Anthropic的成本优势、ShopMy的15亿美元估值与灭火无人机 | 2025年11月10日

欢迎各位收看Information的TI TV。

Welcome everyone to the Information's TI TV.

我叫阿卡什·帕什里查。

My name is Akash Pashricha.

今天是11月10日，星期一。

It is Monday, November 10.

今天我们为大家准备了一场精彩的节目。

We have got a great show lined up for you today.

首先，我们将对比Anthropic和OpenAI对未来几年的预测。

First up, we are comparing Anthropic and OpenAI's projections for the coming years.

接着，我们将采访ShopMy的创始人兼首席执行官，这是一家刚刚以15亿美元估值完成新一轮融资的创作者电商平台。

We're then talking to the founder and CEO of ShopMy, a creator e commerce company that just raised new funds at a $1,500,000,000 valuation.

我们还将深入探讨人工智能如何帮助应对野火。

We're also taking a closer look at how AI could help fight wildfires.

我们还将邀请AWS探讨如何加速AI代理的普及。

We are also bringing on AWS to talk about how to accelerate the adoption of AI agents.

最后，我们将以一场我非常期待的讨论结束节目，主题是围绕AI数据中心构建的007风格安全系统。

Finally, we are ending with a discussion I am really excited for around James Bond style security systems that are being built around AI data centers.

这是一期内容非常充实的节目。

It is a jam packed show.

那么，我们马上进入正题。

And so let's get right on into things.

Anthropic和OpenAI都对其业务做出了令人惊叹的收入预测。

Anthropic and OpenAI have both made some stunning revenue projections for their business.

但我们今天发布的一篇新文章深入探讨了未来三年内两家公司的利润率结构如何比较，以及为什么Anthropic在某些方面可能更高效。

But a new story we published today takes a deeper look at how how their margin structures could compare over the coming three years and why Anthropic could in some ways be more efficient.

现在邀请到的是撰写这篇文章的Shri Mupiti。

Joining me now is Shri Mupiti, who wrote the story.

Shri，欢迎来到节目。

Shri, welcome to the show.

很高兴你能来。

It's great to have you.

很荣幸能来这里。

Excited to be here.

那我们来谈谈Anthropic和OpenAI。

So let's talk about Anthropic and OpenAI.

我们做了很多出色的报道。

We have done a lot of great reporting.

我应该说，你对这两家公司的收入预测做了大量出色的报道。

You have done a lot of great reporting, I should say, on the company's revenue projections.

你今天发表了一篇新文章，谈到了它们的利润率结构和计算处理能力。

And you had a new story today out that talks a little bit about their margin structure, about their compute processing.

跟我们分享一下你的发现吧。

Tell us a little bit of what you found.

当然。

Of course.

所以上周，

So last week,

我们写了一篇关于Anthropic上调其收入预测的文章，比之前更高。

we wrote a story about Anthropic raising its revenue projections to be higher than before.

我们发现的一些结果是，Anthropic在其最乐观的预测中，预计2028年收入将达到700亿美元，现金储备为170亿美元。

So some of the findings that we found was that Anthropic projected to have 70,000,000,000 in revenue and 17,000,000,000 in cash in 2028 in its most optimistic projections.

但同年，OpenAI预计收入将达到约1000亿美元。

But that same year, OpenAI expects to generate about a 100,000,000,000.

Anthropic还预计最早在2027年实现现金流为正，并产生约30亿美元的现金。

Anthropic also expects to be cash flow positive as soon as 2027 and generate about 3,000,000,000 in cash.

与OpenAI相比，OpenAI在2027年的现金消耗高达350亿美元，直到2030年才能实现现金流为正。

Comparing it to OpenAI, OpenAI actually has 35,000,000,000 in cash burn in 2027 and won't be cash flow positive till 2030.

至于效率，也就是今天文章的主题，我们发现Anthropic的计算成本实际上远低于OpenAI。

As for efficiency, which was about today's story, we find that Anthropic's compute costs are actually much lower than OpenAI.

例如，今年Anthropic预计在训练和运行其AI模型的计算上花费约60亿美元，而OpenAI预计花费约150亿美元。

For instance, this year, Anthropic expects to spend about 6,000,000,000 across compute for training and running its AI models versus OpenAI expects to spend about 15,000,000,000.

