUs that that has happened and you folks have done a fantastic job on making sure that you're changing the scarcity model of GPUs to some degree and especially on the inferencing side so walk us through what happens on the firstly do you sell TPUs to will you sell TPUs to the open market or are you going to keep it within the GCP domain

[译文] [Jeetu Patel]: 那么让我们谈谈 TPU，因为我认为它……它是 XPU（专用处理器）取得的第一次巨大成功，你们在确保某种程度上改变 GPU 的稀缺模式方面做得非常出色，特别是在推理（inferencing）方面。所以请给我们讲讲发生了什么，首先，你们会向公开市场出售 TPU 吗？还是会将其保留在 GCP（谷歌云平台）领域内？

[原文] [Amin Vahdat]: tpus are a GCP product offering uh entirely but I do want to note that Nvidia and GPUs are a huge GPU uh GCP product offering in other words for us what we're focused on is solving customer problems and deep partnership with Nvidia uh a lot of our success at Google has been as a result of that partnership with Nvidia and GPUs

[译文] [Amin Vahdat]: TPU 完全是 GCP 的产品服务。但我确实想指出，Nvidia 和 GPU 也是 GCP 巨大的产品服务。换句话说，对我们而言，我们专注于解决客户问题，并与 Nvidia 建立深度合作伙伴关系。呃，Google 的很多成功都是与 Nvidia 和 GPU 合作的结果。

[原文] [Amin Vahdat]: so what we're focused on is solving customer problems whether internal or external and there may be situations where one product is more appropriate than another for a particular use case so that sort of vertical integration that whole stack solution that you talked about starts with what problem is the customer trying to solve and then we work our way up and down in order to deliver ideally of course the best solution with whatever it is that we put together

[译文] [Amin Vahdat]: 所以我们专注于解决客户问题，无论是内部还是外部的。可能在某些情况下，对于特定的用例，某种产品比另一种更合适。所以那种垂直整合（vertical integration），即你谈到的全栈解决方案，始于客户试图解决什么问题，然后我们上下求索，以便理想地交付……当然是用我们要组合的任何东西提供最佳解决方案。

[原文] [Amin Vahdat]: so for me actually the most exciting part of this is how we're able to specialize in other words for me whether it's GPUs TPUs product offerings from many many other uh companies and actually it's exploding as as you all know the real phase change here is actually now we don't have to have general purpose one-sizefits-all architecture anymore and we're able to specialize for the individual use cases and even invent new wholly new architectures hardware or software for it

[译文] [Amin Vahdat]: 所以对我来说，实际上这其中最令人兴奋的部分是我们能够如何进行专业化（specialize）。换句话说，对我而言，无论是 GPU、TPU，还是来自许多其他公司的产品服务——实际上正如大家所知，这方面正在爆发——这里真正的阶段性变化（phase change）实际上是，现在我们不再需要通用的、“一刀切”（one-size-fits-all）的架构了。我们能够针对个人用例进行专业化，甚至为此发明全新的架构、硬件或软件。

[原文] [Jeetu Patel]: do do you see that for every class of model and for every variation that you could actually start to etch different silicon that's optimized for that over time or is that

[译文] [Jeetu Patel]: 你是否认为对于每一类模型和每一个变体，随着时间的推移，实际上可以开始蚀刻（制造）针对其优化的不同芯片？还是说……

[原文] [Amin Vahdat]: it's a fantastic question it's a fantastic question and as we know um the more you specialize the more efficient you can be for a particular workload

[译文] [Amin Vahdat]: 这是一个极好的问题，这是一个极好的问题。正如我们所知，你越专业化，针对特定工作负载的效率就越高。

章节 4：挑战物理极限——缩短硬件迭代周期

📝 本节摘要：

面对AI模型的快速迭代，Amin Vahdat 提出了一个激进的设想：将硬件从设计到规模化部署的周期从目前的3年缩短至 3个月。他认为缩短周期是实现极致“专业化”（Specialization）的关键，而专业化能带来 10倍（Factor of 10） 的能效提升。针对主持人 Jeetu 关于供应链复杂性和设备折旧（Amortization）的担忧，Amin 反驳称目前的折旧年限并非不可打破的“自然法则”。他强调，虽然3个月目前极难实现，但只要能大幅缩短周期，就能通过牺牲通用性（Generality）来换取特定工作负载下的极致性能与效率。

[原文] [Amin Vahdat]: now actually the biggest thing that I' uh the answer to your question would be yes if we could only cut down the lead time of hardware design to delivery by a factor of 10 like right now today from the time we say hey here's an amazing new piece of hardware to the time where we have it at scale in the data center not quite speed of light but really fast would be three years

[译文] [Amin Vahdat]: 现在实际上最大的一点是……我对你问题的回答是“是的”，但这前提是我们能把硬件设计到交付的周期缩短10倍。比如此时此刻，从我们说“嘿，这是一个令人惊叹的新硬件”，到我们在数据中心大规模部署它，如果速度真的很快（虽然不到光速），目前大约需要三年。

