为什么十亿美元退出已成为过去

Anthropic 和 OpenAI 每月创造的收入已经超过 Meta、谷歌和微软。

Anthropic and OpenAI are adding more revenue per month than Meta, Google, or Microsoft.

如果这两家公司合起来的年收入达到两千亿美元，我一点都不意外。

And I wouldn't be surprised if the combination of those two companies is doing 200,000,000,000 of revenue run rate.

从2020年到2024年，顶级

Between 2020 and 2024, top

1% 的公司起步时的收入是100亿美元。

1% started at $10,000,000,000.

今年二月我们更新过一次数据，当时是200亿美元。

We updated those numbers in February this year, $20,000,000,000.

我们昨天刚更新了最新数字。

We just updated them yesterday.

现在已经达到320亿美元了。

It's now at $32,000,000,000.

所以我们的收入增长了十倍。

So we've 10xed Yeah.

大概在二十四个月的时间里。

Over the space of kind of twenty four months.

当模型变得非常强大，围绕它们构建的产品也日益成熟时，用户量就会出现爆发式增长。

When the models get really good and the products that get built around them get really good, you see this takeoff in usage happening.

我们确实经历过一场AI泡沫。

Oh, we had an AI bubble.

我很有信心地说，我们现在并不处在泡沫中。

I feel pretty confident saying that we're not in a bubble right now.

唯一可能改变这一局面的是

The one thing that could shift that would be

在过去十年里，风险投资已经适应了企业规模不断扩大、长期保持私有化的趋势。

Over the last decade, venture capital adapted to companies becoming larger and staying private longer.

但人工智能可能正在极大地加速这一趋势。

But AI may be accelerating that trend dramatically.

尽管这些前沿实验室在企业级应用方面仍处于早期阶段，它们的收入增长速度已经堪比全球最大的软件公司。

The frontier labs are already adding revenue at a pace comparable to the largest software companies in the world despite being early in real enterprise adoption.

与此同时，支撑这一转变的基础设施——计算能力、电力、数据中心和人才——正变得越来越紧张。

At the same time, the infrastructure supporting this shift compute, power, data centers, and talent is increasingly constrained.

这种局面迫使投资者重新思考关于规模、护城河、价值捕获，甚至风险投资本身运作方式的一些核心假设。

That combination is forcing investors to rethink some of their core assumptions around scale, defensibility, value capture, and even how venture capital itself works.

a16z的戴维·乔治和VenCap首席投资官戴维·克拉克讨论了人工智能、风险投资以及下一代巨型科技公司。

A16z's David George and VenCap CIO David Clark discuss AI, venture capital, and the next generation of massive technology companies.

在我的职业生涯里，我从没这么快地改变过主意，这固然好，但也让人感到谦卑。

I can't think of a time in my career where I have changed my mind about things at a faster clip, which is good, but it's also humbling.

对吧？

Right?

有两个关键方面：规模和价值捕获。

Two big areas are scale and value capture.

在规模方面，去年十一月左右，我们的业务和整个职场的生产力就发生了根本性变化。

So on the scale side, the world kind of changed in November as it relates to our business and I think sort of productivity in the workforce.

之前我们对大部分AI工作的看法，都像是企业里一个模糊的承诺，我们通常把它和云计算、软件公司以及提升效率这些事情联系起来。

The way that we thought about much of the AI work that was happening before that was a sort of like nebulous promise in the enterprise, but we probably were contextualizing it around things like the cloud and software companies and productivity enhancement.

而在消费端，你可以把AI公司当成一家消费类企业来看，看它有多少用户、定价多少、潜力能有多大。

And then on the consumer side, you could think about AI companies like a consumer business, how many users they have and what the price is and how big that can get.

顺便说一句，我觉得这个市场最终会比大多数人想象的还要大得多，这个我们后面可以聊聊。

And by the way, think that's going to be much bigger than people expect too, which we could talk about.

但到了十一月，我觉得我们对企业在AI上究竟会怎么发展，所有之前的预判都彻底变了。

But as of November, I think all of our priors shifted around what is actually going to happen in the enterprise.

为了帮你理解之后发生了什么，简单说一句，Anthropic和OpenAI每月新增的收入，已经超过了Meta、谷歌或微软。

But just maybe to contextualize what's happened since then, basically, Anthropic and OpenAI are adding more revenue per month than Meta, Google, or Microsoft.

它们的月收入增长规模已经达到了这个级别，而这项技术真正渗透到实体经济中的程度，却还微乎其微。

They are already at that scale of revenue getting added, and actual diffusion of this technology into the real economy is tiny.

还不到百分之五。

It's like less than 5%.

当然，在编程和科技前沿的公司里，确实进展快得多。

Now, within coding and in tech forward companies, yes, it's much more advanced.

但说到企业里其他所有部门，真正全面发挥AI的潜力，我们现在还差得远。

But as it relates to every other function in the enterprise, full utilization of the capabilities, we're nowhere right now.