到2028年，Anthropic预计支出为270亿美元，而OpenAI则为1110亿美元，后者还包括约400亿美元用于备用服务器，以应对任何意外的技术突破或爆款产品出现。

And then by 2028, Anthropic expects to spend 27,000,000,000 versus OpenAI's 111,000,000,000, which also includes about 40,000,000,000 in backup servers that OpenAI has reserved for any unexpected breakthroughs or viral product hits.

我们有没有了解过，为什么它们的战略差异如此之大？

And do we have a sense as to why their strategies are differing so much?

有的。

Yeah.

我认为部分原因可能是两家公司服务的客户类型不同。

I think part of the reason could be the different types of customers that both companies serve.

例如，Anthropic 大约 80% 的收入来自通过 API 向企业客户销售其 API 或 AI 模型，而 OpenAI 的大部分收入则来自 ChatGPT，其中许多用户实际上是免费用户。

So for example, Anthropic serves roughly 80% of its makes about 80% of its revenue from selling its API or AI models through an API to its business customers, versus OpenAI generates a vast majority of its revenue from ChatGPT, many of which users are actually free users.

第二点可能也在于，Anthropic 或 OpenAI 使用的芯片类型不同。

The second piece could also be just the fact that OpenAI or Anthropic, for example, has a different type of chips that they use.

例如，它们使用 NVIDIA、亚马逊和谷歌的芯片，而 OpenAI 主要使用 NVIDIA 芯片。

For example, they use NVIDIA, Amazon, and Google chips versus OpenAI primarily uses NVIDIA chips.

而且，Anthropic 的管理层还表示，他们会将这些芯片与特定任务相匹配，这也有助于降低计算成本，相比 OpenAI 来说。

And because Anthropic leaders also say that they match those chips to specific tasks, that could also be helping towards driving down compute costs as compared to OpenAI.

最后我要补充的是，正如我提到的，OpenAI 和 Anthropic 所服务的客户类型不同，OpenAI 的研发成本分散在许多不同的项目上。

And the last piece I'll just add is that, as I mentioned to the types of customers that OpenAI and Anthropic respectively serve, OpenAI's R and D costs are across many different types of bets.

例如，我们最近报道了OpenAI可能正在开发的一个AI音乐模型。

For example, we had recently reported about an AI music model that OpenAI might be working on.

因此，我认为这充分体现了OpenAI与Anthropic在产品和研究方向上的多样性差异。

So I think that really points to the diversity and types of products and research that OpenAI is working on as compared to Anthropic.

从更根本的层面来看，我们在这个节目中曾讨论过OpenAI的广泛战略，涉及多个不同产品，比如浏览器、智能代理，以及带有生成视频功能的社交媒体方面。

At a more fundamental level, one of the things that we've talked about on this show is OpenAI's broad strategy here with a number of different products, whether it's the browsers, the agents, the social media side of things with the generative video features.

我的意思是，OpenAI似乎正朝着某种意义上的全面主导地位迈进。

I mean, OpenAI really is going for what seems like world domination in some senses.

Anthropic的使命则似乎更加聚焦。

Anthropics mission seems to be a little bit more focused.

因此，我想知道，您刚才帮我们梳理出的关于财务方面的策略，是否反映了Anthropic有意将专注范围收窄？

And so I wonder this strategy that you've sort of helped us unearth with respect to financials, is this reflective of Anthropic just saying, Hey, we want to be a little bit more narrow with our focus?

还是说，这是OpenAI在表态：我们要做所有事情？

Or is it OpenAI saying, We want to do everything?

帮我们理清一下，究竟是哪一方的策略在这里更占主导地位。

Help us understand sort of which side of that is more dominant here.

我认为，从两家公司发布的产品类型以及大部分收入来源来看，这一点非常明显。

I think we see that with the types of products that both companies have released and also just where a lot of the revenue seems to be coming.

正如我提到的，Anthropic 的收入有 80% 来自企业客户通过其 API 的使用，而 OpenAI 的大部分收入则来自 ChatGPT。

As I mentioned, Anthropic generates 80% of its revenue from business customers of it through its API versus OpenAI has a vast majority of it coming from TapGPT.

OpenAI 还表示，其他潜在的收入来源包括新产品或免费用户的变现，我们了解到这可能涉及广告或联盟费用。

Other areas that OpenAI has also said that it could generate revenue from is, like, new products or free user monetization, which we've learned to potentially be advertising or affiliate fees.

