2026年第一季度风险投资达3000亿美元纪录

欢迎收听本播客。

Welcome to the podcast.

我是主持人贾登·谢弗。

I'm your host, Jaden Shafer.

今天节目中，我们想聊聊2026年第一季度的风险投资数据。

Today on the show, we wanna cover the q one twenty twenty six venture funding numbers.

这些数据刚刚发布。

They've just dropped.

数据相当惊人。

They're pretty eye watering.

仅在这一季度，就有大约3000亿美元投资于初创企业，其中80%直接流入了人工智能初创公司。

There's about $300,000,000,000 that has been invested into startups in this single quarter, and 80% of that is going straight into AI startups.

所以我想详细分析一下资金流向了哪些人，以及为什么资金会流向这些领域。

So I wanna break down all the numbers of who's getting money, why it's going where.

我们还得谈谈Meta的失控AI代理事件。

We also have to cover Meta's rogue AI agent incident.

这在公司内部触发了一次严重的安全警报。

It triggered a really serious security alert internally.

我想聊聊一个关于人工智能在医疗领域应用的精彩故事，这是我最喜爱的话题之一。

And I wanna talk about a very cool story, when it comes to AI in health care, one of my favorite topics.

有一个叫做NOAA实验室的机构。

There's something called NOAA Labs.

它刚刚获得了FDA突破性设备认定，这是一种针对能够通过五秒语音记录检测心力衰竭的人工智能系统的认证。

It just got FDA breakthrough, which is basically designation for an AI that can detect heart failures from a five second voice recording.

非常厉害。

Very cool.

接着我们还要关注佐治亚州，今天那里有三项与人工智能相关的法案提交到了州长案头，其中包括一项聊天机器人儿童安全法案，这很有趣。

And then we also are going to Georgia where they are sending three AI related bills to the governor's desk today, including a chatbot child safety bill, which is interesting.

我们现在开始看到一波由各州立法机构推动的人工智能聊天机器人监管浪潮。

And we're starting kind of to see this broader wave of AI chatbot regulations that are going through state legislatures right now.

目前共有78项法案在27个州推进，内容非常丰富。

There's 78 bills across 27 states, so there's a lot to get into.

在我们开始之前，简单提一下。

Before we jump in, quick mention.

如果你还没试过AI Box，我一直在经常使用它。

If you haven't checked out AI Box yet, it's something I've been using pretty regularly.

它让你在一个地方访问70多个AI模型。

It gives you access to over 70 AI models in one place.

所以你不用分别支付ChatGPT、Claw、Gemini和其他所有服务的订阅费，每月只需9美元或8.99美元就能全部获得。

So instead of paying for separate subscriptions to ChatGPT, Claw, Gemini, and everything else, you get it all for 9 or $8.99 a month.

我真正喜欢的是它内置了自动化构建器。

The thing that I actually like about it is that there is an automation builder built in.

你可以用通俗的语言描述你想要什么。

You can describe what you want in plain language.

我不是开发者。

So I'm not a developer.

你不需要懂任何编程或技术设置，它会为你自动构建工作流程。

You don't need to know any coding or any sort of technical setup, and it builds the workflow for you.

所以我一直用它来处理我自己的内容创作，它真的非常有用。

So I've been using it for a bunch of my own content production stuff, and it has been incredibly useful.

如果你想试试，描述里有个链接。

If you wanna try it out, there's a link in the description.

这是我自己创办的初创公司，AI Box dot a I。

It's my own startup, AI Box dot a I.

好了。

Alright.

我们开始本期内容吧。

Let's get into the episode.

我觉得过去几个月里，悄然兴起的是一股针对AI聊天机器人的州级监管浪潮。

So I think something that's been building quietly over the last few months is that there's a massive wave of AI chatbot regulation at the state level.

根据隐私未来论坛的2026年追踪数据，目前已有78项聊天机器人安全法案在27个州提出。

According to the Future of Privacy Forum twenty twenty six tracker, there's now 78 different chatbot safety bills that are introduced across 27 different states.

我认为其中很多都遵循着非常相似的模式。

And I think a lot of those follow a very similar playbook.