[原文] [Jeetu Patel]: predicting and what do you think that that gets to be in in a 10ear period yeah

[译文] [Jeetu Patel]: 预测一下，你认为在10年内这会变成多久？

[原文] [Amin Vahdat]: so in a 10-year period I think it I don't know where it gets to be but if we could in a 10-year period get that down to let's say three months

[译文] [Amin Vahdat]: 在10年内，我想……我不知道具体会变成多少，但如果我们能在10年内把它缩短到，比如说，3个月。

[原文] [Jeetu Patel]: oh wow three months that's aggressive it's I don't know how to do it right i don't know if anybody does

[译文] [Jeetu Patel]: 哇，3个月，这太激进了。我不知道该怎么做，对吧？我不知道是否有人能做到。

[原文] [Amin Vahdat]: but if we could get it down to three months actually we from a efficiency capability change the world perspective it would be a radical radically different place

[译文] [Amin Vahdat]: 但如果我们能将其缩短到3个月，实际上从效率、能力和改变世界的角度来看，那将是一个根本上完全不同的世界。

[原文] [Jeetu Patel]: but would would that make uh um consumption really hard at at a three-month cycle of chips how would you because the entire value chain has to shift right because the way in which you actually incorporate the chips in the data center your capacity planning how you actually get that to the next because your your shelf life um and the amortization of those chips still needs to be you know three five seven years

[译文] [Jeetu Patel]: 但是，如果是3个月的芯片周期，这会不会让“消费”变得非常困难？你怎么处理……因为整个价值链都必须转变，对吧？因为你将芯片整合进数据中心的方式、你的容量规划、你如何进行下一次迭代……因为你的产品寿命，呃，以及这些芯片的摊销（amortization）仍然需要是，你知道的，3年、5年或7年。

[原文] [Amin Vahdat]: well so this is uh it may need to be three five seven years i don't think that 6 years or 5 years or whatever we have encoded right now is a law of nature in terms of what the depreciation cycle would be

[译文] [Amin Vahdat]: 嗯，这……可能确实需要3年、5年或7年。但我不认为我们现在设定（encoded）的6年或5年折旧周期是某种自然法则（law of nature）。

[原文] [Jeetu Patel]: and and are you saying that that'll be programmability based or is it going to be actual t you know like you'll have different tape outs every 3 months that

[译文] [Jeetu Patel]: 那么你是说这将基于可编程性（programmability），还是说会是实际的……你知道，就像每3个月就有一次不同的流片（tape outs）？

[原文] [Amin Vahdat]: Well so that that's the key question i would imagine that you would need different uh uh and if it's programmable then it's not going to be specialized right right right and if it's not specialized it's not going to be uh optimized for the workload but so three months is radical i don't know how to do it two years seems achievable that's a third of the time yeah right and then now okay 18 months okay you probably every many people in the audience are starting to get nervous maybe that's impossible 12 months seems not doable but the point here is the more we can pull it in the more we can specialize the more we can specialize the more efficiency we can deliver

[译文] [Amin Vahdat]: 这就是关键问题。我想你需要不同的……呃，如果是可编程的，那它就不是专用的（specialized），对吧？（Jeetu: 对，对）。如果不专用，它就不会针对工作负载进行优化。虽然3个月很激进，我不知道怎么做，但2年似乎是可以实现的——那是（目前）时间的三分之一。好的，或者是18个月。观众席里的很多人可能开始感到紧张了，觉得那是不可能的，12个月似乎不可行。但这里的重点是，我们能把周期缩得越短，我们就能越专业化；我们越专业化，就能提供越高的效率。

[原文] [Amin Vahdat]: and so in other words if you think whether it's um power if I can deliver something that is 2x 5x more power efficient because I've specialized it for a particular workload in other words if you whether it's GPUs or TPUs or your other favorite accelerator for a particular workload power efficiency is at least a factor of 10 now okay what if I had something that was even more specialized i I would imagine actually a factor of 10 is achievable but then I'd have to predict the future three years up

[译文] [Amin Vahdat]: 换句话说，如果你考虑能源，如果我能交付某种能源效率高2倍、5倍的东西，因为我针对特定的工作负载进行了专门设计。换句话说，无论是 GPU、TPU 还是你最喜欢的其他加速器，针对特定工作负载的能效至少能提升10倍。现在，如果我有更加专业化的东西，我想实际上10倍的提升是可以实现的，但我必须提前三年预测未来（才能做到这一点）。