因此，OpenAI 预期的盈利方式具有多样性，同时正如我之前提到的，我们在音乐、浏览器技术等领域也看到 OpenAI 正在开展多样化的研究。

So there is a diversity in the way that OpenAI expects to generate money, but also the types of research that we're seeing OpenAI work on, as I mentioned around music and browser technology and things like that.

这类预测通常会多久改变一次？

How often do projections like these end up changing?

我认为，公司很可能每季度都会进行预测。

I would say that companies probably are doing quarterly projections.

这些

These

我们之前报道的这些预测是今年夏天的数据。

projections that we had reported on are from over the summer.

而且

And

对。

so Right.

就在过去的几个月里，OpenAI 和 Anthropic 的计算支出很可能已经增加。

Even just in the last couple of months, it's likely that OpenAI and Anthropic's compute spending has gone up.

我知道，例如，Anthropic 有一些尚未公布的未签署协议。

I know, for example, Anthropic has a number of unsigned deals that haven't been announced yet.

而正如我们所知，OpenAI 已经签署了价值 1.4 万亿美元的协议，以增强其计算资源获取能力。

And OpenAI, as we know, has signed 1,400,000,000,000.0 in deals just to boost its compute access.

因此，这些预测很可能已经上调，计算支出可能比我们之前报告的还要高。

And so these projections have likely gone up, and compute spending is probably higher than what we even reported on.

但即便如此，如果 Anthropic 确实在朝着这个方向推进，这似乎对其融资努力非常有利。

But even still, this seems to bode pretty well for Anthropic's fundraising efforts, should they sort of be marching towards that?

是的。

Yeah.

上周我们报道过，鉴于Anthropic今年早些时候的融资过程中已上调了收入预测，如果Anthropic现在进行融资，其估值可能达到3000亿至4000亿美元，这比其上一轮估值有所提升。

Last week we had reported that if Anthropic were to raise, just given that they had increased their revenue projections from the last fundraise process earlier this year, Anthropic could raise at a valuation of between 300 and 400,000,000,000, and that's a step up from its last valuation.

因此，我认为这些预测对投资者来说非常有帮助，让他们能够了解这两家公司的现状，并决定是否在当下进行投资。

So I think that these projections are really helpful for investors to know what's up with both companies and decide if they want to invest at the time.

很好。

Great.

好了，Shree，感谢你前来做客。

Well, Shree, I want to thank you for coming on.

这是一个快速变化的故事，等你有更多消息时，我们再请你回来分享。

It's a fast moving story, and we'll have you on again when you've got more to share with us.

谢谢。

Thanks.

好的。

Okay.

创作者经济和电子商务领域正变得越来越融合，专注于这两个行业交叉领域的公司正获得投资者越来越多的关注。

The creator economy and ecommerce sectors are becoming increasingly intertwined, and businesses building at the intersection of these two industries are getting a lot more attention from investors.

ShopMy 是这一领域的一家公司。

ShopMy is one company in that category.

这家精心策划的创作者电商公司上个月以15亿美元的估值筹集了7000万美元。

The curated creator e commerce company raised $70,000,000 last month at a $1,500,000,000 valuation.

现在加入我们的是该公司创始人兼首席执行官哈里·雷恩。

Joining me now is Harry Raine, founder and CEO of the company.

哈里，欢迎来到TI TV。

Harry, welcome to TI TV.

很高兴有你来。

It's great to have you.

谢谢你们邀请我。

Thanks for having me.

我很喜欢你们在《The Information》上所做的工作，这让我感到非常荣幸。

I'm a big fan of what you guys do with The Information, so it's an honor.

非常感谢。

Well, thanks a lot.

我一直在你们平台上寻找一些新的穿搭。

I I've been looking on your platform for some new outfits that I can wear.

我知道你们的选择很多，创作者也为我提供了很多选项，所以我也在关注你们的动态。

I know that there's a lot of selection, and I know that the creators have a lot of options for me, so I I also am I'm watching what you do.

我们来聊聊 ShopBy 吧。

Let's talk about Shop By.

对于我们这些不太了解的人，这家公司是卖给谁的？

Tell us, for those of us who aren't familiar, who does the company sell to?