所以，如果把这一点和他们每月新增的收入已经超过三大云服务商的事实结合起来，再考虑到目前这项技术在整体经济中的渗透率还不到5%，我觉得未来的前景会非常惊人。

So, if you pair that up with the fact that they're already getting bigger in terms of revenue added than the hyperscalers, and you're at less than 5% diffusion into the economy, I think the outcomes are going to be extraordinary.

因此，我们开始尝试去思考一个问题：未来可能达到的上限在哪里？那就是企业最终必须为这项技术买单。

So the thing that we've started to try to look at to gauge what can possibly happen, like what's the upper bound, is enterprises are gonna have to pay for this somehow.

是的。

Yeah.

如果你只看财富500强或者标普500企业，它们的规模其实已经很接近了。

And so if you just look at the Fortune 500 or the S and P 500, they're actually pretty close.

这些公司每年合计产生的利润大约是两万亿美元。

They generate like 2,000,000,000,000 of profit per year, the collective.

我不觉得奇怪，到今年年底，这两家公司各自的年收入加起来可能就达到两千亿美元了。

And I wouldn't be surprised if the combination of those two companies is doing 200,000,000,000 of revenue run rate by the end of this year.

对。

Yeah.

更别提还有人用开源方案或者其他供应商的服务了，这些加起来还能再往上添不少。

Not to mention people using open source, other vendors, so, like, you can add even more on top of that.

所以我们现在讨论的已经是占财富五百强企业利润的10%了。

So we're already talking about, like, a 10% profit into the Fortune five hundred.

因此，我认为这笔钱最终会从哪里来，才是关键所在。

And so, I think the upper bound is gonna be where the dollar's gonna come from.

而购买这些东西的一个重要影响是……

And one of the implications, like, to buy this stuff.

你知道，我们以前有很多理论，说开源和本地化会特别重要。

You know, one of the implications of this is we had all these theories why open source and local were gonna be really important.

结果现在成本问题迎面砸过来，反而让这些东西比我们预想的更早变得至关重要。

And it turns out that, like, cost is gonna hit us in the face and make them really important sooner than we thought.

所以，关于规模，我们已经彻底更新了之前的判断，对这个机会的规模和影响力有了全新的认识。

So, scale, we've updated our priors to get really pilled on this outcome thing, on the size of the prize and the scale.

从数据上，你已经能看到一些早期迹象了。

And you can see the early signs of it in the numbers.

但说到底，目前这些技术在实体经济中的渗透率几乎可以忽略不计。

But basically, almost no diffusion into the real economy.

其他这些职能也会变得越来越好。

It's gonna get great for all these other functions.

顺便说一句，编程领域发生的变化，你现在也能在一些其他白领工作中看到苗头了。

By the way, what's happened in coding, you can kinda start to see it in some other white collar jobs.

比如，法律领域已经开始出现这种趋势了。

So, like, it's starting to happen in legal.

当然，法律行业的规模比编程小得多。

Legal space is much smaller, obviously, than coding.

但当模型变得足够强大，围绕它们构建的产品也越来越成熟时，使用量就会突然爆发。

But, you know, when the models get really good and the products that get built around them get really good, you see this takeoff in usage happening.

我觉得在未来十二个月内，这种变化会在组织里的多个职能和行业里陆续发生。

And I think it's gonna happen in a bunch of different functions in organizations and verticals over the next twelve months.

你觉得这其中有多大一部分会是原生的AI应用呢？

And how much of that do you think is gonna be native kind of AI applications?

因为我总是会想起克里斯·迪克森说过的话，头三四年里，我们看到的都是一些模仿旧形式的应用

Because I kind of always go back to Chris Dixon's point around, like, the first three or four years, you kind of see these skeuomorphic applications that kind

开始出现。

of come in.

是的。

Yeah.

我们已经看到了。

And we've seen that.

大多数人现在用人工智能来更高效、更快、更便宜地完成现有工作。

The minute most people are using AI to do their existing job in a way that's more efficient, faster, cheaper.

但我们已经开始看到一些原生应用出现，尤其是在自主型人工智能方面。

But we're kind of starting to see some of the native applications come in with particularly around agentic AI.

对。

Yeah.

你觉得这会如何改变格局？

How do you think that alters the landscape?

我认为

So I think

企业里真正要发生巨变的是，我们到现在还没搞清楚公司该怎么运营。

the big thing that's gonna change in enterprise is we're kinda nowhere on how companies are run Mhmm.

现在的情况跟以前完全不同了。

Differently today.

我觉得那些最前沿的公司，眼下看到的裁员潮，其实更像是在砍掉过去积累的冗余。

And so the most cutting edge companies, I happen to think that what's happening with some of the layoff things that we're seeing is kind of like trimming of previous fat.

我不认为这真的是效率提升。

Like, I I don't think it's actually efficiency gains.

顺便说一句，这些公司内部正发生一件挺有意思的事：资源投入最多的地方，尤其是那些优秀公司，其实都在产品和新事物上，而不是用来自动化内部运营。

And by the way, there's some a really interesting thing that's happening inside these companies where most of the resource devotion, at least for really good companies, is actually on product and new things, as opposed to like automating the way they're run.