[原文] [Jeetu Patel]: so how does how do economics change with XPUs uh

[译文] [Jeetu Patel]: 那么 XPU（专用处理器）的经济性会如何变化呢？

[原文] [Amin Vahdat]: so again I think it's this factor of 10 factor factor of 10 uh and again picture XPU by the way I include GPUs to to me and I know there's different terms uh etc but to to me specializing the way that we have right now gets you a factor of 10 at least now that might be cost that might be scale that might be power across all dimensions i get that massive uplift but I give up generality i wouldn't go run a database on a XPU correctly

[译文] [Amin Vahdat]: 所以再说一次，我认为是这个“10倍因子”（factor of 10）。再想象一下 XPU——顺便说一句，我把 GPU 也包括在内，对我来说，我知道有不同的术语等等，但对我来说，像我们现在这样进行专业化，至少能给你带来10倍的提升。这可能是成本，可能是规模，可能是能源，在所有维度上。我获得了巨大的提升，但我放弃了通用性（generality）。我不会去 XPU 上运行数据库，对吧。

章节 5：终极前沿——太空数据中心构想

📝 本节摘要：

对话进入了最具科幻色彩的领域——太空数据中心。Amin Vahdat 证实 Google 正在认真探索这一方向。他详细阐述了在太空部署算力的核心优势：利用“太阳同步轨道”实现 24/7 全天候太阳能供电（无需电池、无云层遮挡），利用真空环境解决冷却难题，以及光在真空中传播速度快于光纤从而带来的 50% 延迟降低。尽管面临维护难题（需依赖机器人），且距离实现吉瓦级（Gigawatt）规模可能还需要 5年以上，但 Amin 坚信这是解决地球能源与基础设施扩展瓶颈的必要投资方向。

[原文] [Jeetu Patel]: space the final frontier talk about data centers in space and we we talked about it with Matt Garmin earlier um talk about your point of view on that

[译文] [Jeetu Patel]: 太空，最后的疆域（the final frontier）。谈谈太空数据中心吧，我们之前和 Matt Garmin 也聊过这个。谈谈你对这个的看法。

[原文] [Amin Vahdat]: it's really exciting actually and uh as probably you're aware uh Google is looking into the space a number of uh companies are looking into the space from uh no pun intended from a first principles perspective holds a lot of appeal I mean a sun-synchronous orbit with 247 solar cooling figured out in space yeah so I'll get I'll get to it I'll get to it

[译文] [Amin Vahdat]: 实际上这非常令人兴奋。呃，你可能知道，Google 正在研究这个领域（space），很多公司也在研究这个领域。呃，这并非双关语（pun intended），从第一性原理的角度来看，它非常有吸引力。我是说，太阳同步轨道（sun-synchronous orbit），配合24/7全天候太阳能，以及在太空中解决冷却问题。是的，我会谈到这个，我会谈到这个。

[原文] [Amin Vahdat]: um power um part of that by the way three-year hardware perspective is actually being able to deliver the building and the and the power necessary to hold that chip so we can talk about hey if I can go build the chips in uh three months how am I going to house it all so removing a bottleneck which is perhaps if we can solve all the issues to be able to put XPUs GPUs TPUs whatever it is in space with 247 solar power no need for batteries I mean no no cloud cover right no sunset etc

[译文] [Amin Vahdat]: 呃，电力。顺便说一下，那个“三年硬件周期”的一部分实际上是能够交付容纳芯片所需的建筑和电力。所以我们可以讨论“嘿，如果我能在3个月内造出芯片，我该如何安置它们？”所以这（太空计划）是在消除一个瓶颈。如果我们能解决所有问题，能够把 XPU、GPU、TPU 或任何东西放到太空中，拥有24/7的太阳能电力，不需要电池——我的意思是，没有云层覆盖，对吧？没有日落等等。

[原文] [Amin Vahdat]: now okay cooling and you're 30% more efficient in 30% more efficient uh networking like actually assuming that we're going to connect these satellites together in space you get 50% reduction in latency for for speed of light not having to go through fibers etc now there are many many problems to to solve uh cooling is one of them maintenance is another one maintenance right

[译文] [Amin Vahdat]: 现在的确，冷却方面你能获得30%的效率提升。呃，网络方面，假设我们将这些卫星在太空中连接起来，你能获得50%的延迟降低，因为光速（在真空中传播）不需要穿过光纤等等。当然现在还有很多很多问题要解决，呃，冷却是其中之一，维护是另一个，维护，对吧？

[原文] [Amin Vahdat]: in other words these uh these uh computers accelerators are wonders of nature and uh at least on on our side we're we're certainly expending spending effort in maintaining them but I I view this I think robotics in space would be the way to do it

[译文] [Amin Vahdat]: 换句话说，这些计算机加速器是自然的奇迹。呃，至少在我们这边，我们肯定在花费精力维护它们。但我对此的看法是，我认为太空机器人（robotics in space）将是解决之道。

[原文] [Amin Vahdat]: so uh in yes so if you look at the scale at which uh we're growing whether it's uh terrestrial or in space I I do believe that the current way that we're going about deploying and maintaining our infrastructure is unlikely to scale and it's unlikely to get to the level of reliability velocity and everything else that uh we we need to do it