它们主要关注披露要求。

They're kind of focused on disclosure requirements.

还有很多内容涉及儿童安全保护。

There's a bunch of stuff that, you know, plays towards child safety protection.

当聊天机器人与未成年人互动时，对其行为有所限制。

There's limits on what chatbots can do when they're interacting with minors.

我认为这些法案的共同点是，如果你在和AI对话，你应该知道你是在和AI对话。

I think the common thread across basically all of these bills is that if you're talking to an AI, you should know you're talking to an AI.

我认为很多人感到担忧，尤其是在进行客户服务或销售时，他们会说：看。

I think a lot of people are concerned, you know, particularly when you're doing something like customer support or sales, and they're like, look.

如果我在和一个客户服务机器人对话，至少告诉我它不是真人。

If I'm talking to a customer support bot, like, let me know it's not a, you know, not a real human.

当然，他们也希望为使用AI聊天机器人的儿童提供充分的安全保障。

And, of course, they want a lot of safety for children using AI chatbots.

因此，俄勒冈州、夏威夷州、科罗拉多州、亚利桑那州和内布拉斯加州目前都有类似法案正在立法进程中。

So states like Oregon, Hawaii, Colorado, Arizona, and Nebraska all have versions of this moving through their legislation right now.

我觉得有趣的是，这并不仅仅来自某一个政治方向。

I think what's interesting is that it isn't just coming from kind of one political direction.

你看到保守派和进步派都在参与。

You're seeing conservatives and progressives.

这两方都在推动这些法案。

Both of these are kinda pushing these bills.

我认为，关于AI和儿童的担忧，是少数几种真正存在两党共识的议题之一。

I think the concern about AI and kids is one of the really rare cases where there's kind of genuine bipartisan energy.

这些78项法案是否都会成为法律，我觉得这是另一个问题。

Whether all 78 of these are actually gonna become law, I think that's definitely a different question.

但我认为趋势已经非常明确了。

But I think the direction is pretty clear.

各州并没有等待国会采取行动。

States aren't waiting for congress to act.

我认为，对于那些AI公司来说，这种各州层面的监管拼凑局面将成为一个巨大的合规噩梦。

And I think for the AI kind of companies out there, this patchwork of state level regulation is gonna become a really big compliance nightmare.

你可能会在每个州遇到不同的披露规则，我认为这正是最终迫使联邦采取行动的那种情况。

You could end up with different kind of disclosure rules in every state, which I think is exactly the kind of thing that eventually forces federal action.

所以所有这些州都在提出这些规定。

So all of these states are putting it out there.

我个人不认为所有这些州将来都能监管好这一切。

I personally don't believe that all of these states are gonna be able to regulate all of this in the future.

我认为联邦政府和国会最终会出台一些法律，比如：

I think the government the federal government is gonna come and Congress is gonna come and just pass laws that are like, okay.

这些就是关于人工智能披露的所有法律，你知道的。

These are all of the laws as far as, like, you know, disclosure of AI.

这些就是关于儿童披露和安全防护的所有法律。

These are all of the laws as far as disclosure and and kind of guardrails for children.

然后所有人都必须采纳这些统一规定，而不是去应对各州杂乱无章的法规，这无疑会是一场噩梦。

And then everyone's going to have to just adopt that instead of having to work through a patchwork of all the states, which could be a nightmare for sure.

好的。

Okay.

佐治亚州目前正在向州长提交三项不同的AI法案。

Georgia right now is sending three different AI bills to the governor.

说到这个，有三项与AI相关的法案。

Speaking of that, there's three AI related bills.

它们都已送至州长布莱恩·肯普的办公桌上。

They all have made it to governor Brian Kemp's desk.

第一项是SB540号法案。

The first one is SB five forty.

这是一项关于聊天机器人披露和儿童安全的法案。

It's a chatbot disclosure, child safety bill.

这是一个很好的例子。

So this is kind of a good example.

第二项法案禁止仅凭AI做出医疗保险覆盖决策。

The second one is prohibiting health care insurance coverage decisions from being based solely on AI.