毕竟资源就这么多，顶尖公司都明白，产品端一旦做对，回报巨大；而且这些公司里最优秀的人才，比如顶尖工程师，也更想投身于产品开发。

So, they only have so many resources, and the best ones know that the size of the prize of getting something right on the product side, and by the way, the best people at those companies, best engineers, wanna work on that side of things.

对。

Yeah.

产品端的回报确实巨大，而最优秀的人才自然会往那边聚集。

The size of that prize, and the best people are gonna work on that.

所以，大部分工作其实都集中在那边。

And so that's kinda where most of the work is happening.

更成熟的公司反而更适合尝试自动化内部运营，但它们的反应速度通常比较慢。

The more mature companies would be the ones who probably would be better suited trying to automate the way their business is done internally, but they're the slower adopters.

我们在投资的公司里看到一个潜在的机会，就是能进一步提升效率。

There's kind of this latent opportunity that we see in our portfolio companies to get more drive efficiency gains and stuff.

但这些工作并不是顶尖人才在做，而且目前新增的投入也还没往这边倾斜。

But it's not the best people working on it, and it's not where the incremental dollar is going to go just yet.

我接触过那些公司里最前沿的人，他们现在还处在文档整理阶段——就是把所有东西都写成 Markdown 文件，尽可能多地记录背景信息，然后看看怎么在不牺牲客户体验的前提下，提升运营效率。

The most cutting edge folks inside those companies who are trying to do this that I've talked to are kind of in the documentation phase, which is just turn everything into markdown files, have as much context capture as you can possibly get, and then see where you can kind of still manage your business appropriately, not make sacrifices on customer experiences, but drive efficiencies.

我们离真正成熟还早得很。

So, we're very, very, very early in that.

我觉得那些天生就用 AI 驱动的公司，运作方式完全不一样。

I would say that the native AI companies run themselves totally differently.

它们的创始人，骨子里就不一样。

The founders are just built different.

我们观察到上一代创业者的一个特点，比如看SaaS公司，我以前写过，当时根本没意识到他们运营得有多低效，直到后来才看清楚。

One of the things that we've observed about the previous generation of founders, like if you look at SaaS companies, for example, I've written about this, like, we didn't realize how inefficiently they were running until until much later.

他们能多快增长

It's like the How much

才是关键

more quickly they could grow is

他们能多快增长

the Or way how much more quickly they could grow.

顺便说一句，我们发现他们所处市场的规模，跟我们现在看到的模型公司相比，简直小得可怜。

And by the way, it turns out that the magnitude of their market we're already seeing is just is so small compared to what we've seen in the models.

对。

Yeah.

这些AI驱动的公司创造的收入增量，已经超过了整个上市软件行业的总和。

The model companies are adding more than the entire public software universe in terms of revenue added combined.

它们的运营未必精细，但商业模式确实出色。

And so, they're not particularly tightly run, but they had great business models.

所以他们能快速扩张，表现不错，人人都有预算买更多软件，员工人数也不断增长，一切顺理成章。

And so, they could grow, they could do well, and everyone had a mandate to buy more software, and headcount grew, and so everything kind of worked out.

新成立的公司都非常精简、极具进取心，而且全天候工作。

The new companies are very lean, very aggressive, and they work all the time.

所以，看看最前沿的那些公司，真的很有趣。

And so, it's fun to see like the most cutting edge companies.

你进去一看，所有研究员都坐在那儿，轻声跟系统对话，根本不是在跟人工助手交流。

When you go in, all their researchers are sitting there and they're whispering in to the Yeah, so not talking to their agents.

他们连键盘都不碰。

They're not even typing.

效率高得惊人。

Like, they're so efficient.

他们轻声指令，同时运行着一大群智能代理。

They're like whispering in, and they're running swarms of agents.

我觉得，这大概就是未来的模样。

And I think that's kinda gonna be the future.

现在还太早了。

It's just really early.

我觉得现在的交互模式还属于那种仿真实阶段，所有东西都是被动响应的。

I think the skeuomorphic phase, I would say it's like everything that is reactive today.

我觉得未来会转向主动型的模式。

Like, I think there's gonna be a shift of proactive Yeah.

无论是在消费端还是企业端，都会如此。

Engagement, both in consumer and in enterprise.

没错。

Yeah.

我们在一些前沿的初创公司里已经开始看到这种趋势了，但真的还非常非常早期。

And we're starting to see it in some of the cutting edge early stage companies that we're doing, but it's really, really early.

是的。

Yeah.

回想一年前我们的预设，有几件事我觉得已经发生了变化。

When I think of our priors sort of twelve months ago, there's a couple of things that I think have kind of changed.

有一点得到了印证，就是我们一直认为，最大的公司规模会比以往任何一轮高出一个数量级。

One's been reinforced, which was we always thought that the largest companies were going to continue to be an order of magnitude larger than we'd seen in prior cycles.

没错。

Yes.

