"是否存在AI泡沫？" Gavin Baker与David George

我们正处于人工智能泡沫中吗？

Are we in an AI bubble?

我不认为我们现在处于人工智能泡沫中。根据不同的视角，我有幸或不幸地在2000年泡沫时期担任科技投资者，那实际上是一场电信泡沫。我认为将当下与2000年进行对比非常有帮助。2000年的互联网泡沫或电信泡沫是由所谓的'暗光纤'定义的。高峰期时，97%的

I do not believe we're in an AI bubble today. I was, depending on how you look at it, the privilege and the misfortune of being a tech investor during the year two thousand bubble, which was really a telecom bubble. And I think it's really helpful to compare and contrast today to the year 2000. The year two thousand internet bubble or telecom bubble was defined by something called dark fiber. At the peak, 97% of

已铺设的光纤处于闲置状态。与今天形成鲜明对比的是，现在没有闲置的GPU。每个重大技术周期都会引发同样的疑问：这是真实的还是泡沫？

the fiber that had been laid was dark. Contrast that with today. There are no dark GPUs. Every major technology cycle raises the same question. Is it real or are we in a bubble?

今天，你将听到Runtime节目中Atreides管理公司董事总经理兼首席投资官Gavin Baker与A16z普通合伙人David George的对话，探讨AI如何重塑全球经济。从资本配置、基础设施支出到商业模式和利润率，这是一次基于数据的深入分析，揭示我们当前在AI周期中的实际位置以及未来可能的发展方向。让我们开始吧。

Today, you'll hear a conversation from Runtime between Gavin Baker, managing director and CIO of Atreides Management, David George, General Partner at A16z, about how AI is reshaping the global economy. From capital allocation and infrastructure spending to business models and margins, it's a detailed, data driven look at where we actually are in the AI cycle and what's likely to happen next. Let's get into it.

这就引出了我们的开场炉边谈话。我们将直接从一个禁忌问题开始。你准备好了吗？如果AI是当今世界最重大的趋势，其证据在哪里？为什么它才刚刚开始在经济中显现？

And that brings us to our opening fireside chat. We're gonna start with a taboo question right out of the gate. Are you ready for it? If AI is the biggest trend in the world right now, where is the evidence for it? Why is it only just beginning to show up in the economy?

正如Andre Karpathy所问：智能体真的只是幻影吗？为了开启这个话题并帮助我们解答这个问题，请和我们一起欢迎Atreides管理合伙人兼首席投资官Gavin Baker。有些人可能知道Gavin是推特上那位见解深刻的博主。每当有重大AI新闻发布时，我知道不少人都会指望Gavin来解释到底发生了什么。非常感谢Gavin今天与我们相聚。

And as Andre Karpathy asked, are agents really just ghosts? To kick this off and to help us answer this question, please join us in welcoming Gavin Baker, managing partner and CIO of Atreides. Now some of you may know Gavin as that really thoughtful guy on Twitter. Anytime some big piece of AI news comes out, I know more than a few people who count on Gavin to explain what the f is really going on. So a huge thank you to Gavin for being with us today.

与他一同参与讨论的是我们A16z的普通合伙人David George。

Joining him is our very own David George, general partner at a sixteen z.

谁知道那音乐是从哪来的？

Who knows what that music was from?

很高兴我们的激励音乐选对了。

Glad to get our pump up music right.

是的。是《太空堡垒卡拉狄加》，1977年的原版，以防我们几年后都得和赛昂人战斗。

Yes. Battlestar Galactica, the original nineteen seventy seven one, in case we have to all fight Cylons in a few years.

是啊。这话题转得不错。谢谢你今天能来，我一直很喜欢和你聊天。

Yeah. Good segue into the topic, I guess. So thank you for being here. I always love talking to you.

我也是。非常感谢你的邀请，也感谢你同事们的款待。我非常期待接下来两天的交流，相信能学到很多，谢谢。

Same. Really grateful to you for inviting me, grateful to your colleagues for having me here. I really look forward to the next two days. I think I'm gonna learn a lot, so thank you.

好的。那么主要议题是AI泡沫，算是宏观视角。先列几个数据铺垫一下，然后我想听听你对当前形势的看法。

Yeah. Okay. Alright. So the big topic is AI bubble, kinda macro view of things. So maybe just to start with a couple stats to set the stage, and then I wanna get your take on where we're at.

美国现有约1万亿美元的数据中心投资。计划未来五年再追加3到4万亿美元。过去三年我们在数据中心上的投入金额已超过整个美国州际公路系统——那可是花了四十年才建成的（经通胀调整后）。仅OpenAI一家就签了超万亿的协议订单，这个可以稍后详谈。但另一方面，这些基础设施的天文数字确实令人担忧，人们会说‘泡沫来了’。

So we have about a trillion dollars of data centers in The US. The plan is to add 3 to $4,000,000,000,000 in the next five years. Over the past three years, we have already built out in data center capacity a larger amount of dollars than the entire US interstate highway system, which took forty years just in terms of dollars, and that's inflation adjusted. OpenAI alone, I think, has more than a trillion dollars of deals set up that they've committed to and we can talk about that. But at the same time, so those are all like big numbers on infrastructure and they're scary and they say, oh bubble.

谷歌最近发布了一项统计数据，显示过去17个月里处理的令牌数量增长了150倍。一方面，这种扩张速度听起来既疯狂又可怕；另一方面，实际使用量确实在激增。那么我们是否正处于AI泡沫中？

And Google released a stat recently that they have seen a 150 x increase in the amount of tokens processed in the last seventeen months. So on the one hand, you've got this crazy, scary sounding build out. On the other hand, you actually have a bunch of usage that's happening. So are we in an AI bubble?

我认为当前并不存在AI泡沫。作为经历过2000年泡沫（实质是电信泡沫）的科技投资人——这既是特权也是不幸——我认为对比当下与2000年很有启发。首先，思科当年市盈率峰值达150-180倍，而英伟达目前约40倍，估值水平截然不同。

I do not believe we're in an AI bubble today. I had, depending on how you look at it, the privilege and the misfortune of being a tech investor during the year 2000 bubble, which was really a telecom bubble. And I think it's really helpful to compare and contrast today to the year 2000. First, I think Cisco peaked at 150 or 180x trailing earnings. NVIDIA's at more like 40x, so valuations are very different.