因此，保险公司可以将AI作为流程的一部分，但最终决定必须有人工参与。

So an insurer can use AI as part of the process, but a human has to be involved in the final call.

第三项是SR789号决议，它主要是成立一个研究委员会，以评估人工智能对本州的广泛影响。

And the third is SR seven eighty nine, and it basically is creating a study committee to look into AI's broader impact on the state.

所以，这个更像是一个兜底条款。

So, I mean, that one's just kind of, a catchall.

但我认为保险法案特别值得关注，这就是我想提这件事的原因——过去一年里，我们看到了大量关于健康保险公司利用人工智能拒赔的报道，而且他们是在大规模地这样做。

But I think the insurance bill is one that's very interesting to watch here, that's why I wanted to bring this up because we've seen a lot of stories over the last year about health insurers using AI to deny claims, and they're doing it at scale.

佐治亚州基本上是在说，这类决策必须有人参与，不能完全由人工智能决定。

Georgia is essentially saying that, you know, not without a human in the loop can you make those types of decisions.

如果这三项法案全部获得签署，佐治亚州将成为在人工智能监管方面最积极的州之一。

If all of these get signed, Georgia's gonna become one of the more active states on AI regulation.

我认为其他州很可能会把保险法案当作模板，因为它契合了我们一直关注的更广泛主题：人工智能正被部署到一些高风险决策领域，而对完全自动化结果的反对声音正变得越来越响亮。

I think I'd expect other states to look at the insurance bill as a template because it kind of ties into this broader theme we've been tracking, is like AI is getting deployed in some of these really high stakes decision areas, And I think the pushback on fully automated outcomes is getting a lot louder.

但归根结底，这些保险公司完全可以声称：我们对许多决策确实有人工介入，或者也许只是点了一个勾，写着‘我接受人工智能的决定’，所以这并不是百分之百自动化。

Now, at the end of the day, these insurance companies can just say, hey, look, know, we we have a human in the loop for a lot of these decisions, or maybe they click a little checkbox that says, yes, I accept the AI's decision, and so it's not gonna be just like a 100% automated.

所以说实话，我不确定我们究竟能获得多少透明度，或者这实际上能带来多大改变——尽管这项法案的初衷非常崇高，但我认为要绕开它其实很容易，只要说‘我们用人工智能工具来辅助我们’就行了。

So I don't really know how much disclosure we're we'd ever get into that or how much of a change it's actually gonna make, if I'm being honest, know, despite how noble a cause that bill might be, think it'd be pretty easy to get around and just say, hey, look, like, use AI tools to help us.

我的意思是，算法和软件可以帮助我们处理一切，你根本无法阻止这种趋势。

I mean, algorithms and software to help us with everything, and you can't really kill that.

所以，是的，当然。

So, yeah, sure.

人类会点击那个标有“根据此结果拒绝索赔”的复选框，但实际上AI在很大程度上完成了思考过程，因为软件在这些大多数事情中都承担了主要的计算任务。

A human will click the checkbox that says denied claim based off of this, but the AI is really doing a lot of the thinking because software does a lot of the thinking in most of these things.

所以，我不确定这是否会带来太大的影响。

So I don't really know if it's gonna make that big of an impact.

接下来我想谈谈国家海洋和大气管理局实验室。

The next thing I wanna talk about is NOAA Labs.

他们刚刚获得了FDA的批准。

They just got FDA approval.

这是基于语音的心力衰竭检测的重大突破。

It's a big breakthrough for voice based heart failure detection.

这对我来说太不可思议了。

This is crazy for me.

本质上，他们有一个名为Vox的产品，可以分析五秒钟的语音录音，并检测出心力衰竭恶化的迹象。

Essentially, they have a product called Vox, and it analyzes a five second recording of your voice, and it can detect signs of worsening heart failure.

它的原理是，人工智能从你的语音中提取与肺部充血和体液过载相关的声学特征，然后基本上会检测你胸部和喉咙中的这些物理变化，这些变化会影响你的声音。

The way it works is that AI extracts acoustic features from your voice that are linked to pulmonary congestion and fluid overload, and then, it's basically going to have, you know, be detecting these physical changes in your chest and your throat that affect how you sound.