而且这种趋势还在加快。

And if anything, that's accelerating.

你之前发布过一些数据，提到顶级1%的退出案例每五年左右就翻一番。

So you've put out some data around the size of a top 1% exit doubling every five years or so.

从2020年到2024年，顶级1%的退出估值起点是100亿美元；今年二月我们更新了数据，发现2025年以及2026年前两个月的顶级1%退出估值已经涨到200亿美元；我们昨天刚又更新了一次。

So between 2020 and 2024, 1% exit started at $10,000,000,000 We updated those numbers in February this year, and a top 1% exit for '25 in the first two months of '26 was then $20,000,000,000 We just updated them yesterday.

如果只看已经完成的退出交易，现在的门槛已经达到了320亿美元。

And if you look at just the exits that have closed, it's now at $32,000,000,000.

所以，320亿美元现在就是顶级1%的门槛线。

So Wizz is the threshold for the top 1%.

如果再考虑OpenAI和Anthropic可能的退出，到九月份，这个数字可能突破1000亿美元。

And then if you then think about OpenAI and Anthropic coming in, potentially, we could be north of a $100,000,000,000 by September.

这太惊人了。

That's incredible.

我们已经增长了十倍。

So we've 10xed Yeah.

就在短短二十四个月里。

Over the space of kind of twenty four months Yeah.

这就是顶级1%的退出规模。

What a top 1% exit looks like.

是的。

Yeah.

这些大公司的总和，我觉得比整个罗素2000指数还要大，我不确定是不是这样。

I mean, just the combination of those large companies, I think, is larger than the entire Russell two thousand, if I'm not mistaken.

对。

Yeah.

所以确实如此。

And so Yeah.

这些公司的规模真的变得太大了。

The magnitude of these companies has just grown so great.

而且说实话，我们公司的发展正是回应了这一趋势。

And look, we've built our firm kind of in response to that.

我们相信，随着新趋势的出现，下一代崛起的公司会比它们的前辈更大。

We believe the next subsequent generations of companies that get bigger as new trends happen are going be bigger than their predecessors.

我们其实做过类似的分析，研究了过去六年所有获得风投支持的上市公司。

We actually did a similar analysis where we looked at all of the VC backed IPOs that happened over the last six years.

把这些公司加起来，总市值超过了一万亿美元。

And if you sum all of them up, they're a little over a trillion dollars.

但这个数字，可能还比不上我们预计即将出现的三家大公司中任何一家的估值。

That's probably gonna be smaller than any of the three of the large IPOs Yeah.

这些大公司上市，我们预计很快就会发生。

That we expect to happen.

是的。

Yeah.

所以我觉得，观察到的现象是，结果变得越来越大，但发生的速度也快得多。

So I'd say the observation is the outcomes keep getting bigger, but it's happening much faster.

对。

Yeah.

价值创造的速度

The pace of value creation is

简直惊人。

extraordinary.

像Whiz和Cursor这样的项目，原本可能需要四五年才能从零起步，达到三百亿，甚至可能达到六百亿。

Particularly with something like whiz and cursor, you'd kind of like four, five, six years to get from nothing to, well, 30,000,000,000 and then potentially $60,000,000,000

我也觉得，我们经常讨论部署速度、基金规模这些事。

I would say similarly, there's a lot that we talk about all the time about deployment pace and how big our funds are and things like that.

如果你往前推演一下，会发现过去的趋势规模小了十倍，而最终的结果却大得多。

And if you, you know, extrapolate out and you say, hey, previous trends are kind of 10x smaller and the outcomes get much bigger.

而且值得一提的是，赢家公司之间的集中度非常高。

And by the way, there's a tremendous amount of concentration in the, in the companies that are the winners.

现在我们认为，正是投资的好时机。

You know, now, we believe, is a great time Yeah.

投身市场，现在正是时候。

To be in the market investing.

你知道吗？ChatGPT的爆发还不到四年前。

You know, ChatGPT moment is, I think, less than four years ago.

我们这才刚刚看到，基于这项基础技术所催生的最令人兴奋的成果。

So, we're just now seeing some of the most interesting things happening on the back of the foundational technology.

确实如此。

Yeah.

关于谁能最终抓住这波机会，这个话题我们可以聊很久，因为过去的经验一直在变。

We could have a long talk about who captures it, which is another thing where priors change all the time.

但我们相信，现在正是那些将成为未来十年标志性企业的公司诞生的时刻。

But we believe, you know, now is the moment where the companies are getting created that are going to be the generational companies of the next ten years.

没错。

Yeah.

另一件让我看法有所改变的事，是关于变化的速度，以及领先企业的护城河会如何演变。

So the other thing that where my priors have kind of shifted a little bit as well is just around the speed of change and what happens to the defensibility of the leading companies.

因为从前几代技术来看，真正赚到钱的往往不是第一个进入市场的人。

Because we've seen in prior generations that it's not necessarily the first movers that ultimately capture the economic value of a market.