但最关键的是，2000年的互联网泡沫或电信泡沫有个标志性现象叫'暗光纤'。经历过那个时代的人都懂——暗光纤就是埋在地下却未启用的光纤。没有两端的光学设备和路由器，光纤毫无价值。我至今清晰记得Level three、环球电讯等公司季度报告说'本季度铺设了20万英里暗光纤'的场景。

Most important, however, is that the year two thousand Internet bubble or telecom bubble was defined by something called dark fiber. And if you're a veteran of the year 2000, you will know what that was. But dark fiber was literally fiber that was laid down in the ground and not lit up. Fiber is useless unless you have the optics and switches and routers that you need on either side. And so I vividly remember companies like Level three or Global Crossing or Worldcom would come in and they say, we laid 200,000 miles of dark fiber this quarter.

这太疯狂了。人们总说'互联网将改变世界'、'我们迫不及待要启用这些光纤'。泡沫巅峰时，美国97%已铺设光纤都处于黑暗状态。再看现在的情况——

This is so amazing. The Internet's gonna be so big. We can't wait to light these up. At the peak of the bubble, 97% of the fiber that had been laid in America was dark. Contrast that with today.

现在根本没有'闲置GPU'的概念。随便翻开技术论文就能发现，当前最大难题之一是GPU过热熔毁。有个简单方法能看透本质：观察那些公开上市的GPU大买家的资本回报率(ROIC)。这些公司增加资本支出后，ROIC普遍提升了约10倍。

There are no dark GPUs. All you have to do is read any technical paper, and that one of the biggest problems in a trading run is that GPUs are melting. And there's a very simple way to kinda cut to the heart of all of this. It is return on invested capital of the biggest spenders on GPUs who are all public. And those companies, since they ramped up CapEx, have seen, call it, a 10 increase in their ROICs.

迄今为止，所有AI投入都获得了正向回报。虽然关于Blackwell架构的巨额投资能否持续盈利存在争议，但我个人持乐观态度。无可争议的是，当前AI投资回报非常健康，从估值角度看根本不算泡沫。

So thus far, the ROI on all the spending has been really positive. It's an interesting and open debate about whether or not it will continue to be positive with the quantum of spend we're gonna have on Blackwell. I personally think it will, but there's no debate that thus far the ROI on AI has been really positive. And valuation wise, we're just not in a bubble.

我完全同意。另一个关键区别在于技术采用率——当年互联网推广其实非常困难，需要同时构建网站和用户群这种双边网络，远比现在困难得多。

I couldn't agree more. The other thing that I would say is you can contrast the actual adoption and usage of the technology from then. Right? The internet was actually really hard because you had to build a two sided network. Like you had to build websites and then you had to get users and it's much more difficult.

对于AI工具来说，你只需通过API激活它们，或者在网站上启用ChatGPT，大家就能立即使用，对吧？它们构建在云计算和互联网之上，能实现即时分发，瞬间触达十亿用户。没错。另外就是交易对手方——你提到的这些公司，可以说是人类历史上最优秀的企业，对吧？

In the case of the AI tools, all you have to do is kind of light them up via API or turn on your website ChatGPT and everybody has access to them, right? Built on top of cloud computing, on top of the internet, and you can get to instant distribution, a billion people right away. Absolutely. So the other thing is the counterparties. So you mentioned this, they happen to be the best companies in the history of the world, right?

我认为那些自掏腰包、为这些资本支出开支票的人，他们集体每年能产生约3000亿美元的自由现金流。是这样吗？大致方向没错吧？取个整数。对。

I think collectively the people who are coming out of pocket, the writing checks for this CapEx, I think they collectively generate like $300,000,000,000 of free cash flow a year. Is that right? Some directionally? Round numbers. Yeah.

而且他们资产负债表上还有5000亿美元现金。所以当人们惊呼'天啊这是泡沫要破了吗'时，我觉得其实还好。要知道，激活1千兆瓦的算力大概需要400到500亿美元。

And they have $500,000,000,000 of cash on the balance sheet. So whenever people are like, oh my god, it's a bubble or is it gonna pop? I'm like, I think it's kinda fine. I mean, it costs, like, 40 or $50,000,000,000 to light up one gigawatt.

是啊，如果你用的是英伟达芯片。英伟达芯片？

Yeah. If you're on NVIDIA chips. On NVIDIA chips?

对，没错。所以你看，这就像有个8000亿美元的缓冲池，每年还增长3000亿美元。

Yeah. Yeah. So, you know, there's kind of like an $800,000,000,000 buffer growing $300,000,000,000 every year.

确实。我是说，其中部分企业的自由现金流可能已经开始...你知道的...

Yeah. I mean, free cash flow at some of them has begun to maybe, you know

嗯，这就涉及到你关于投资资本回报率的观点了。

Well, this is this goes to your point on return on invested capital.

我们应该看到下一个环节稍微降下来一点

We should see that next creep down a

对，稍微降一点

little bit. Yeah.

是的，在建设过程中存在一些不匹配。但要知道，拉里·佩奇曾在内部说过，我宁愿破产也不愿输掉这场竞赛。我认为这绝对是谷歌乃至Meta的心态。他们视之为生死存亡之战，必须赢得胜利。

Yeah. A little bit of mismatch in the build out. But, know, Larry Page apparently internally said, I'm happy to go bankrupt rather than lose this race. And I think that is the mentality for sure at Google and perhaps Meta. It's just seen as existential, and you have to win.

好的。关于这种循环交易已经有很多讨论了。因为循环交易是互联网建设中非常可怕的概念，曾造成大问题。你对此怎么看？

Okay. So lots has been written about these round tripping deals. So because round tripping is a very scary concept from the internet build out. That was a big problem. Do you make of it here?

这确实在客观发生。资金是可替代的。如果英伟达与OpenAI签订协议，他们可以说'你不能用我们的钱买我们的芯片'，但资金是可替代的。不过目前规模还很小。是的。

It is objectively happening. Money is fungible. So NVIDIA, if they sign a deal with OpenAI, they can say, hey, you can't use our money to buy our chips, but money is fungible. But it's happening at a very small scale. Yes.