[译文] [Amin Vahdat]: 所以呃，是的。如果你看看我们增长的规模，无论是在地球上还是在太空中。我确实认为，我们目前部署和维护基础设施的方式不太可能扩展（scale），也不太可能达到我们需要做到的可靠性、速度和其他一切水平。

[原文] [Amin Vahdat]: in other words I think the way that we're deploying infrastructure today is not radically different from how we did it when let's say Google built its first data center in 2002 or so and that was a 10 megawatt like and I remember when this happened people said 10 megawws right that that now of course is people are talking about 10 gigawatts like it's um done deal etc so yeah a factor of a thousand really means that you have to reconsider how you do everything

[译文] [Amin Vahdat]: 换句话说，我认为我们今天部署基础设施的方式，与比如说2002年左右 Google 建立第一个数据中心时并没有根本的不同。那时是10兆瓦（megawatt），我记得当时人们说“10兆瓦？！”对吧？现在当然，人们在谈论10吉瓦（gigawatts），好像这已经是板上钉钉的事了等等。所以是的，一千倍的增长确实意味着你必须重新考虑你是如何做每一件事的。

[原文] [Jeetu Patel]: you think a gigawatt in space is a decade away huh Huh this is a good question um

[译文] [Jeetu Patel]: 你认为太空中实现吉瓦级（算力）还需要十年吗？嗯？这是一个好问题，嗯……

[原文] [Amin Vahdat]: I think that it is greater than 5 years away uh at at this scale but what it is too early to put a um time frame on it is is what I would say i think that it is an idea that is absolutely worth investing in and going after with gusto all right because um we're going to learn a ton from it and I think that it's going to advance the uh state-of-the-art no matter what

[译文] [Amin Vahdat]: 我认为这需要超过5年的时间，呃，在这个规模上。但我会说现在给它定一个时间框架还为时过早。我认为这是一个绝对值得投资并以此为目标全力以赴（going after with gusto）的想法。好吧，因为我们将从中通过学习获得大量知识，而且我认为无论如何它都会推动最先进技术的发展。

这里是为您整理的访谈第6章内容。

章节 6：增长的焦虑——速度、能源与效率悖论

📝 本节摘要：

在这一节中，Amin Vahdat 坦言目前最让他焦虑的是“速度”（Velocity）——即技术迭代与交付的速率，以及能源供应和供应链压力。当被问及行业内普遍存在的“错误共识”时，Amin 提出了一个反直觉的观点：许多人寄希望于“效率提升”来解决算力瓶颈，但现实是，模型能力的增强（如Agent、编程辅助）让用户能够处理更复杂的任务，这导致每一次效率的提升都会被瞬间爆发的新需求所吞噬。他认为单纯依赖效率“救场”可能比人们预期的要遥远得多。

[原文] [Jeetu Patel]: what do you worry about the most right now uh

[译文] [Jeetu Patel]: 你现在最担心的是什么？呃……

[原文] [Amin Vahdat]: it changes by the week it changes by the week and it's velocity is what you worry about velocity is uh uh the generic one that uh cuts across everything in terms of how we're able to do things how we're able to deliver how we're able to iterate

[译文] [Amin Vahdat]: 这每周都在变，每周都在变。而在其中，“速度”（velocity）是你担心的。速度是一个通用的问题，它贯穿于一切之中——关于我们如何做事，我们如何交付，我们如何迭代。

[原文] [Amin Vahdat]: um energy will be on the list in in many uh cases in many weeks uh supply chain uh I mean that the rate that we're all looking to grow the number of things that we discover

[译文] [Amin Vahdat]: 呃，能源在很多情况下、很多周里都会在（担忧）清单上。呃，供应链。我的意思是，既然我们都希望以这样的速度增长，既然我们发现了这么多的事物……

[原文] [Jeetu Patel]: what do you think on memory prices you think uh um so so that's that's been on my list of worries in in certain weeks as well very very exciting times in terms of uh especially as you know the uh split between DRAM and HBM and uh how that affects uh do you see a light at the end of the tunnel or Lipu was saying it's going to be until end of 28

[译文] [Jeetu Patel]: 你怎么看内存价格？你认为……呃，这也是某些周里我担忧清单上的事项。这是非常非常激动人心的时刻，特别是你知道 DRAM 和 HBM 之间的分野，以及这如何影响……你看到隧道尽头的曙光了吗？还是说像 Lipu 说的那样，要等到2028年底？

[原文] [Amin Vahdat]: lipu would know more about it than I do so I think that uh um I I hope he's wrong but but but he knows more than I do so I think I think we might have to pencil that in

[译文] [Amin Vahdat]: Lipu 比我更了解这一点。所以我认为，呃……我希望他是错的，但他比我懂得多。所以我认为我们可能得把这个（2028年）暂定下来。