比如，谷歌并不是第一个搜索引擎，Facebook 也不是第一个社交平台。

So, you know, think Google wasn't the first search engine, Facebook wasn't the first social media site.

我们每年都会关注《福布斯》发布的 AI 50 初创公司榜单。

And and, you know, one of the things we we track is, you know, every year Forbes comes out with their AI 50 startups list.

明白。

Mhmm.

去年和今年相比，榜单上有四成公司直接消失了。

And what was really interesting was, you know, from last year to this year, 40% of the companies that were on that list last year dropped off.

真没想到。

Wow.

这些公司的生命周期，感觉短得惊人。

So that the half life of these companies feels kind of incredibly short.

没错。

Yes.

我觉得我们的预期确实有些变化了：没错，我们觉得最终的成果会大得多，但要预测谁能抓住这些机会，却变得难多了。

You know, I think where our kind of priors have evolved a bit is, yes, we think the outcomes are gonna be much larger, but trying to predict who captures that feels like it's getting much harder.

对。

Yeah.

你们在自己的投资组合里，也看到这种情况了吗？

Is that something that you guys are seeing, like, internally in your portfolios?

是的。

Yeah.

确实越来越难了，因为技术的变革发生得太快了。

It is getting it is getting much harder because the shift in the technology has happened so much faster.

所以我们总是跟创始人聊，说他们脚下的地基正在不断变化。

And so, you know, we we always talk with our founders about, you know, the the shifting sands underneath you.

确实，这一点非常非常真实。

Like, that is very, very true.

我们对价值将在哪里被捕捉这一点的看法，已经彻底更新了。

And our priors have been updated a ton about where value is gonna be captured.

对。

Yeah.

你知道，我们最早投资OpenAI的时候，那时候还没出ChatGPT。

You know, when we first we we invested in, as you know, OpenAI before ChatGPT.

在早期，我们一度觉得，模型公司将会成为一切。

And, you know, there were moments of time in the early days where we said, like, model companies are gonna be everything.

不会再有应用公司了。

There's never gonna be any more application companies.

所有的应用公司都会消失。

They're all gonna go away.

但后来我们又经历了一个阶段，觉得每个领域都会冒出应用公司。

And then we went through a cycle where we said there's gonna be application companies for everything.

模型公司就只是提供API而已。

The model companies are just gonna be APIs.

而现在，我们又回到了这样一个阶段：模型公司正逐步渗透到应用层。

And then now we're back in this moment where the model companies are kind of legging their way up into the application.

而这正是它们提升用户粘性的最主要方式。

And, you know, this is their biggest way to drive stickiness.

所以，要评估一家公司在这个生态中的位置，首先你得处在令牌处理的路径上。

So as it relates to assessing something's place in the world, first of all, like, right now, you have to be in the token path.

是的。

Yeah.

这是我们目前为旗下公司判断价值的首要标准。

Like, that is the number one thing that we're looking to for our companies.

之所以这么重要，是因为我之前提到过的原因。

And the reason that's so important is what I had said earlier.

实际上，科技产品的购买方已经面临越来越大的成本压力。

So, there's actually cost pressure happening at buyers of technology already.

这种变化发生得非常迅速。

Like, very it's happened very fast.

所以，他们不会给那些旧一代软件增加预算。

So, they're not gonna be increasing budget for things that are like previous generation software.

事实上，就连AI带来的成本增长，他们都没法靠削减其他开支来覆盖。

In fact, they can't even cover the growth in their costs that is happening from AI with that, with reductions in that.

因此，这些方面的压力会越来越大。

So, there's gonna be pressure on those.

说实话，解决办法可能只有两种：要么提高价格，要么重组人力结构。

You know, honestly, it's probably gonna have to come from either higher prices that they can charge or restructuring of of the labor force.

目前价值最有可能被捕捉的地方，我觉得是一个完全无法预测的因素——那就是模型公司的市场结构。

The biggest driver of where value is gonna get captured right now is, I would say, something that is totally unknowable, which is what is the market structure of the model companies?

竞争有多激烈？

How much competition is there?

如果前沿只有少数几家公司，那么令牌价格很可能会更高。

If there's a couple at the frontier, token prices will probably be higher.

对。

Yeah.

如果前沿领域有五家公司，令牌价格可能会更低。

If there are five at the frontier, token prices will probably be lower.

令牌价格更低，对整体经济可能更有利，因为这样就不必那么急着调整劳动力结构，尤其是在技术快速扩张的时候。

Token prices being lower probably is better for the overall economy because there's not this pressure to kind of restructure the labor force as quickly as things get really, really big.

对。

Yeah.

现在的情况是，数量还比较少。

You know, right now, the number is smaller.

远不到五家。

It's not five.

目前，前沿人工智能的供给弹性非常低。

There's a tremendous amount of inelasticity for frontier intelligence right now.

还有一个问题是，这种情况未来会如何变化。

There's also a question of how much does that change over time.

很多工作其实用上一代模型就能很好地完成，对吧？

Like, are a lot of the jobs that can be done fine to be done with previous generations of models?