对，我想我明白这个情况

Yeah. And I think I know this is

就像加密货币或区块链那样

like a crypto or blockchain.

确实。很好。是的。我认为推动这一点的并非是需要为GPU或数据中心采购融资，而是竞争态势。所以英伟达最大的竞争对手不是AMD，不是博通，肯定不是Marvell，也不是英特尔，而是谷歌。

Exactly. Good. Yeah. And I think what is driving this isn't the need to finance GPU or data center purchases, but it's actually competitive dynamics. So NVIDIA's biggest competitor, it's not AMD, it's not Broadcom, it's certainly not Marvell, it's not Intel, it's Google.

更具体地说，是谷歌，因为谷歌拥有TPU芯片。这目前可能是训练领域唯一能替代英伟达的方案，也可能是推理领域的最佳替代品。谷歌是个棘手的竞争对手，因为他们还拥有DeepMind公司，以及产品Gemini。可以说他们是当前领先的AI企业。过去两三个月里，他们可能已抢占15到20个百分点的流量份额，这还只是Gemini的访问量。

And more specifically, it is Google because Google owns the TPU chip. And this is by far, maybe perhaps today, the only alternative to NVIDIA for training and maybe the best inference alternative. And Google's a problematic competitor because they also own a company called DeepMind, and they have a product called Gemini. And I think you could argue that they are the leading AI company today. I think they've taken 15 or 20 points of traffic share in the last two or three months, and that's just traffic to Gemini.

这还不包括搜索概览。我怀疑按实际流量计算，谷歌现在比OpenAI、Anthropic等任何公司都大。而这些业务都将运行在TPU上。另外还有三家值得关注的实验室：Anthropic，它受亚马逊和谷歌控制。

It does not include search overviews. I suspect on a actual traffic basis, Google is bigger than OpenAI, Anthropic, anyone today. And that business is going to run on TPUs. And then we have three other labs that are relevant today. There's Anthropic, and that's an Amazon and Google captive.

Anthropic实际上将运行在TPU和Trainium芯片上。因此最前沿就剩下xAI和OpenAI。如果谷歌向Anthropic这样的实验室表示要协助融资并提供芯片，出于竞争考虑，英伟达很难不作出回应。正如黄仁勋所说，他认为这会是个好投资。所以我认为对资金循环的担忧有些过头了。

Anthropic is really going to run on TPUs and Trainiums. And so you're left with xAI and OpenAI at the forefront. And if Google is going to a lab like Anthropic and saying, I'm going to help you fundraise and give you chips, for competitive reasons, it's very hard for NVIDIA not to respond. As Jensen said, he thinks it's gonna be a good investment. So I think the round tripping concerns are pretty overblown.

是的。英伟达真正需要的是Meta能重整旗鼓，或者美国出现另一个开源参与者，又或者中美在AI领域达成某种缓和。

Yeah. I mean, what NVIDIA really needs is they need Meta to get their act together or another American open source player to emerge or maybe some sort of detente with China in AI. Yeah.

当人们问我关于英伟达的所有动作和资金循环时，我的反应是：他们所做的一切都完全合理。

When people ask me about NVIDIA and all the moves and the round tripping, my reaction is everything they've done is completely rational.

百分之百合理。

A 100% rational.

对，长期来看。对，确实。长期来看。

Yeah. Long term. Yeah. Sure. Long term.

他们的举措可能在资本回报率上不如其他选择高，但从战略角度看，我认为这些都是正确的行动。

Things they do may not have as high of a return on capital as other things, but strategically, I think they're all kind of the right moves.

黄仁勋是我所知与埃隆并列的两位最杰出CEO之一，而且我认为他正在把一手好牌打得非常漂亮。

Jensen's one of the two best CEOs along with Elon I have ever known, and I think he's playing a strong hand really well.

好的。那么你开始接触模型公司了。我们直接聊聊模型吧。之后可以再讨论芯片、内存和网络，因为我想听听你的看法。

Yeah. Alright. So you started getting into the model companies. Let's just talk about the model. So we can come back to chips and memory and networking because I wanna get your take on that.

不过，既然我们谈到模型方面，你认为市场结构会如何演变？哪些领域会胜出？你最看好谁？又有哪些担忧？

But, you know, since we're on the model side, what do you think happens with market structure? Who wins where? Who are you most optimistic about? Where do you have concerns?

我认为谦逊是投资者重要的美德。如果打个比方，说ChatGPT之于AI就像网景浏览器之于互联网——在互联网热潮的这个阶段，谷歌尚未成立，扎克伯格还在上中学，卡兰尼克还在幼儿园。所以现在真的非常早期。

So I think humility is an important virtue for an investor. And I'm just if we're going to make an analogy and say that ChatGPT is to AI, as Netscape Navigator was to the Internet, at this point in the Internet boom, Google had not been founded. Mark Zuckerberg was in middle school. Travis Kalanick was in kindergarten. So it's just very early.

因此我认为在应用层做出高置信度预测时保持谦逊很重要。这也是为什么我觉得基础设施层在技术浪潮初期往往是个相对安全的投资领域。

So I think it's important to be humble about making high confidence predictions at the application layer. It's one reason I think the infrastructure layer is often maybe a safe place to be at the beginning of one

关于这些新技术浪潮。实际上要讨论它们在基础设施层扮演的角色，因为它们有一部分明显是作为基础设施层为其他应用提供商提供支持，同时它们自身也有应用。所以我认为，是的，我会画一个

of these new technology waves. Well, actually talk about the role they play at the infrastructure layer because there's a piece of them that obviously they serve as an infrastructure layer powering other application providers, and then they also have their own application. So I think Yeah. I would draw a

我是说，谷歌最能体现这点，但除了观察到互联网曾是非常颠覆性的创新外，很难有十足把握。有合理观点认为AI可能是持续性创新，因为数据这种原材料、购买算力的资金以及分发渠道这些必要条件，当今所有科技巨头都一应俱全。只要执行得当、招揽人才并制定合理策略，你可能会看到它对MAG七巨头中的多数成员成为持续性创新。但另一方面，我也认为这是生死攸关的。如果执行不力，你知道，IBM的结局或许还算不错。

I mean, that's most true of Google, but I think it's hard to have high conviction other than to observe the Internet was a very disruptive innovation. I think there's reasonable arguments that AI could be a sustaining innovation because the raw ingredients of kind of data, the capital to buy compute, and distribution, which is what you need, all of today's biggest tech companies have all of those in spades. So as long as they execute well, hire good people, and have a sound strategy, like, think you could see it be a sustaining innovation for a lot of members of the MAG seven. On the other hand, I do think it's existential. And if you don't execute, you know, IBM might be a might be a good fate.