你们的实际产品是什么？

What is the actual product?

你们是怎么赚钱的？

How do you make money?

跟我们说说这方面的情况。

Talk to us about that.

嗯。

Yeah.

我非常喜欢。

I love it.

我们正在为未来精选电商构建基础设施。

So we so we're building the infrastructure for the future of curated commerce.

我们服务三大不同的利益相关方。

We serve three different stakeholders.

首先是内容创作者。

First is obviously content creators.

我们帮助他们通过推广自己喜爱的产品获得报酬。

We help them get paid promoting the products they love.

他们通过联盟链接、我们构建的数字 storefront 产品以及品牌合作来实现这一点。

The way they do that is through affiliate links, these digital storefront products that we built, as well as brand partnerships.

第二部分是直接为品牌提供服务。

Second piece is we serve brands directly.

我们为他们提供一切所需，以打造世界级的营销计划。

So we give them everything they need to build a world class career marketing program.

因此，这涵盖了从基于数据发掘人才、礼品赠送、联盟追踪、社交聆听到性能工具等一切你所能想象的方面。

So that's everything from discovering talent based on data, that's gifting, that's affiliate tracking, social listening, performance tooling, everything you might imagine.

而这正是我们的核心业务——为品牌提供的SaaS产品。

And that's really our bread and butter is a SaaS product for brands.

第三部分是我们现在引入了消费者。

The third piece is now we've involved shoppers.

我们为他们提供了一种结构清晰、轻松便捷的购物体验，让他们可以轻松购买这些顶级策展人的推荐商品。

So we're giving them a shopping experience, incredibly well structured, easy way to shop the recommendations of these top curators.

你们目前的营收是多少？

And how much revenue are you guys generating right now?

过去几年，我们的增长势头非常强劲。

We we've been on kind of a tear the last couple years.

两年前，我们的营收约为400万美元。

So we ended, two years ago about 4,000,000 in revenue.

去年是2700万。

Last year, 27.

今年，我们的收入大约是8000万美元。

This year, we're about 80.

我们为品牌合作伙伴带来了约十亿美元的成交总额（GMV）。

We were driving about a billion dollars in GMV to our brand partners.

平台上拥有约20万名内容创作者，每月有1000万购物者在平台购物。

We have about 200,000 content creators on the platform, 10,000,000 shoppers shop each month.

一切都在迅猛发展。

Things have been kind of explosive.

我认为这恰恰反映了购物方式的变革，以及商业格局的显著变化。

And I think it's just it's just indicative of the way that shopping's changing and the commerce landscape's obviously overwhelming.

这种由策展层将海量选择过滤并简化为终端消费者可消费内容的理念，已成为商业中一股强大的力量。

And this idea of this curator layer taking the in the massive amount of options and filtering it down for the end consumer has been like a really powerful force in commerce.

所以，到2025年，你们预计全年收入将达到8000万美元，是吗？

And so $80,000,000 in 2025, that's where you're expecting to end the year.

你们盈利吗？

And you're profitable?

嗯。

Yeah.

嗯。

Yeah.

我们从2024年开始就盈利了。

We've been profitable since 2024.

所以显而易见的问题是，为什么我们要融资？

So obvious question is why do we raise?

我认为这里的机遇太大了。

I think the answer is the opportunity here is so big.

人工智能和大语言模型正在彻底颠覆这场商业变革。

This commerce is shaking up so massively with LLMs of AI.

我们认为人类的品味在这一未来中扮演着重要角色，我们希望占据这一领域。

And we think humans are and taste play a big role in that future, and we kinda wanna own that lane.

所以占据这一领域。

And so own that lane.

那么，你们计划把这笔钱花在什么地方？

So what do you plan to spend the money on?

嗯。

Yeah.

有几件事。

There's a few things.

所以，实际上，退一步谈谈我们是如何达到15亿美元估值的，以及我们在市场上做了哪些改变，可能会有所帮助。

So it might be helpful actually to take a step back and talk about kinda how we got to the $1,500,000,000 valuation and what we changed in the market.

如果你回顾过去十年，站在一个品牌的立场上，从营销角度来看，只有两种真正有效的增长方式。

So if you think of the last 10, put yourself in the shoes of a brand, there's been two real ways to grow from a marketing standpoint.

一种是创作者营销。

One is creator marketing.