在你注意到症状之前，它很多时候就已经能察觉到这些变化了。

It'll be able to notice this before you even notice symptoms a lot of the times.

他们的算法是在超过300万条语音样本上训练的，并已在五个多中心临床试验中得到验证，合作机构包括梅奥诊所和加州大学旧金山分校。

So their algorithm was trained on over 3,000,000 sample voices, and it's been validated across five multicenter clinical trials with partners including the Mayo Clinic and UCSF.

我认为这确实是一个重大突破，但这一认定意味着FDA目前认为这项技术有足够的潜力，从而可以加速对该公司的审查流程。

I think this is definitely a big breakthrough, but this designation means that the FDA FDA right now can see enough promise in this to basically expedite the review process for this company.

他们预计将在2026年年中获得欧盟批准，并在之后不久启动美国的临床试验。

They're expecting EU approval by mid twenty twenty six and a US clinical trial to kick off soon after.

我认为这是展示人工智能在医疗领域令人兴奋潜力的典型案例之一。

I think this is one of those, like, stories that shows you AI in health care is super exciting.

当它被正确实施时，实际上可以表现得极其出色。

It can actually look it can be like, it can work really incredibly when it's done correctly.

它并不是要取代医生。

It's not replacing doctors.

它主要是为患者提供一种在家自我监测的方式，而无需花费大量时间和金钱去医院。

It's basically giving patients a way to monitor themselves at home without something, without having to go in and kind of take all of this time and money.

我认为目前有大约六百万美国人正在经历心力衰竭，因此我认为这种早期预警系统对很多人来说可能意义重大。

I think about six million Americans are living with heart failure right now, and so I think this kind of early warning system could be really significant for a lot of people.

而且我认为这不仅仅局限于通过语音检测心力衰竭。

And I also don't think it's only gonna be in, you know, voice detection heart failure.

你可以将这项技术推广到许多行业和不同的医疗问题上，所以我个人对此非常期待。

Like, you could roll this out to many industries, many different medical problems, so personally very excited about this.

好的。

Alright.

我们得聊聊Meta的失控AI代理。

We gotta talk about Meta's rogue AI agent.

最近Meta已经多次出现AI代理方面的混乱情况。

Meta's had a bunch of these AI agent kind of snafu moments lately.

这件事有点疯狂。

This one was kind of crazy.

基本上，一名员工在内部论坛上发布了一个技术问题。

Basically, an an employee posted a technical question on an internal forum.

在Meta内部，他们有这种内部论坛，可以向其他开发者提问，然后另一位工程师让一个AI代理帮忙分析这个问题。

So inside of Meta, they have these kind of internal forums that they can ask questions for other developers, and then a different engineer asked an AI agent to help analyze the question.

该代理随后生成了回复并直接发布，而且没有征得工程师的同意，这显然有问题。

The agent then generated a response, to this, and it posted it, and it did this without asking the engineer for permission, which obviously is an issue.

而原始的员工随后根据该代理的建议采取了行动。

And the original employee then took actions based off of the agent's guidance.

对吧？

Right?

所以我们现在是在跟踪这件事，对吧？

So so are we tracking this right?

开发者说，嘿。

Developer says, hey.

那我该怎么处理这件事？

What do I do about this?

另一个开发者的代理直接告诉开发者该怎么做，然后开发者真的去执行了。

Another developer's agent goes and tells the developer what to do about it, then they go and actually do the thing.

所以这个过程里虽然有人参与，但真正破坏指挥链的，是这个代理。

So there's, like, human in the loop for part of this, but it's the agent that's, like, poisoning the poisoning the the chain of command here.

但关键是，这件事之所以严重，是因为当这名员工按照AI代理的指示行动时，无意中将大量公司和用户数据泄露给了无权访问这些数据的工程师。

But, basically, the reason why this is a big deal because normally that's, you know, would be whatever, is because when the employee took action based off of what the AI agent told it to do, they inadvertently exposed massive amounts of company and user related data to engineers who were not authorized to access it.