是啊。确实。确实。这这很难。是的。

Yeah. Yeah. Yeah. That's that's tough. Yeah.

数据分发、算力、资金、人才。

Data distribution, compute, dollars, talent.

没错。而且他们完全有资格胜出。

Yeah. And like They have every right to win.

是的。他们完全有资格胜出。而且现在比以往更明显的是，他们对此非常重视。特别是谷歌，但

Yeah. They have every right to win. And it seems now more than before, they're taking it quite seriously. Yeah. Maybe Google in particular, but

哦，不。不是的。

Oh, no. No.

显然，Meta Meta也在采取他们那些引人注目的行动。

Obviously, Meta Meta is making the dramatic moves they're making too.

不，在我看来，ChatGPT对谷歌而言就像珍珠港事件，我们将看到他们如何应对。他们已经开始慢慢做出反应了。

No. To me, ChatGPT was Pearl Harbor for Google, and we're gonna see how they responded. And they're slowly starting to respond.

是啊。那么你对他们的平台业务部分、基础设施部分有什么预测？你认为在商业模式和市场结构方面会如何演变？你觉得最终会像云计算或飞机制造商那样成为高利润业务，还是像航空公司那样陷入低利润的激烈竞争？

Yeah. And then what do you think what's your forecast for that sort of the platform piece of their business, the infrastructure piece? What do you think how do you think it shakes out in terms of, like, business model, market structure? So do you think they end up as high margin businesses like the clouds or, like, aircraft manufacturers, or do you think they end up very competitive in low margin businesses like airlines?

我认为他们不会沦落到航空公司的地步，但任何人只要看看2021到2022年间SaaS公司的损益表就会发现90%的毛利率。而由于规模效应法则——理查德·萨顿的'越大越好'理论——AI本质上是更依赖计算的，因此它们的毛利率在结构上会更低。但这不意味着它们不能成为伟大的企业。只是我认为要看到某个AI实验室或前沿实验室的毛利率接近SaaS或互联网时代的水平，还需要很长时间。

I don't think they'll be airlines, but you can anybody can just look at the p and l, you know, of a SaaS company circa 2021 and 2022, and you see, you know, 90% gross margins. And the nature of AI because of scaling laws, Richard Sutton's the better the better lesson. They're just more compute intensive, so their gross margins are structurally going to be lower. But that doesn't mean they can't be great businesses. I just I think it's gonna be a long time before we see a truly kind of, you know, an AI lab, a frontier lab with gross margins anywhere near SaaS or Internet era margins.

现在它们的运营支出可以低得多，或许这就是平衡点，但毛利率本质上是不同的。除非规模效应法则改变，或者测试阶段计算的重要性等因素发生变化——我认为这不会发生——否则它们的利润率注定会更低。

Now their OpEx can be a lot lower, and, you know, maybe that's how you square it, but just the gross margins are fundamentally different. And until scaling laws change and the importance of test time compute, things like that to change, which I don't see happening, they're they are gonna be lower margin.

好的。那我们谈谈应用层吧。你刚才稍微提到了SaaS业务。不知道你是否参与了推特上这场论战，每隔几个月就会有人跳出来说'SaaS糟透了'、'SaaS已死'、'这行当要完蛋了'。

Yeah. Okay. So let's talk about application layer. So you just you just kinda got into it a little bit with the SaaS businesses. And I don't know if you've waded into this fight on Twitter, but it's sort of, you know, the the like, you know, every few months it comes up and it's like, SaaS is terrible and it's dead and, you know, it's all gonna go away.

而且安德烈·德克什最近的访谈出来后，市场反应很积极，就像鞭打效应一样。你觉得SaaS和软件行业会怎么发展？

And then, you know, with Andre's Durkesh interview he just did, it's, know, like, the market's reacting positively to it, and it's like a whipsaw reaction. So what do you think happens with SaaS and software?

要知道，我大概在24年初就说过，应用型SaaS与基础设施SaaS可能完全不同。现在我的观点更细致了——我认为应用型SaaS领域会出现大赢家，特别是服务于分散的中小企业客户时。谷歌正在让客户能轻松使用自己的数据构建任何SaaS应用，且数据不会外泄。但许多零售商在应对亚马逊时犯的关键错误是：他们看到亚马逊的利润率就说'我们不想做这种生意'，这显然是个灾难性误判。

You know, I think I, you know, first said probably in early twenty four that I thought all of application SaaS might be a zero different than than infrastructure SaaS. I I would say I have a more nuanced view now, and I think there could be some really big application SaaS winners, especially if you serve, like, a more fragmented SMB customer base. You know, Google is making it really easy if you're a customer of theirs to use your data and essentially make any SaaS app you want, and then your data isn't shared with anyone else. But the critical mistake that I think a lot of retailers made in dealing with Amazon is they looked at Amazon's margins and they said, we don't want to be in that business. And that was obviously a terrible mistake.

25年后的今天，亚马逊零售业务利润率非常健康。我担心应用型SaaS公司正试图维持现有毛利结构，认为毛利下降会导致股价下跌。根据我们刚才的讨论，要在AI领域成功必然面临毛利压力。微软和Adobe已证明软件公司能妥善应对利润率下降，我不明白他们为何担忧。当年企业因云服务毛利更低而不敢转型，现在云服务毛利确实更低。

And here we are twenty five years later, and, you know, Amazon has really healthy retail margins. And I worry that application SaaS companies are trying to preserve their existing gross margin structures because they believe that if their gross margins go down, their stocks will go down. It is definitionally impossible given what we just discussed to succeed in AI without gross margin pressure, and I do not know why they have concerns because we have an existence proof that a software company can deal well with declining margins in Microsoft and Adobe to the whole AI thing came along. You know, it used to be that companies were scared to go from on premise to the cloud because margins were lower. Cloud margins are are are lower.