[原文] [Jeetu Patel]: um where are the constraints and the bottlenecks going to be in the next um next couple years and by the way actually let me let me take a step back things that are absolutely held beliefs of truth right now that you fundamentally disagree with

[译文] [Jeetu Patel]: 呃，接下来几年约束和瓶颈会在哪里？顺便说一句，实际上让我退一步问：目前有哪些被绝对奉为真理的信念，是你根本不同意的？

[原文] [Amin Vahdat]: uh you know I think that one of the things that um a lot of people are counting on right now is efficiency uh winning the day whether that's software efficiency model efficiency hardware efficiency we're investing hugely in in all these domains power efficiency

[译文] [Amin Vahdat]: 呃，你知道，我认为很多人现在指望的一件事是“效率”（efficiency）能最终获胜。无论是软件效率、模型效率、硬件效率——我们在所有这些领域都投入了巨资——还有能源效率。

[原文] [Amin Vahdat]: the amazing thing right now is uh as our capabilities grow as these models become more and more powerful people are doing more with them so in other words every efficiency we deliver and it's uh I mean the rate of improvement on the efficiency side I've never seen anything like it but it it gets consumed and more instantaneously because the capabilities I know you had Mike a very interesting conversation ear earlier the capabilities that these not just the models but now the orchestration around them agents coding so much more are delivering instantly consumes the efficiencies

[译文] [Amin Vahdat]: 现在令人惊奇的是，呃，随着我们能力的增长，随着这些模型变得越来越强大，人们正在用它们做更多的事情。所以换句话说，我们交付的每一份效率——我是说，在效率方面的改进速度，我从未见过像现在这样的——但它被瞬间吞噬了，甚至更多。因为这些能力——我知道你早些时候和 Mike 有过一次非常有趣的谈话——这些不仅仅是模型，还有围绕它们的编排、Agent（智能体）、编程等等提供的能力，瞬间就消耗掉了这些效率。

[原文] [Amin Vahdat]: so I mean I think that a um belief or a hope depending on how you look at it is the efficiencies are going to save the day they will eventually but I'm maybe with lipu on that might be further out than some people think

[译文] [Amin Vahdat]: 所以我的意思是，我认为有一种信念或希望——取决于你怎么看——是效率将会救场。它们最终会的，但在这一点上我可能和 Lipu 站在一起，那可能比一些人认为的要更遥远。

章节 7：AI的真正价值——从“原创性”到“全人类的导师”

📝 本节摘要：

在本节中，对话深入探讨了AI的核心价值。面对关于AI能否产生“原创性洞察”的提问，Amin Vahdat 结合自己曾经的学术经历指出，真正的价值往往不在于绝对的原创，而在于如何高效地通过“站在巨人的肩膀上”来组合观点。他认为AI目前最大的变革在于它能让每个人瞬间拥有跨学科的顶尖专家资源（如计算机博士也能瞬间获得生物学见解）。双方一致认为，“研究与学习”是AI被低估但影响最深远的场景，其终极愿景是实现“为每位患者提供医生，为每位学习者提供导师”的个性化普及。

[原文] [Jeetu Patel]: and then if you think about original insights that we don't have in the human corpus of knowledge that that AI starts to generate becomes the force multiplier rather than just doing something that we do today but doing it slightly better um is are we now generating meaningful original insights from AI that you feel are uh going to start to solve really complicated problems or you feel like we're still a little ways away from that happening

[译文] [Jeetu Patel]: 那么如果你思考一下目前人类知识库中不存在的“原创性洞察”（original insights），AI 开始生成这些洞察并成为“力量倍增器”（force multiplier），而不仅仅是把我们今天做的事情做得稍微好一点。呃，你觉得我们现在是否正在从 AI 那里获得有意义的原创性洞察，从而开始解决真正复杂的问题？还是说你觉得我们离那个阶段还有一段距离？

[原文] [Amin Vahdat]: it's a great question but I think that to me it's um um the the more impactful question right this moment is as a former uh professor uh relatively longtime professor and academic the question I would always ask myself and in academia you get judged on the originality of your idea yeah which which always used to drive me nuts because for me it was always standing on the shoulder of giants like was I really original or did I consumerization of the idea that might be more right and and did you put the right set of ideas together in a way etc

[译文] [Amin Vahdat]: 这是一个很好的问题。但在我看来，此时此刻更具影响力的问题是……作为一个前教授，一个相对长期的教授和学者，我总是问自己这个问题。在学术界，你是根据你想法的原创性来被评判的。（Jeetu: 是的。）这过去总是让我抓狂，因为对我来说，这总是关于“站在巨人的肩膀上”。比如，我真的是原创吗？还是我对这个想法的“消费化”（consumerization）可能更准确？你是否以某种方式将正确的一组想法组合在了一起等等？