这些数据在被发现前暴露了大约两个小时。

So the data was exposed for about two hours before it was caught.

Meta将此事定性为SEV一级事件，这是他们第二高的严重等级。

Meta classified this as a SEV one incident, which is their second highest severity level.

而且这并不是孤立事件。

And this wasn't even an isolated thing.

Meta自己的安全主管Summer Yu之前曾提到过——我想我以前在播客里也讲过，但现在很多人都在讨论——她曾使用的一个AI代理，尽管她明确要求代理在采取任何行动前必须先征得她的确认，却还是删除了她的整个收件箱。

One of Meta's own safety directors, Summer Yu, had previously shared, and I think I've talked about this on the podcast in the past, but a lot of people are talking about this, but she kind of previously shared that an AI agent she was using deleted her entire inbox even though she explicitly told it to confirm with her before taking any action.

所以这里确实出现了一些问题。

So there there are some issues happening here.

我觉得更广泛的数据相当惊人。

I think the broader numbers are pretty crazy.

Hidden Layers 2026年的报告发现，自主代理现在占企业报告的AI泄露事件的八分之一以上，另一份报告则发现，47%的组织已观察到AI代理表现出非预期或未经授权的行为。

A hidden layers 2026 report found that autonomous agents now account for more than one in eight reported AI breaches across enterprise, and a separate report found that 47% of organizations have observed AI agents exhibiting unintended or unauthorized behavior.

我觉得这正是在演示时做这些事和实际部署它们之间的差距真正显现的地方。

I think this is kind of where the gap between, like, when you do these things as a demo and you actually are deploying them, it kinda really shows up.

这些代理在受控环境中非常出色，但当你将它们放入包含真实数据和真实权限的企业环境中时，一旦出错，后果可能非常严重。

The agents are very impressive in controlled settings, but when you put them into a real enterprise environment with real data and real permissions, if if something goes wrong, it can go wrong in a big way.

我要说的是，我认为这里确实存在一个问题，那就是你究竟使用了什么样的AI模型，以及它有多智能。

Now what I will say is I think there's definitely an issue here with how just honestly, what AI model you're using and how smart it is.

如果我像使用Claude Cowork那样完成很多我日常的任务，但用的是较旧版本的Claude，甚至是ChatGPT，它根本没那么聪明。

If I was doing, you know, some a lot of the tasks that I do with Claude Cowork, for example, and I was using kind of like an older version of Claude or even ChatGPT, it's just not smart enough.

我不太想把Meta推上被告席，但他们的Meta AI和一些LAMA模型确实属于这一类。

I sadly not trying to, you know, throw Meta under the bus, but their Meta AI is kind of in that camp with some of their LAMA models.

它们就是不够聪明。

They're just not as smart.

它们的能力也不够强。

They're not as capable.

所以如果你试图用它们来完成这些非常重要的任务，肯定会出错。

And so if you have you know, if you're trying to use them to do these kind of highly important tasks, it's gonna mess up.

它们会幻觉。

It's gonna hallucinate.

我认为这些问题中的很多——虽然不是全部——实际上已经被最新的Claude Opus 4.6版本解决了，它表现得非常出色。

Now I think a lot of these problems, and not all of them, have been actually solved by some of the latest versions of Claude Opus 4.6, has been amazing.

在过去一周里，我没有任何疯狂的事故，也没有出现任何疯狂的幻觉或问题。

I have not had any crazy incidents or any crazy kind of hallucinations or problems.

在过去一周里，我用Cloud Cowork和Cloud Code构建了两个网站和两个完整的SaaS产品。

In the last week, I've built two websites and two full SaaSes in the last week using Cloud Cowork and Cloud Code.

我不是开发者。

I'm not a developer.

我对代码一无所知，但已经能做出一些非常疯狂、了不起的东西。

I don't, you know, know anything about the code, but I've been able to do some really crazy phenomenal stuff.

没有出现幻觉。

No hallucinations.

它帮我设置了服务器。

It sets up my servers.

它帮我设置了云环境。

It sets up my cloud.