但依然可观。微软从永久许可维护模式转向云服务，股价十年表现良好。所以对应用型SaaS公司，我想说：别害怕，毛利下降应被视为成功的标志，而非耻辱或恐惧。

They're still good. And Microsoft, they transitioned, you know, from, you know, on premise perpetual licenses with maintenance to a cloud model, and it was a pretty good stock for ten years. So I don't if you're an application SaaS company, like, what I would just say is don't be scared and look at declining gross margins kind of has a mark of success rather than, you know, a badge of shame or something to be feared.

你说这个真有意思，每次我们讨论企业时，来路演的公司都说'我们是AI企业'。我们总看它们的毛利率——现在低毛利率反而成了它们的荣誉勋章。

It's actually so funny you say that because whenever we have these discussions about companies, basically, every company that comes to present to us is like, we're an AI company. And we always look at their gross margins and it's become like a badge of honor for them to actually have low gross margins.

就像是...

Kind of like,

天啊，说明真的有人用你们的AI产品！但如果有人说'我是AI公司'却带着82%的毛利率，你会觉得根本没人用他们的产品。

oh my god. People are actually using your AI stuff. Yeah. But if you show up and you're like, I'm an AI company and it's like, I got 82% gross margins. You're like, I don't think anybody's really using You were not.

确实很有趣。作为上市公司，你更想要1000万美元收入（90%毛利）还是5000万美元收入（60%毛利）？

Yeah. It's it's interesting. Yeah. If you're if you're one of these public companies, would you rather have, like, $10 of revenue with 90% gross margins or $50 of revenue with 60% gross margins?

不难。

Not hard.

就是说，没那么复杂，真的。不算太复杂。但在公开市场上很难操作。

Like, it's not that comp yeah. Not that complicated. It's hard to do in the public market.

市场。在公开市场确实很难，但如果沟通得当，就能避免与云转型相提并论。我是投资人，我会为此感到兴奋的。你懂吧？对。

Market. It's hard to do in publics, but if you communicate it, you drop parallels to the cloud transition. I mean, I'm an investor, and I would be excited about it. You know? Yeah.

而且我觉得世界上不止我这么想。这些传统SaaS应用公司的最大优势在于，他们确实拥有这些高利润的现有业务。所以你可以让新AI产品收支平衡，然后追赶行业领头羊等等。我只是惊讶更多人没这么做——为什么没有一家上市编程公司尝试与Cursor竞争？

And I don't think I'm alone in the world. And then the big advantage these legacy application SaaS companies have is they do have these really profitable existing businesses. And so you can run your new AI products at breakeven and, you know, catch up to the leaders, etcetera, etcetera. And I'm just surprised more people have not done that. Like, why are none of the public coding companies even trying to compete with Cursor?

现实是Cursor现在拥有海量token，等到他们积累足够多的编程token时，追赶将变得困难。但我认为目前，如果某家上市编程公司宣布要全力投入，以收支平衡方式运营，依托现有业务将其整合到所有产品线...

And the reality is Cursor now, they have a trillion trillion tokens, and, you know, there there will be a point where they have enough coding tokens that it's tough to catch them. But I think today, if you're a public coding company and you said, I'm gonna lean in. I'm gonna run at breakeven. I have an existing business. I'm gonna attach it to everything.

嘿，你还有机会。而且奖品显然非常丰厚。我看到马丁已经...

Hey. You have a chance, and, you know, the prize is clearly really big. I see Martin has kept

马丁马丁刚说你还有机会。

Martin Martin just said you have a chance.

我说有机会。他也说有机会。

I said a chance. So he said a chance.

就像《阿呆与阿瓜》里说的，你是在告诉我还有机会，不是

It's like a dumb and dumber, you're telling me there's a chance, not

一个真正的机会。

like a real chance.

你是在告诉我还有机会。

You're telling me there's a chance.

所以是的。完全正确。是的。这就像

So yes. Exactly. Yes. It's like a

没错。完全同意。我们实际上看到...比如Figma上市时，他们的毛利率极高，然后他们说会积极分发AI工具，毛利率会下降。

Yeah. Exactly. I totally agree. Yeah. We actually saw I mean, know, we see it, you know, we may if we if we, you know, Figma, for example, like when they went out, they are extremely high gross margin and they're like, hey, we're gonna, you know, pretty aggressively distribute our AI tools and our gross margins are gonna go down.

投资者问了些问题后意识到这其实是好事。我惊讶公开市场没更多人这么做。

And, you know, investors asked a few clarifying questions and then they were like, oh, that actually would be a good thing. And so I'm surprised more people in the public markets aren't doing it.

我们为他们安排得还不错。

We worked out okay for them.

进展顺利。这是一场持久战。那么在消费者端，应用层面呢？显然，谷歌曾是互联网的门户，某种程度上现在依然是

It's working out well. Long game to play. What about on the consumer side, the application layer? So obviously, Google was the portal to the internet is kind of still is

互联网的门户。而且

the portal to the internet. And the

整个商业模式都建立在捕捉用户意图并将你引导至其他网站的基础上，由它们来为你提供服务。这种方式即将改变。在AI时代已经有所不同。虽然我今天试用了浏览器，想完成一些基本购物操作，发现仍有改进空间，但我相信会逐步完善。那么你认为消费互联网公司的市场结构会如何演变？

whole business model was predicated upon taking some intent and directing you to someone else's website where they would do stuff with you. It's kind of not going to be that way. It already is not that way with AI. Although I tried the browser today and I tried to do some pretty basic shopping stuff and it's, you know, still some work to do, but I think it will get there. So what do you actually think happens with the sort of market structure of the consumer internet companies?

它们会被整合进聊天机器人界面成为组件吗？还是你认为会有其他发展路径？

Do they get subsumed into a component of a chatbot interface, or do you think it's something else?