[原文] [Amin Vahdat]: so I I think that for me the most exciting thing about this moment with AI and Genai is um even for people like me who are super privileged super lucky to have access to information to have access to experts I'm able to get near instantaneous insights for relatively advanced questions not original insights in other words if I were able to instantaneously contact worldleading experts or top 1% experts I'm sure I'd get equal or better but it's at the point now where across so many different fields I'm able to access incredible information near instantaneously

[译文] [Amin Vahdat]: 所以我认为，对我来说，关于 AI 和生成式 AI（GenAI）在这个时刻最令人兴奋的事情是，即使像我这样非常优越、非常幸运能够接触到信息和专家的人，我也能够针对相对高级的问题获得近乎瞬时的洞察。这不一定是“原创性洞察”。换句话说，如果我能瞬间联系到世界领先的专家或前1%的专家，我肯定能得到同等或更好的答案。但现在的情况是，在如此多的不同领域，我能够近乎瞬时地获取令人难以置信的信息。

[原文] [Amin Vahdat]: and so in other words even if you look at it from a um business impact I might ask a question at work and I'm fortunate to have amazing teams it might take many many people many many days to ask answer my simple question they're they're the smartest people they're working super hard and but now being able to get that same answer in seconds minutes without having to consume the time of many right smart people is that original no no but it's a game changer massively it's a game changer right

[译文] [Amin Vahdat]: 所以换句话说，即使你从商业影响的角度来看，我也许在工作中问一个问题——我很幸运拥有令人惊叹的团队——但这可能需要许多人花许多天来回答我的这个简单问题。他们是最聪明的人，他们工作非常努力。但现在，能够在几秒钟、几分钟内得到同样的答案，而无需消耗许多聪明人的时间。这算是“原创”吗？不，不算。但这彻底改变了游戏规则（game changer），巨大的改变，对吧？

[原文] [Amin Vahdat]: so to me I'm not so worried about is AI going to now be able to outdo humanity with original ideas that's uh we couldn't do on our own maybe that's going to happen maybe um that's not when is it going to happen you stitch them well together you'll actually have meaning right this moment and having access to them right in other words I I I might have a PhD in computer science very fortunate i might understand concepts in computer science but I don't have a PhD in biology in chemistry in finance in medicine etc but having access to to that myself that I think is the game changer

[译文] [Amin Vahdat]: 所以对我来说，我并不那么担心 AI 现在是否能够用我们自己无法做到的“原创想法”来超越人类。也许这会发生，也许……重要的不是它何时发生，而是如果你能把现有的东西很好地缝合在一起，你实际上在此时此刻就能获得意义，并且能够访问它们，对吧？换句话说，我很幸运拥有计算机科学博士学位，我可能理解计算机科学的概念，但我没有生物学、化学、金融或医学等领域的博士学位。但是我自己能够（通过 AI）获得这些领域的知识，我认为这才是改变游戏规则的地方。

[原文] [Jeetu Patel]: and so this is actually a very interesting um use case because if if I were to think about what is the most additive use case in my life it's been research and learning yes right we talk a disproportionate amount about coding we talk a lot about customer support we talk a lot about lot of those use cases research and learning is one that is actually least talked about it is actually the most used it is the most prolific in humanity exactly and we have actually somehow gotten so used to it that we've stopped stopped even giving it credit

[译文] [Jeetu Patel]: 这实际上是一个非常有趣的用例。因为如果我要思考我生活中最具“增益性”（additive）的用例是什么，那就是研究和学习。（Amin: 是的，对。）我们不成比例地谈论编程，我们谈论很多客户支持，我们谈论很多那些用例。但“研究和学习”实际上是被谈论最少的一个，而在人类活动中它实际上是用得最多的，也是最丰富的。（Amin: 确实。）实际上我们不知何故已经对此习以为常，甚至不再认为它有什么了不起。

[原文] [Amin Vahdat]: and I think that that's I mean the the way I like to cast it uh when I when I talk to my team is we have the opportunity we're not there and there's lots of challenges i also heard the discussion on safety but we have uh and security we have the opportunity to deliver a doctor for every patient and a teacher for every learner and be able to specialize it to the needs of the individual the business use cases we talked about as well so I mean this will be a gamecher in other words as you said for you you're you're learning oriented you want to learn you're you're very good at it etc uh being able to open this up to everybody and I think we're actually at the cusp of that

[译文] [Amin Vahdat]: 我认为这就是……我的意思是，当我与团队交谈时，我喜欢这样描述它：我们有机会——虽然还没到那一步，还有很多挑战，我也听到了关于安全的讨论——但我们有机会为每一位患者提供一名医生，为每一位学习者提供一名老师，并能够根据个人的需求进行专业化定制。我们也谈到了商业用例。所以我的意思是，这将彻底改变游戏规则。换句话说，正如你所说，你是以学习为导向的，你想学习，你非常擅长学习等等。呃，能够将这种能力向所有人开放，我认为我们实际上正处于这个时代的风口浪尖（cusp）。