创作者营销的价值在于它非常真实。

And the value of creator marketing is it's very authentic.

你是通过与受众建立关系的创作者来销售产品。

You're selling through content creators that have a relationship with their audience.

但很难规模化，因为你是在和人合作，而与人合作本身就很难分配有意义的预算。

But it's hard to scale because you're working with people and inherently working with people is hard to kind of allocate meaningful budget.

第二种选择是效果营销。

The second option is performance marketing.

所以你可以制作几个广告，进行非常高效的实验，并且能够很好地扩展。

So you build a couple ads, you can experiment incredibly well, and it scales very effectively.

你可以每月投入数百万美元，并看到直接的投资回报。

So you can put in millions of dollars a month and see direct ROI.

缺点是不够真实。

Downside, not authentic.

这是某人制作关于自己的广告，推广说：‘看看我，看看我的新产品。’

It's someone making an ad about themselves, pushing out like, Hey, look at me, look at my new product.

所以我们所做的，是结合了两者的最佳优势，我们称之为创意效果营销，它让你既能像效果营销那样大规模扩展，又能保持真实性。

So what we've done is taking kind of the best of both worlds, and we're effectively calling creative performance marketing, which allows you to scale at the scale of performance marketing while maintaining it authentic.

我们实现这一点的方式是通过大量自动化追踪。

And the way we do that is through a ton of we track everything automatically.

我们有一个定价和预算分配引擎，用于确定你应该与哪些创作者合作，以及应该支付他们多少费用。

We have a price and budget allocation engine to figure out what creators you should work with, how much you should pay them.

因此，它的扩展性非常强。

So it scales incredibly effectively.

我认为，任何与我们合作并大力投入的品牌，都已看到它成为他们真正主导的新渠道。

And I think any brand that we work with that has really leaned in has seen it become their real dominant new channel.

不过，当你谈到扩展时，我注意到的一点是，ShopMy与TikTok商店等平台的一个区别在于，你们所推广的产品和合作的创作者明显更偏向高端市场。

When you talk about scaling, though, one of the things I noticed is is that one of the ways in which ShopMy differentiates itself from something like a TikTok shop is you guys are a lot more upmarket in terms of the products that you seem to be advertising with your brands and creators.

那么，如果产品范围相对狭窄，同时又受限于能够推广这些产品的创作者数量，你们该如何思考扩展呢？

How do you sort of think about scaling then if it's sort of a narrow set of products, but then also it's sort of limited by the number of creators that are available to tout these products.

对吧？

Right?

是的。

Yeah.

是的。

Yeah.

我认为我们所处的产品领域需要品味和一定程度的人类信任。

The I would say the we work in the product space where it requires taste and a level of human trust.

所以这包括你身上佩戴的任何东西，你穿的衣服，以及你如何装饰你的家。

So that's anything that you put on your body, anything you like, your what you wear, how what you how you decorate your house.

TikTok商店更像是低端产品，属于快速冲动型购买。

The TikTok shop is kinda more like lower end products, quick impulsive buys.

这是一个更加有意图、更注重思考的世界，而且规模巨大。

This is a much more intentional, thoughtful world, and it's enormous.

我的意思是，人们在需要品味的产品上花费的金额是巨大的。

I mean, the amount people spend on things that require taste is massive.

你说得对，我们确实受限于策展人的数量。

You're right that we're limited by the number of curators.

我认为，我们在这里投入资金的主要目标之一就是扩大创作者群体，以满足品牌们试图分配的更多预算。

And I think, like, one of the main things we're spending money on here is to try to expand that creator pool and serve more of the brand's budgets that they're trying

你是否担心这种趋势：人们转向聊天机器人来寻找产品推荐，而我们过去依赖创作者，现在依然如此。

to Are you concerned at all with this shift towards people turning to chatbots to find recommendations for products and the idea that we used to use creators, and we still do.

我不是说它完全消失了，但确实有人在讨论，人们是否会更多地转向ChatGPT来寻找产品推荐，而不是依赖创作者？

I'm not saying it's gone away entirely, but there is a bit of a discussion around, hey, could people turn more to ChatGPT to find recommendations rather than turning to creators?

你对此感到担忧吗？

Are you worried about that at all?

是的。

Yeah.

这是个很好的观点。

It's a great point.