第一，保持谦逊，难以断言。第二，我认为推出AI浏览器的那些AI公司可能会后悔，因为有个叫Chrome的产品坐拥50亿用户。如果你是谷歌，只需回顾Google Buzz的教训就知道他们非常谨慎。目前他们还深陷与政府的诉讼中。

So one, humility, hard to say. Two, I would just say I think the AI companies that have launched these AI browsers may come to regret it because there's something called Chrome that has, whatever it is, 5,000,000,000 users. And if you're Google, you know, you can just go look at what happened with Google Buzz. You know, they are very cautious. You know, there's you know, they're currently in in litigation with the government.

他们本可以轻松跟进甚至做得更好，但不愿做第一个吃螃蟹的人。现在有两家原生AI公司推出了自己的浏览器，让它们先跑上三到六个月取得些微优势，然后——瞧，我们不得不跟进，但效果如何还很难说。或许对那些不拥有Chrome的公司来说情况会不同。

And they could easily do this and probably do it even better, but they didn't wanna be first. So now you have two AI native companies with their own browsers, let them run for three to six months, get a little head start, and then, wow, here we are. We had to do this, and I don't know how that's gonna work. Maybe for the companies other than Google who don't own Chrome.

我想数据和分发在这方面相当强大。

I guess data and distribution is pretty powerful in that.

是啊，事后诸葛亮。我想说的是，现在要押注那些拥有庞大现有用户基础的公司确实很难。而且我认为推理能力从根本上改变了这些前沿模型的经济模式。要知道，在推理能力出现之前，我常说：如果你是一个前沿模型，却无法获取独特且有价值的数据和互联网规模的分发渠道，那你就是历史上贬值最快的资产。

Yeah. Hindsight's 2020. And the one thing I would say is I do think it's tough to bet against the companies with large existing user bases today. And I also think reasoning has fundamentally changed the economics of these frontier models. You know, pre reasoning, I often said, if you are a frontier model without access to unique valuable data and Internet scale distribution, you're the fastest depreciating asset in history.

推理能力确实改变了这一点，因为强化学习在后期训练中的运作方式——现在拥有大量用户基础就像解锁了每个伟大消费互联网公司的飞轮核心：你有一个好产品，吸引大量用户，用户让算法变得更好，算法又让产品更出色，如此循环。虽然AI领域这个飞轮还没完全转起来，但已经能看到雏形了。

Ike reasoning really changed that because the way RL works during post training, having a big user base now kind of unlocks that flywheel that was at the center of every great consumer Internet company where you have a good product. You get a lot of users. The users make the algorithm better. The algorithm makes the product better, and it just spins. And that it's not quite spinning yet in AI, but you can squint and see it.

所以我认为这从根本上改变了Anthropic、xAI和OpenAI的经济模式。不过扎克伯格也很努力。我们拭目以待吧。

And so I think that fundamentally changes the economics for Anthropic, for x AI, for OpenAI. But, I mean, Mark Zuckerberg is trying hard. Yeah. And we'll see.

是啊是啊，确实

Yeah. Yeah. Yeah. A lot

现在里面有很多聪明人。确实。我觉得令人担忧的是——这也是另一个有趣的点——某种程度上，中国的开源模型生态系统对美国那些试图追赶四大领先实验室的公司来说简直是天赐良机。因为问题是，如果你没有Gemini 2.5 Pro或它的后续版本，或者Grok未公开的新版本，或者GPT Checkmate的新检查点来训练下一代模型，你就会处于巨大劣势。

of smart people in there now. Yeah. For sure. I think the worry is, and I think this is another interesting thing, is if you don't, like, in a strange way, the Chinese open source model ecosystem is a godsend to any American company that's trying to catch those four leading labs. Because the problem is if you don't have Gemini 2.5 Pro or a later checkpoint of it or a later checkpoint of Grok that we don't see or a later GPT Checkmate checkpoint training the next model, you're at a big disadvantage.

哦对了，有件事让我特别抓狂——那些说GPT-5是规模扩展终点的人。GPT-5是个更小的模型，它本就不是为了变得更强而设计，而是为了让OpenAI和微软能更经济地运行。任何把GPT-5和规模扩展终点扯上关系的说法都太疯狂了。

Oh, by the way, one thing I just wanna say that drives me crazy is all these people who say that GPT five is the end of scaling loss. GPT five is a smaller model. It was not designed to be better. It was designed to be more economical for OpenAI and Microsoft to run. It any reference to GPT five and scaling loss is crazy.

嗯。抱歉。吐槽完毕。

Yeah. Sorry. Rant. Rant over.

如果你需要的话，我们可以把底座放在这里。对。没错。握个手。是的。

We get the pedestal up here if you want. Yeah. Exactly. Shaking your hand. Yes.

表现好点。那样就很好。你想聊聊芯片吗？

Be good. That'd be good. Do you wanna talk about chips?

当然。

Sure.

好的，我知道你热爱英伟达。谈谈你对英伟达、AMD、TPU、ASIC的看法，以及你认为市场结构如何影响各玩家所拥有的竞争优势。

So, okay, I know you love NVIDIA. Talk about, you know, your view of NVIDIA, AMD, TPUs, ASICs, and how do you think sort of market structure shakes out their, you know, competitive advantage that the various players have.

是的。我认为这实际上是英伟达与谷歌TPU之间的较量。但有一点常被忽视的是博通和AMD实际上正在联合进军市场。英伟达已不再只是一家半导体公司——明天你听黄仁勋演讲时就会明白。它从半导体公司转型为拥有CUDA的软件公司，再到提供机架级解决方案的系统公司，现在甚至可以说是数据中心级公司，通过横向扩展、跨网络扩展等架构设计实现升级。

Yeah. I think it goes I think it is really it's a fight between NVIDIA and the Google TPU. And then something that I don't think is broadly appreciated is the extent which Broadcom and AMD are effectively going to market together. NVIDIA is no longer just a a semiconductor company as I'm sure you'll hear from Jensen tomorrow. You know, not it was a semiconductor company, then a software company with CUDA, now a systems company with these rack level solutions, and now arguably, you know, a data center level company with the, you know, level of architecting they're doing with scale up, scale across and scale out, scale across networking.