[原文] [Jeetu Patel]: yeah and then similarly being able to proactively deliver is it already opened up at this point i think that it is very very close i uh um if you have the internet you've got it if you have the internet and if you now I think the last bit that is missing is the uh personalization to in other words and it's already we seeing hints of this right where okay G2 he likes information presented to him like this i mean more or less the same but a little bit differently yeah yeah i think we're almost there necessarily if we could do it for health if we can do it for business intelligence etc we're and it's not as far away as uh some might think

[译文] [Jeetu Patel]: 是的，同样地，能够主动地交付。在这一点上它已经开放了吗？我认为非常非常接近了。如果你有互联网，你就拥有了它。如果你有互联网……我认为现在最后缺失的一块是“个性化”。换句话说——我们要么已经看到了这方面的迹象，对吧？比如“好的，Jeetu（G2）喜欢这样呈现给他的信息”。我的意思是，大体相同，但又有些微不同。是的，是的，我认为我们几乎就要到了。如果我们能在健康领域做到这一点，如果我们能在商业智能等领域做到这一点……这并不像一些人想象的那么遥远。

章节 8：摩尔定律的回归与未来展望

📝 本节摘要：

在访谈的最后部分，Amin Vahdat 对未来几年的技术演进做出了大胆预测。他将当前 AI 模型的迭代速度比作 “摩尔定律全盛时期” 的 CPU 发展，甚至认为现在的速度（每 3-6 个月性能翻倍）比当年的摩尔定律（每 18 个月）还要快。他强调，这不会是一个 “赢家通吃” 的市场，竞争（如 Claude、ChatGPT、Gemini）正在推动整个行业共同进步。最后，他指出这场 AI 革命的影响力将远超互联网，对于基础设施从业者而言，现在是 “书写未来规则” 的绝佳历史时机。

[原文] [Jeetu Patel]: so if if you were to um get people excited about the next next few years and what's happening in Gemini 4 and Gemini 5 without revealing road maps um but go a little bit more specific than they're just going to get better as models which we all know um what do you think we could expect and what turn of crank can we expect is it a 10x better with every model is it 25% better is it 100x better and does that um does that improvement compound over time or does it actually start to flatten as a curve at this point

[译文] [Jeetu Patel]: 那么，如果你要让人们对未来几年感到兴奋，关于 Gemini 4 和 Gemini 5 正在发生什么——在不透露路线图的前提下，但要比“模型只是会变得更好”（我们都知道这点）更具体一点。呃，你认为我们可以期待什么？我们可以期待什么样的迭代幅度（turn of crank）？是每个模型好10倍？是好25%？还是好100倍？这种改进是随着时间推移复利增长，还是在这个点上曲线开始趋于平缓？

[原文] [Amin Vahdat]: I don't see any slowdown and so what I would say is I mean the the capability of the models it's hard to put numbers on them uh but I would say that it definitely feels the same as it did in the heyday of um uh CPUs Moors law we have Lipu in the audience where you know every 18 months you couldn't wait to get your hand on the latest CPU because everything got twice as good for the same cost

[译文] [Amin Vahdat]: 我没有看到任何放缓的迹象。所以我得说，我是指模型的能力，很难给它们定一个具体的数字。但我会说，这种感觉绝对像是在 CPU 摩尔定律（Moore's Law）的全盛时期。Lipu 也在观众席里，你知道那时候每隔18个月，你就迫不及待地想拿到最新的 CPU，因为所有东西都在同样的成本下变得好了一倍。

[原文] [Amin Vahdat]: i think that where we are with models every three six months I don't have a quantification around it but things feel like they're getting twice as good even faster than

[译文] [Amin Vahdat]: 我认为我们在模型方面目前处于一种状态，大概每三到六个月——我没有具体的量化数据——但感觉事物变得好一倍的速度甚至比那时候（摩尔定律）还要快。

[原文] [Jeetu Patel]: and do you think eval are getting better proportionately to actually gauge whether or not the models are in fact improving at the level that we

[译文] [Jeetu Patel]: 那你认为评估（evals）是否也在成比例地变好，以便真正衡量模型是否确实在我们预期的水平上改进？

[原文] [Amin Vahdat]: the eval are getting really good evals are getting really good because I think they're increasingly focused on real world use and actually we have enough data where the models didn't do well in the past and so we can actually now say hey you know what for these um hard cases the ones where the models maybe struggled uh some before they they generally do well for for many many use cases how much improvement do they deliver so it's hard to say okay across all evals the model is twice as good

[译文] [Amin Vahdat]: 评估正变得非常好，评估变得非常好。因为我认为它们越来越专注于现实世界的用途。实际上我们有足够的数据显示模型过去在哪里做得不好，所以我们现在实际上可以说：“嘿，你知道吗，对于这些困难的案例——那些模型以前可能有些挣扎的地方——它们（现在）通常在很多很多用例中都做得很好。”虽然很难说“好吧，在所有评估中模型都好了一倍”……