它一直在做些了不起的事情。

It's been doing some amazing stuff.

所以一方面，我喜欢这些故事，因为我觉得了解这些很重要，有点像是提醒你：小心点。

So like on the one hand, I do like these stories because I think it's interesting and important to know and it's kind of like, look, watch out.

但另一方面，我觉得很多人会把这些故事当作AI末日论的依据。

But at the other side of this, I do feel like a lot of people will use these stories as kind of this AI doomerism.

是的，这就是为什么你不能在没有人类监督的情况下信任AI。

Like, yeah, that's why you can't trust AI without a human in the loop.

而且说实话，我并不是想刻意表达什么，但我真心认为，我们最终不得不在没有人类干预的情况下信任AI。

And look, I'm not trying to be any sort of way, but I honestly believe we're gonna have to trust AI without a human in the loop.

无论我们是否愿意，它都会被部署，并且会真正做得很好。

Whether we do or we don't, it's gonna get rolled out, and it's gonna actually do a good job.

我认为，最终——我的意思是，就在未来六个月内，许多许多任务都将被完全自动化。

I think eventually, and by eventually, mean, like, within the next six months, many, many, many tasks will be completely automated.

这完全是基于我目前在实际中看到的工具和正在使用的系统。

This is just from the tools I'm seeing on the ground and what I'm using.

所以，总之，我不想陷入全盘AI末日论。

So, anyways, don't wanna go all AI doomerism.

有时候，这种说法还挺搞笑的，或者也许在Meta发生这么大的安全事件并不好笑。

Sometimes that's kind of, I guess, funny, or or maybe this isn't funny having a big security incident like this at Meta.

但如果你使用的是糟糕的模型，这种事情就一定会发生。

But if you're using bad models, that's what's gonna happen.

如果你使用的是最顶尖的模型，我就看不到未来这类事情会频繁发生。

If you use some of the best models, I just don't see this happening very much in the future.

我认为这种情况会变得越来越罕见。

I think this is something that'll be let more and more rare.

好的。

Okay.

下一个我想深入讨论的话题是2026年第一季度的风险投资数据已经公布了。

The next story I wanted to dive into is the Q1 twenty twenty six venture funding numbers are out.

它们打破了所有纪录，达到3000亿美元。

They've shattered all records at $300,000,000,000.

这些数据来自Crunchbase。

So this comes from Crunchbase.

他们每个季度都会发布这些数据，我认为这简直是我们从未见过的情况。

They publish this every quarter, and I think this is basically unlike anything we've ever seen.

今年第一季度，投资者向全球初创企业投入了约3000亿美元。

Investors poured about $300,000,000,000 into startups globally in the first quarter of this year.

这一数字比上个季度和去年同期都增长了150%。

That is up 150% both quarter over quarter and year over year.

为了让你理解这个数字的意义，仅这一季度就占了2025年全年风险投资总额的约70%。

So to put that into context, this single quarter accounted for about 70% of all venture capital invested throughout the entire year of 2025.

伙计们，这太疯狂了。

Guys, this is crazy.

对于那些说‘哦，这只是一个泡沫’的人来说，

And for people that are like, oh, you know, it's just a bubble.

他们认为资金只流向了大公司。

It's, you know, they're only giving it to the the big companies.

但实际上，资金正在广泛地分布到各个公司。

This is getting spread out to everybody.

我认为人工智能领域的集中度才是真正有趣的地方。

And I think the AI concentration is where it gets really interesting.

在这3000亿美元中，约有2420亿美元，也就是80%左右，投向了人工智能公司。

About 242,000,000,000 of that $300,000,000,000, which is eight, you know, percent, went to AI companies.

此前人工智能领域风险投资的最高占比纪录是55%，发生在去年的第一季度。

So the previous record for AI share for venture funding was about 55%, and that was in q one of last year.

所以我们今年超过了，比例从50%上升到了80%。

So we beat it this year, and it went from 50 to 80%.

你知道，这简直太疯狂了。

You know, that that's pretty wild.