因此网络、架构、软件都至关重要。博通对Meta这类公司说的是：'我们将为你打造一个理论上能与英伟达架构抗衡的网络架构，后者混合了NVLink与InfiniBand或以太网技术。我们将基于以太网构建，这将是开放标准。'

So the networking, the fabric, the software, it's all important. And what Broadcom is saying to companies like Meta is, hey. We will build you a fabric that can theoretically compete with NVIDIA's fabric, which is a mixture of NVLink and either InfiniBand or Ethernet. It will build it on Ethernet. It's gonna be an open standard.

嘿，我们会为你打造属于你的TPU版本——顺便说一句，谷歌花了三代产品才让它正常运作。你知道吗？如果你的ASIC不够好，直接换上AMD就行。但我个人认为大多数ASIC都会失败，

And, hey, we'll we'll make you your version of of TPU, which, by the way, took Google three generations to get working. And you know what? If your ASIC isn't good, you can just plug AMD right in. But I personally believe most of those ASICs are gonna fail,

特别是经过相当长的时间后？或者说假以时日？

particularly if it's in the fullness of like, over a period of time or in the fullness of time?

未来三年内。我认为你会看到一大批高调的ASIC项目被取消，尤其是如果谷歌开始对外销售TPU——这消息在X平台上已经传遍了。到时候，谁知道具体会怎样发展呢？比如Anthropic刚被传出想购买价值数百亿美元的TPU。如果你是Anthropic，可能不想让谷歌看到你的核心算法，但总有办法规避。

In the next three years. I think you'll see a bunch of high profile ASIC programs canceled, especially if Google starts selling TPUs externally, which has been all over x. And then, you know, they you know, who knows exactly how that would work? Because if you're Anthropic, it was just rumored Anthropic wants to buy tens of billions of TPUs. If you're Anthropic, maybe you don't want Google seeing your secret sauce, but there's ways around that.

所以我认为这实际上是谷歌与其TPU（目前由博通支持）之间的博弈，谷歌随时可以收回TPU控制权。是的，他们现在做不了博通那样的以太网技术，但他们掌握着TPU。本质上这是谷歌+TPU对阵英伟达的战争。要知道，亚马逊那边可是有支非常厉害的团队。

So I think this is really a battle between Google and its TPU enabled by Broadcom for now, and Google can take the TPU away from Broadcom whenever they want. Yeah. Now they can't do the Ethernet networking that Broadcom is is doing, but they control the TPU. So it's really Google and the TPU versus NVIDIA. You know, with with, you know, Amazon, like, that's a very talented team.

你们是超大规模厂商中最强的芯片团队吗，安娜普尔纳团队？我觉得Tranium三代可能会比二代强得多。谷歌花了三代才搞定TPU。而AMD永远会是备选方案，但你确实需要备选方案。

Are you the most talented silicon team at any hyperscaler, the Annapurna team? Like, I think the Tranium three will probably be a much better chip than the Tranium two. It took Google three generations to get the TPU right. And then AMD will, you know, will always be kind of the second source, and you need a second source.

好的。真令人兴奋。你认为接下来会怎样...我想回到商业模式的话题。大家广泛讨论的一个重点是颠覆性机遇的来源——在座多数CEO都是初创企业领袖，他们要么想击败行业巨头，要么在寻找新市场机会。

Alright. Exciting. What do you think happens okay. So I wanna go back to business models. So one of the big things that is widely discussed is, like, you know, source of disruption, and most of the CEOs in this room are CEOs of startups who are trying to go beat some incumbent or find, you know, some new market opportunity.

最成熟的机遇往往出现在平台重大转型伴随商业模式变革时。有几个领域在我看来很明显，比如我们投资的Decagon做客服支持——很容易想象按任务解决量收费的商业模式，因为这非常可量化。在编程领域，很多商业模式已转向按使用量计费，尤其是面向开发者的服务，这种模式已被广泛接受。

And the most ripe opportunities tend to come when you have a big platform shift that is also accompanied with a business model shift. And so there are a couple of areas where I can see it. I feel like in an obvious way. So, you know, we're investors in Decagon, customer support, like you can pretty easily see a business model that is priced on the resolution of a task because it's so measurable. You can see, you know, like in coding, like a lot of the business model has now shifted to consumption and, you know, obviously, especially for developer facing things like that's comfortable and pretty well known.

其他行业的情况如何？因为我感觉现在有种敷衍了事的趋势，大家都在说‘我们要拿下所有服务’，但具体怎么做？这其实相当困难。你对这个发展趋势有什么预测吗？

What about the rest of the industry? Because I feel like there's sort of this hand wave thing that is going on, which is like, we're gonna go get all of services. But it's like, okay, so how do you actually go do that? It's gonna be pretty hard. So do you have any prediction on how that plays out?

以客服为例——这是个简单的初始案例——我们拥有大量文本数据。大语言模型擅长处理文本。你可以通过强化学习确保它们获得有效奖励，比如客户满意度、首次通话解决率等指标。我认为这种模式会普及开来。人类本质上是按成果获取报酬的，而AI既会增强人类能力，也会取代部分岗位，最终都将根据成果获得报酬。

Well, I think what you're seeing in customer service, which is kind of like an easy first example, we have a lot of textual data. The LLMs are good at text. You could kind of, you know, probably really easily run some RL to make sure that they, you know, get a good verified reward, you know, verified reward being a happy customer, first call resolution, or whatever it is. And but I do think you will see that played out. Like, humans were fundamentally played for out paid paid based on outcomes, and a lot of AI will be augmenting humans, but probably also replacing some humans, and that will involve being paid paid for outcomes.

回到消费者商业模式，现在人人都在讨论联盟营销费。我肯定会拥有自己的AI助手——基于Grok的定制版本，毕竟我们都曾是该公司的股东。这个版本的Grok会了解并喜欢我，比如下次度假时，它能推荐三家我偏好的酒店。

You know, going back to the consumer business model, you know, everybody's talking about affiliate fees. And for sure, I'm gonna have, you know, my own AI. It will be a version of Grok because we're both ex AI shareholders. It will be a version of Grok that knows me and it likes me. And, you know, when I when I wanna know, you the next time I wanna go on vacation, it will know the hotels that I like to go to and it'll say, hey, three hotels.