现在没人是这么使用令牌的。

That's not the way anyone is consuming tokens today.

对。

Yeah.

所以，这根本没法预测。

So, you know, that's an unknowable.

市场结构也是个未知数。

The market structure's an unknowable.

开源到底能起什么作用呢？

You know, what role does open source play?

那情况可不太稳定。

You know, that's a tenuous situation.

你能在本地运行多少？

You know, how much can you run locally?

用小模型又能运行多少？

How much can you run with small models?

这些全是悬而未决的问题，我觉得它们将决定谁最终能抓住价值。

Like, these are all the open questions that I think will determine who captures value.

但为了让整个生态系统繁荣起来，恐怕还是得靠竞争，才能把代币价格压得低一些。

But for the broader ecosystem to thrive, it's probably competition that that keeps token prices, you know, lower.

对。

Yeah.

我有几个同事现在正在中国。

So so a couple of my colleagues are in China at the minute.

听他们反馈，跟我们在美国看到的情况对比，真的挺有意思的。

It's been really interesting just getting their feedback, you know, relative to what we're seeing in The US.

他们提到，中国领先的大型语言模型在能力上大概比美国落后六个月，但价格却便宜了十倍。

And one of the things that they were saying was that the leading LLMs in China were probably six months behind where we are in The US in terms of the capability of their models, but they're 10x cheaper.

没错。

Yes.

所以其中一个不确定的因素是，这些公司到底能占多大市场份额？

And so that one of the unknowns, I think, the minute and is is, know, to what extent are they know, what percent of the market will those type of companies capture?

未来十年，我们做的很多事情，到底必须依赖最前沿的模型，还是说，次一级的模型就够了？

How much of of what we end up doing over the next decade will need to be done through the very frontier models Yeah.

而那些稍微低一档的模型又能拿下多少市场？

And what can be captured by that next level down.

这不就是经典的创新者困境吗？

And it's, you know, it's the classic innovator's dilemma, isn't it?

新一代产品虽然只能完成前沿模型80%的功能，但成本却只有十分之一。

That that you get the next generation product that that can do 80% of what the the frontier product can do, but at 10% of the cost.

对。

Yeah.

随着时间推移，这些中端模型的能力会不断追赶，而守住最前沿的门槛就越来越难了。

And over time, those capabilities extend, and it's harder to be at at at that frontier.

是的。

Yeah.

到现在为止，我们对市场对最前沿技术的渴求程度还是感到挺意外的。

As of right now, we've been surprised at how voracious the appetite is for the absolute frontier.

这可能部分是因为我们还没进入优化阶段，对吧。

That's probably partially because we're not in, like, the optimization phase Mhmm.

但优化阶段的到来，可能会比我们预想的快得多。

Yet, but the optimization phase is probably gonna happen sooner than we would have expected.

是的。

Yeah.

我的感觉是这样。

It's my sense.

还有很多其他待解的问题，比如开源的未来——这些玩家到底有多强的本事，能把大模型压缩得那么好？

There's all these other open questions about, you know, the future of open source, like, how how capable are these players of distilling the big models?

毕竟，那些大模型公司并不希望自己的模型被压缩。

Like, the big model companies don't want their models distilled.

对。

Yeah.

没错。

Yeah.

所以，大概只需要花费模型预训练成本的2%左右，就能完成蒸馏。

And so, you know, it probably costs in the order of, like, 2% of the actual training cost pre training cost of a model to distill it.

如果这种情况能持续下去，那对开源来说显然是个好兆头。

And so, you know, if that continues to hold and be possible, you know, that probably bodes well for open source.

但如果做不到，那对开源就不那么乐观了。

If not, it probably doesn't bode well for open source.

所以目前来看，你说得一点都没错。

And so, as of right now, yeah, you're exactly right.

每单位token的成本，每年都在下降超过十倍。

The sort of per token cost like for like is going down more than 10 X year over year.

但前沿领域对token的需求，按美元计算却增长得快得多。

But the appetite for tokens on the frontier is massively exceeding that in terms of dollars.

是的。

Yeah.

对。

Yeah.

那你在评估这些公司价值的时候，怎么考虑这一点呢？

How do you factor that in when you're then thinking about valuations of of these companies?

我觉得其中一个担忧，有点像2021年的情况。

Because I think one of one of the concerns that that that I would have is a bit like in in 2021.

那时候我觉得市场已经到了新兴基金管理的顶峰，因为很多管理人刚做完种子轮，老牌机构就跟进投资，种子轮刚结束几个月就大笔砸钱，几乎零亏损。

I thought 2021, the market there was kind of peak emerging manager because a lot of these managers had done the seed round, you know, established firms were coming in and writing things up, you know, months after the seed round had done, and there was basically a zero loss ratio.

但我们都知道，风险投资本来不是这么运作的。

And, you know, we know that's not how venture works.

现在感觉我们又有点陷入类似的局面，只不过这次是老牌机构主导，因为正是这些成熟机构拿下了AI领域早期的突破性项目。

It feels like we're in a little bit of that situation today, but with the more established firms, because it's the established firms that have been by and large capturing the the early breakouts in the AI space.