我认为当你看钱实际流向哪里时，它们高度集中在顶级AI公司身上。

I think when you look at where the money actually went, it's heavily concentrated in the top AI companies.

我知道我之前说过，你看。

Now I know I did say earlier, like, look.

这其实已经相当分散了。

This is pretty spread out.

我认为相比之下，十年前风险投资的金额和分布情况是怎样的？现在虽然资金还是分散到很多小公司，但它们筹集的金额确实小得多，比如五千万、两千万这样的小额度投资。

And I think comparatively, like, you look ten years ago and how much money is getting spent in venture and where, like, this is very, the amount is still getting very spread out to a lot of the smaller companies, but they're just I mean, smaller companies are gonna raise, you know, 50,000,000, 20,000,000, smaller smaller check sizes.

然后你就发现，所有其他事情都显得被大公司远远甩在后面了。

And then you it's just everything is gonna feel like it gets dwarfed by the big companies.

比如OpenAI，就拿走了其中很大一部分。

So OpenAI, for example, took a huge chunk of this.

你知道，他们刚刚融资了1250亿美元。

You know, they just raised a $125,000,000,000.

Anthropic融资了300亿美元。

Anthropic raised 30,000,000,000.

X AI融资了200亿美元。

X AI raised 20,000,000,000.

Waymo融资了160亿美元。

Waymo raised 16,000,000,000.

仅这四轮融资就占了本季度全球风险投资总额的约1800亿美元，即65%。

Those four rounds alone account for about a 180,000,000,000 or 65% of all global venture investments for the quarter.

所以，显然，OpenAI那笔巨大的、重磅的融资会极大地影响整体数据。

So, obviously, OpenAI's big, you know, blockbuster mammoth round is gonna kind of put a put a big weight on the scale.

但我认为地理上的集中度也十分有趣。

But I think the geographic concentration is also really interesting.

美国公司融资总额达2500亿美元，占全球总额的83%，而一年前这一比例仅为71%。

US based companies raised 250,000,000,000, which is 83% of the global total, and that's up from 71% a year ago.

所以，投资正越来越集中在美国，而且我认为这主要是后期阶段的投资。

So the the investments are getting more concentrated in The United States, and I think it's overwhelmingly late stage.

2460亿美元流向了后期轮融资，其中2350亿美元流向了融资额超过1亿美元的公司。

246,000,000,000 went to late stage rounds with 235,000,000,000 of that going to companies that are raising a 100,000,000 or more.

那么，这究竟意味着什么？

So what does this actually mean?

我认为这里有几件事值得拆解一下，首先是显而易见的一点。

I think if there are a few things worth kinda pulling apart here, first is kinda the obvious.

对吧？

Right?

人工智能行业吸引资本的速度在历史上前所未有。

The AI industry is attracting capital at a pace that has no historical parallel.

对吧？

Right?

这在科技史上从未发生过，无论是互联网泡沫时期，还是移动时代，还是加密货币热潮期间。

This has never happened in tech before, not, you know, during the dot com boom, not during the mobile era, not during crypto.

没有任何事情能与之相比。

Nothing comes even close to this.

其次，这种集中度非常显著。

And second, the concentration is really significant.

当一个季度65%的全球风险投资都流向了四家公司时，我认为这表明投资者对这个市场有着非常特定的看法。

When 65% of a quarter, you know, of a quarter's global venture investment goes to four companies, I think it tells you that there is you know, investors are really thinking about this market in a specific way.

他们相信自己在下大注，并且相信基础模型层将主要由少数几家公司在主导，他们愿意开出巨额支票，以确保这些公司拥有所需的算力和人才。

They believe they're making big bets, and they believe that the foundational model layer is gonna be basically dominated by a small number of players, and they're willing to write these enormous checks to make sure that those players have the compute and talent that they need.

我从这一点中抽出的第三点，是我认为经常被忽视的，那就是：这对其他人意味着什么？

The third thing that I kinda pull out of this is something I think gets overlooked a lot, and that is basically, you know, what does this mean for everyone else?