加文会来帮我，他能提供最优价格和最佳房型。

I have Gavin, you know, I have Gavin coming who's got the best price and the best room.

他会大幅提升你送给贝姬的礼物档次。

He's gonna massively upgrade the gifts that you give to Becky.

没错。顺便说下——贝姬就在现场，她非常喜欢你之前引用的《阿呆与阿瓜》梗。不过最终可能还是要支付些联盟费用。

Yes. Case she's As Becky Becky's in the audience, she really appreciated your Dumb and Dumber reference. I'll have you know. But yeah. And then there will probably be some sort of affiliate fee.

这本质上还是为结果付费的闭环模式，可能会轻微削弱原有商业模式。为什么谷歌从不做交易平台？因为商家总是高估通过谷歌获客后将其转化为自然客户的能力，导致持续超额支付广告费。这种低效环节终将被挤压出去。

And, again, that's just being paid for an outcome and kind of closing that loop, which will be probably a little bit of a business model degradation because the great why why did Google never start a marketplace? Because people overvalue systematically their ability once they've acquired a customer through Google to keep it as an organic customer. So they systematically overpay, and they continue doing that. That's why Google never went to outcomes or marketplace because advertising leads to the advertisers systematically overpaying. So that inefficiency will be squeezed out.

不过，是的，我们会关注结果。你知道，我觉得今天马斯克发推说工作将变成可选项。明白吗？就像，与其在超市买蔬菜，你完全可以自己种个菜园子。虽然不知道实现要多久，但以这项技术的强大程度，听起来并非天方夜谭。

But, yeah, we'll go to outcomes. And, you know, I think Elon tweeted today that, you know, work would become optional. You know? Like, instead of buying your vegetables, you know, at a at a supermarket, you can grow your own garden if you want. Now who knows how long it takes us to get there, but I that doesn't sound wildly implausible to me for how powerful this technology is.

我刚被震惊了，卡帕西两天前还被描绘成怀疑者，因为他认为通用人工智能还要十年。开玩笑吗？太离谱了。十年？得了吧。

And I was just struck, Karpathy, you know, whatever, two days ago, you know, was being painted as, like, a skeptic for saying AGI is ten years away. Are you kidding? That's insane. Ten years? Yeah.

太夸张了。没错。算我一个。大多数人预期的时间线更短些，求之不得。

That's wild. Yeah. Sign me up. Most of have shorter timelines, please.

是啊。呃，其实不。太棒了。既然我们聊到激动人心的未来科技话题，机器人技术，你...

Yeah. Well, so no. That's awesome. While we're on the topic of, like, very exciting futuristic things, robotics, do

有看法吗？非常现实。而且会是特斯拉对阵中国公司，就像电动车领域特斯拉与中国企业的竞争一样。对。

you have a view on Yeah. Very real. And it's gonna be Tesla versus the Chinese in the same way it's Tesla versus the Chinese in in cars. Electric cars. Yeah.

对。我直接说汽车行业吧，不单指电动车。

Yeah. I would just say cars, not electric cars.

嗯。汽车。没错。你对时间线有概念吗？

Yeah. Cars. Yeah. Do you have a sense of timeline?

我是说，你们都可以看看Optimus的视频。我认识的所有机器人专家都对此印象深刻。要知道，现在有个巨大的争论——未来是人形机器人还是非人形机器人的天下？我认为这场争论已经结束了，因为人形机器人能够通过观看YouTube视频来学习，而且人类穿上动作捕捉服来示范动作也会更容易。

I mean, you can you can all watch the Optimus videos. Every roboticist I know is extremely impressed. You know, there's there's a giant debate. Is it gonna be humanoids or not humanoids? I think that debate is over because humanoids can kind of learn, you know, from watching YouTube videos, and then it's easier for a human being, you know, to put on a suit and show the robot how to do it.

我是说，看着50台Optimus机器人执行50种不同任务的视频简直太疯狂了，但原理其实很简单。你懂吗？比如你有没有正确地把玻璃杯放进洗碗机？

I mean, it's kinda crazy to watch the video of all the, you know, the 50 Optimus robots doing 50 different tasks, and then it's very simple. You know? Did you did you put the glass in the dishwasher correctly or not?

这太有趣了，Gavin。我一直很喜欢和你聊天。让我们为Gavin鼓掌。

This is so fun, Gavin. I I always love chatting with you. Let's give a hand to Gavin.

谢谢你，David。非常感谢。好的。

Thank you, David. Thank you. Alright.

接下来，我们将迎来一场关于建设现实世界基础设施的精彩讨论。但首先请给我们几分钟时间，我们需要快速调整舞台布置。谢谢大家。

Next up, we have a very exciting panel on building out real world infrastructure. But first, give us a few minutes. We got to do a quick, stage change here. So thank you. Thanks, everybody.

感谢收听本期a16z播客。如果您喜欢本期节目，请记得点赞、评论、订阅、给我们评分或写评论，并与亲友分享。更多节目请访问YouTube、Apple Podcasts和Spotify。在X平台关注@a16z，订阅我们的Substack专栏a16z.substack.com。再次感谢收听，下期节目再见。

Thanks for listening to this episode of the a 16 z podcast. If you like this episode, be sure to like, comment, subscribe, leave us a rating or a review, and share it with your friends and family. For more episodes, go to YouTube, Apple Podcasts, and Spotify. Follow us on x at a sixteen z, and subscribe to our Substack at a sixteen z dot Substack dot com. Thanks again for listening, and I'll see you in the next episode.

温馨提示：本节目内容仅用于信息分享，不作为法律、商业、税务或投资建议，也不应用于评估任何投资或证券，且不针对任何a16z基金的现有或潜在投资者。请注意a16z及其关联机构可能持有本播客讨论公司的投资。更多详情及投资披露链接请见a16z.com/disclosures。

As a reminder, the content here is for informational purposes only, should not be taken as legal business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any a sixteen z fund. Please note that a sixteen z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a 16z.com forward slash disclosures.