对。

Mhmm.

但当我回头看我们早期基金的表现时，亏损率高达60%。

But when I look at it, like, when we look at our early stage funds, there's a 60% loss ratio.

也就是说，六成的投资项目都没能收回本金。

So 60% of deals don't return the capital that was invested in them.

如果我看过去几年人工智能领域的亏损率，虽然不是零，但大概也就个位数百分比。

If I was looking at the loss ratio for the last couple of years in the AI space, I mean, it's not it's not zero, but it's probably single figures, percentages.

但这个数字会上升的。

And it'll go up.

这样下去是不可持续的。

And, like, that's not sustainable.

对。

Yeah.

那我们该怎么看待眼下这个周期所处的位置呢？

So how how how do you kind of think around around where we are kind of in the in that cycle today?

因为迟早，现实规律会重新发挥作用。

Because at some stage, like, the laws of gravity will will will reassert themselves.

没错。

Yeah.

也许有必要说说我们在早期阶段的投资理念，因为我们其实也不追求低亏损率。

Maybe it's helpful to sort of explain our philosophy at the early stage because we we also don't wanna target a low loss ratio.

不，不是这样的。

Like, that's not No.

如果我们承担的风险太低，那反而不对。所以我们经常开玩笑。

We're we're not taking the appropriate amount of risk if we have a low loss And so, you know, we we joke all the time.

我们这个圈子里有个很有名的风投，他特别引以为豪的一点是，他从来没在任何一笔投资上亏过钱。

There's a, you know, prominent VC around in our ecosystem, and, you know, one of his big points of pride is that he's never lost money on a deal.

我们就说，

And we're like,

这根本不是

that's not

这根本不是什么值得骄傲的事。

that's not a point of pride.

这其实是个糟糕的信号。

Like, that's a horrible data point.

这根本不是我们该追求的目标。

Like, that's not what you want.

是的

Yeah.

对

Yeah.

那是一家私募股权公司。

That's a PE firm.

是的

Yeah.

没错

Exactly.

所以，如果你这么做事，确实可以说你承担的风险太少了。

And so, like, certainly, you can make the case that you're not taking enough risk if that's the way you approach it.

我们过去的做法是这样的，这有点像克里斯·迪克森的理念：只要某个重要领域里有不少非常优秀的创业者在做，我们觉得趋势向好，对技术也有明确判断，那就该去找最好的创始人。

The way we've approached it historically, and this is sort of a, you know, Chris Dixon philosophy, is, you know, any major space where there are multiple very talented entrepreneurs building, where we think there's tailwinds, where we have a point of view on the technology that it's good, we should pick the best founders.

我们应该在早期阶段就支持那些市场领导者。

And we should we should try and back the the leaders at the early stage, the market leaders.

如果这个领域真的成功了，而且我们投的是领头羊，那就再好不过了。

And, you know, if the space happens to work out and we've got the leader, excellent.

如果这个领域没起来，但我们投的是领头羊，也没什么损失。

If the space happens to not work out and we have the leader, no harm, no foul.

其实，这本来就是我们业务的一部分。

Actually, that's part of our business.

是的。

Yeah.

这本来就是我们应该做的。

That's what we should be doing.

对。

Yep.

对。

Yep.

我刚才说的那种糟糕情况是：领域确实火了，但我们选错了公司。

The the bad box of what I described is the space works out, and we pick the wrong one.

是的

Yeah.

这些正是我们特别关注的地方，我们一定会尽力确保做对。

And those are the things that we really scrutinize, and we and we try and make sure that we get right.

是的

Yeah.

所以，说实话，我也说不准。

So, you know, I don't know.

有很多例子表明，有些空间最终并没有成功。

There's there's many examples of spaces that didn't quite work out.

但我们始终支持那些优秀的创业者，他们很有才华，彼此竞争，而且这个领域里玩家很多。

But we did back to leading entrepreneur, they're talented entrepreneurs, and they were competing, and there were lots of players in the space.

这对我们来说完全没问题。

That's totally fine with us.

所以，这就是我们的理念，它解释了为什么我们的失败率会是这样，也说明了我们如何权衡风险。

And so, that's the philosophy that underpins how we could have a loss rate that, you know, and and sort of how we think about balancing taking an appropriate amount of risk.

当然，到了成长阶段，情况会有点不一样。

Obviously, that's a little bit different at the growth stage.

对。

Yeah.

所以我们不应该有那么高的失败率。

And so, you know, we we shouldn't have as high of a loss rate.

目前一切还太早期了，我们还不清楚。

As of right now, everything is so early that we don't know.

是的。

Yeah.

就像你所说的，关于谁最终能抓住价值，还有很多未知数。

There's all these unknowns about who captures value, as you said.

我相信随着时间推移，失败率肯定会上升。

I'm sure loss rates are gonna go up over time.

我们能做的就是专心打造公司，结果自然会随着时间慢慢显现。

All we can think about is how we build the firm, and, you know, the results will play out over time.