[原文] [Amin Vahdat]: but I mean I can tell you and it's the same by the way it's amazing time because whether it's claude whether it's Jad GPT whether it's Gemini they're all getting better and they're all making one the I I would say the competitive environment is also making everyone better uh and this so this is fantastic to see where release after release you're you're now you're feeling like you're gaining more insight you're gaining more capability and it's able to go further deeper faster

[译文] [Amin Vahdat]: 但我的意思是，我可以告诉你——顺便说一句，这同样是一个令人惊叹的时刻——因为无论是 Claude，无论是 ChatGPT，还是 Gemini，它们都在变得更好。而且我想说，这种竞争环境也让每个人都变得更好。呃，所以这是非常棒的，看到一个版本接一个版本发布，你现在感觉你在获得更多的洞察力，你在获得更多的能力，而且它能够走得更远、更深、更快。

[原文] [Jeetu Patel]: what question did I not ask you that I should have and what advice would you give to this audience as well uh

[译文] [Jeetu Patel]: 我还有什么应该问但没问的问题吗？以及你会给这里的观众什么建议？

[原文] [Amin Vahdat]: I mean you you asked the fantastic questions what what I would say is that Oh Gemini yeah well no no no i wouldn't say that actually but I mean for me uh of course I I love Gemini i love uh uh what we're doing there but I don't this is very sincerely this is not going to be a winner takes all environment this is going to be you all have heard it uh the biggest revolution since the internet u I I remember the internet you remember the internet very well it was a pretty big deal like in other words I think we're at the point now where uh those of us who have children they wouldn't be able to recognize how th those of us born before the internet exploded lived right the world is just a radically different place

[译文] [Amin Vahdat]: 我是说，你已经问了非常棒的问题。我想说的是……噢，Gemini（笑）。不不不，我其实不会那么说。我的意思是，对我来说，当然我爱 Gemini，我爱我们在那里所做的一切。但我非常真诚地认为，这不会是一个“赢家通吃”（winner takes all）的环境。这将是——你们都听说过——自互联网以来最大的革命。呃，我记得互联网时代，你也非常清楚地记得互联网时代，那是一件相当大的事。换句话说，我认为我们现在正处于这样一个节点：我们这些有孩子的人，孩子们将无法理解我们在互联网爆发之前出生的人是如何生活的，对吧？世界将是一个根本不同的地方。

[原文] [Amin Vahdat]: this is going to be that but much much much bigger and in technology there's never been a better time to uh be working be contributing to be making an impact uh whether you're working at the top of the stack or whether you're working in infrastructure like uh like myself like of course all the amazing things that Cisco is doing the great thing about infrastructure is whether it's the internet or whether it's AI it's high in demand

[译文] [Amin Vahdat]: 这次也会是那样，但规模要大大大得多。在科技界，从来没有比现在更好的时间去工作、去贡献、去产生影响。无论你是在技术栈的顶层工作，还是像我一样在基础设施领域工作——当然就像 Cisco 正在做的所有令人惊叹的事情一样。关于基础设施最棒的一点是，无论是互联网还是 AI，它都有极高的需求。

[原文] [Amin Vahdat]: so what maybe you didn't ask this but what I would say is the opportunity to write the book for how we um support this revolution from a technical perspective i mean it's singular right and we're we're going to literally be inventing the future in terms of how these services how these agents are going to be delivered

[译文] [Amin Vahdat]: 所以，也许你没问这个，但我想说的是：从技术角度去书写我们如何支持这场革命的篇章（write the book），这个机会是非凡的（singular），对吧？我们将真正地发明未来——关于这些服务、这些智能体（agents）将如何被交付。

[原文] [Amin Vahdat]: i mean I I have to just thank you um I think it'd be remiss if I didn't thank you for all the great work that we've done together as two organizations and I want to see that partnership just flourish over time and it wouldn't have happened without your support and thank you for um your entire team and how they work with our silicon team and our kind of um our networking team so really appreciate it thank you for being here hopefully you'll come back again absolutely would love to yeah thank you thank you very much thanks so much

[译文] [Amin Vahdat]: 我的意思是，我必须感谢你。呃，如果不感谢我们两个组织共同完成的所有出色工作，那就是我的失职。我希望看到这种伙伴关系随着时间的推移蓬勃发展。没有你的支持，这一切都不可能发生。谢谢你，呃，谢谢你的整个团队，以及他们如何与我们的芯片团队和我们的网络团队合作。真的非常感谢。

[Jeetu Patel]: 谢谢你来到这里，希望你能再来。

[Amin Vahdat]: 绝对愿意。是的，谢谢。

[Jeetu Patel]: 非常感谢，多谢。