如果你是一家不从事基础模型开发的初创公司，那么如今融资的实际难度比表面数据所显示的要大得多，因为这些超级轮次极大地推高了平均交易规模。

If you're a startup that isn't building foundational models, raising money right now is actually harder than the headline numbers suggest because the average deal size is being massively inflated by these mega rounds.

所以，如果剔除前十大交易，风险投资的整体格局会显得正常得多，甚至可能略显紧张。

So if you strip out the top 10 deals, the venture landscape looks a lot more normal, maybe even a little tight.

对吧？

Right?

然后还有一个问题，就是这种投资水平是否可持续。

Then there's also the question of whether this level of investment is sustainable.

单个季度就达到三千亿美元。

$300,000,000,000 in a single quarter.

我的意思是，这是一笔巨款。

Like, that is a ton of money.

但那些筹集这笔资金的AI公司也在大量消耗资金，用于计算资源、人才和数据。

But the AI companies that are raising this money are also burning through it on compute, on talent, on data.

他们并不是在积累巨额现金储备。

Like, they're not building huge war chests.

他们正在迅速花掉这些钱。

They are burning this money.

OpenAI、Anthropic、xAI，它们都在为下一代模型展开一场军备竞赛，而资本需求还在持续上升。

OpenAI, Anthropic, xAI, all of them are in an arms race basically for the next generation of models, and the capital requirements keep going up.

我们并不是说，嘿，你看。

It's not like we're, oh, look.

我们找到了让这件事更有效或更高效的方法。

We figured out how to make this more effective or more efficient.

我们可以用更少的钱来训练模型。

We can spend less money to train the models.

终有一天，这些公司需要展现出能证明这些投资合理性的回报，因为这些投资实在是太过庞大了。

At some point, these companies are gonna need to show returns that justify this investment because, I mean, these investments are absolutely massive.

它们还没做到这一点。

They're not there yet.

今年第一季度可能会成为人们回望的时刻，要么是AI投资真正起飞的节点，要么是事情变得有点过热的时刻。

Q one of this year might be the quarter people start to look back on as either the moment that I think AI investments truly took off or maybe the moment that things got a little too hot.

我不确定。

I'm not sure.

我认为这并不是传统意义上的泡沫，因为这项技术是真实的。

I don't think it is a bubble in in maybe the traditional sense because the technology is real.

对吧？

Right?

采用是真实的。

The adoption is real.

很多人在使用这个。

A lot of people are using this.

OpenAI 目前每月收入高达20亿美元。

OpenAI is doing $2,000,000,000 a month right now.

但我认为，资本的集中以及这些少数前沿实验室正在形成一种绝对值得密切关注的态势。

But I think that the concentration of capital and just kind of these handful of Frontier Labs is creating a dynamic that is definitely worth watching carefully.

获得资金的人和没获得资金的人之间的差距正在扩大。

The gap between, you know, who gets this money and who isn't getting this money is getting bigger.

这对竞争、创新以及哪些AI产品能够真正进入市场，都有深远影响。

This has a lot of implications for competition, for innovation, for what kind of AI products actually make it to market.

所以这些都是需要关注的事项。

So these are all things to look out for.

各位，非常感谢今天收听这个播客。

Guys, thank you so much for tuning into the podcast today.

如果你喜欢这一集，请务必在你收听播客的平台留下评分和评论。

If you enjoyed this episode, make sure to leave a rating and review wherever you get your podcasts.

这能极大地帮助这个节目。

It helps the show a ton.

这也能让我有能力为你制作所有这些播客。

It helps me be able to make all of these podcasts for you.

一如既往，如果你想访问80多个顶级的图像、音频、视频模型，包括Sora，请前往aibox.ai，因为Sora很快就要停用了——单个10秒片段就要133美元。

And as always, make sure to go check out aibox.ai if you wanna get access to over 80 of the top image, audio, video models, Sora that will soon be discontinued because it costs a $133 for a ten second clip.

在停用前，赶紧去aibox.ai体验一下吧。

Before it's discontinued, go check it out at aibox.ai.

告诉我你们的想法。

Let me know what you guys think.

非常感谢收听，我们下一期再见。

Thanks so much for tuning in, and I'll catch you in the next episode.