我们再次认为，这里存在着巨大的幂律效应，之前也提到过。

Again, we think there's just massive power laws we talked about.

赢家自然会脱颖而出，至于那些没成功的项目，我们也会尽力做好支持。

And the winners are gonna take care of themselves, you know, we'll do our best for the things that don't work out.

我们打造这家公司的模式，我觉得正是在回应创业者真正需要的东西。

The way that we're building our firm, I think, is catering to what the entrepreneurs want.

你之前问过新兴管理人和我们这样的大型平台之间的区别。

So, you know, you asked about, you know, emerging managers versus, you know, large platforms like ours.

我们之所以把平台建得这么大、这么有规模，就是因为创业者想要这样的支持。

The reason we built our large platform the way we have with a lot of scale is because that's what the entrepreneurs want.

没错，他们确实希望如此。

And, you know, they Yeah.

这种需求体现在很高的项目成功率，以及对关键事务的高持股比例上。

You know, they that expresses itself in, you know, high win rates of deals and large ownership of of things that matter.

这场AI浪潮发展得太快，带来的一个后果就是，初创公司刚起步就面临大公司才会遇到的问题。

So, you know, one of the consequences of how fast this has happened, this AI wave, is the companies run into big company problems very early in their lives.

所以我们得调整公司的发展方式。

And so, we need to adapt the way we build our firm.

这也是为什么我们扩大了招聘规模的原因之一。

And so, you know, that's part of the reason that we've scaled up, you know, some hiring.

我们正在打造一个更全面的平台，涵盖国际市场、渠道支持这些领域——比如我们已经配备了定价专家，还有如何规模化运营Salesforce这类业务的专业团队。

We're building out a much broader platform that that includes things like international, like channel, you know, where we've we've already got, like, experts in pricing and and, you know, how you scale a Salesforce and and all those things.

除了我们一直为初创公司提供的服务，现在还因为公司上市周期变长了，他们在早期就迫切需要这些支持。

In addition to all the things that we've always done for companies, the reason is the companies are staying private longer, and the companies just need it really early in their lives.

你明白吧？

You know?

比如Cursor，已经实现了数十亿美元的收入。

Like, Cursor, as an example, is, you know, billions of dollars of revenue.

但他们团队还很小，公司也才刚刚起步。

And they're very small, and it's very early in their life.

以前的科技公司没这么快发展，所以他们还没到早期就遇到大规模商业谈判、复杂的供应商关系、云服务合作，或者国际扩张这些难题。

And the previous generation of technology didn't happen so fast, so they didn't encounter things like major business deals they had to negotiate, you know, supplier relationships that were complex, cloud deals, you know, international expansion.

所有这些事情都来得早太多了。

It's all just happening so much sooner.

所以我认为，我们所看到的市场份额增长，某种程度上是创业者在表达他们的选择。

And so, I think part of the market share gains, if you will, that that we've seen is just this is entrepreneurs expressing their preferences.

是的。

Yeah.

对。

Yeah.

这挺有意思的。

So it's funny.

我一位同事昨天参加了一个由英国风险投资协会举办的会议，会上他们对现场观众做了一项调查，问大家觉得现在的AI公司估值怎么样。

One of my colleagues was at a conference yesterday that was run by the the UK Venture Capital Association, and they they surveyed the audience saying, you know, what do you think about AI valuations today?

你觉得是偏高、正好，还是偏低？

You know, too high, about right, you know, too low.

80%的人觉得偏高，只有大约6%的人觉得偏低。

80% said too high, about 6% said too low.

当我思考整个AI领域时，感觉这个比例可能刚好合适，因为我觉得现在80%的公司都被高估了，毕竟从历史经验来看，大多数公司最终都会失败。

And as I kind of think of the AI universe, it feels like that's probably about the right balance because I think 80% of companies probably are overvalued today because we know most of the companies aren't gonna work historically.

对。

Yeah.

但其中肯定有一小部分公司被严重低估了，因为它们会成长为行业领军者，未来估值可能会是现在的好几倍。

And there's probably gonna be a small subset of those companies that are massively undervalued because they're the ones that are gonna emerge as the leaders, and we'll see multiples of of, you know, where they are, where they're being valued today.

对。

Yeah.

从有限合伙人的角度来看，我真的很难想象你现在的处境——毕竟要从这么多公司里挑出真正的赢家，我知道你可以组建一个投资组合，但我觉得坐在LP的位置上有个优势，就是能广泛布局，分散投资于AI领域里那些可能的黑马。

And I think from an LP perspective, I I really would struggle to be in your shoes today because, like, having to pick those individual companies, I know you can kind of put a portfolio together, but one of the advantages I think from being in the LP seat is that we can have a really broad and diversified portfolio of the potential outliers in that AI space.

而且我们很清楚，从历史经验看，即使大部分公司最终会失败，这个投资组合的整体价值还是会随着时间增长。

And we know historically that that basket will increase in value over time even as the majority of those companies might fall away.

对。

Yeah.

是的。

Yeah.