我认为从这个角度来看，购物其实有两种形式。

And I think we so there's kind of two forms of shopping in that sense.

一种是我们所说的‘目标型购物’，即我清楚自己想要什么。

One is what we call hunt based shopping, which is I know exactly what I want.

我需要去找到它。

I need to go find it.

黑色高腰牛仔裤。

High rise black jeans.

大型语言模型在这一点上非常出色。

LLMs are phenomenal at that.

它们能精准地找到你想要的东西。

They're going find you exactly what you want.

另一方面，我们关注的是收集者，他们是在世界上探索，发现你可能感兴趣的商品并将其推荐给你。

The other side is what we focus on is more the gatherers, where it's, I'm going out in the world, I'm finding things that you might be interested in buying and bringing them to you.

你不能通过聊天来开始这种行为，因为你并不知道具体要找什么。

You can't really start that with a chat because you don't know what it is.

因此，我们所推崇并帮助融入平台的是这种收集型模式。

So it's gathering form that we're celebrating and helping bring into a platform.

你主要向品牌和创作者销售，对吧？

Now, you mostly sell, you said, to the brands and to the creators.

你还提到了针对消费者。

You also talked about sort of targeting the consumers.

上周末我试着注册了这个应用。

I tried to register for the app actually over the weekend.

所以我看到需要一个进行中的流程，或者我得去申请。

And so I see that there needs to be sort of an in flight process or I got to apply.

所以我的理解是，目前并没有面向消费者的应用。

And so the way I understand it, there's no actual consumer facing app right now.

对吧？

Right?

你们有考虑进入这个领域吗？

You thinking about going into that space?

我们距离上线只有两周了。

We are two weeks away.

所以不幸的是，还有两周左右，敬请期待。

So unfortunately, two weeks around, but stay tuned.

那两周后会有什么新动态？

Well, what's coming in two weeks?

跟我们说说吧。

Tell us about that.

是的。

Yeah.

我们即将推出我们的应用。

So we're launching our app.

我们对此非常兴奋。

We're really excited about it.

这是为购物者提供这些潮流引领者推荐内容的平台。

It's for shoppers to consume the recommendations of these tastemakers.

这是一个令人惊叹的平台。

It's an incredible platform.

你可以收藏商品、标记喜欢的内容、获取信息流、购物，并通过所有明显的AILM策略搜索信息流。

You can wishlist things, favorite things, get a feed, shop, search the feed through all the obviously AILM tactics.

是的。

Yeah.

太棒了。

It's amazing.

我等不及了。

I can't wait.

那么，你如何看待与这些作为发现引擎的社交媒体平台竞争呢？

So then how do you think about then competing with all these social media platforms, which are the discovery engines?

我们的初衷从来不是让创作者离开这些平台。

The this is, the intention is never to bring the creators off of those platforms.

他们仍然通过这些渠道发布内容。

They still post through those channels.

我们只是在后台进行聚合。

We just aggregate behind the scenes.

我们开发这款应用的初衷并不是要成为一个新的娱乐目的地。

Our like, our intention with the app is not to be a new entertainment destination.

而是打造一个专为你而生的购物网站。

It's to be a shopping site built just for you.

因此，由你来选择哪些策展人构成你的推荐内容。

So you're choosing which curators make up your recommendations.

然后你会获得一种相当通用的购物体验，同时还能享受到定制应用中所有那些出色的功能。

And then you get what feels like a pretty general commerce experience with all the, like, kind of amazing parts that you can build into a custom app.

而且界面简洁。

And it's clean.

太棒了。

It's awesome.

所以当你真正想买东西时，你可以过来，说：好吧。

So it's when you're actually looking to buy something, you can come and say, alright.

我已经信任了15个人。

I already have 15 people I trust.

让我搜索一条夏裙，或者你想要的任何东西，提前获得你肯定会喜欢的推荐。

Let me search for a summer dress or whatever you're looking for and get that recommendation that you know you're gonna like ahead of time.

很好。

Great.

哈利，我期待看到这个应用。

Well, Harry, I'm looking forward to seeing that app.

恭喜你完成融资，我们很快会再邀请你回到节目中。

And congrats on the fundraise, and we'll have you back on the show again soon.

非常感谢。

Thank you so much.

一家新的初创公司正在应对一个重大挑战，以减少野火的发生频率和严重程度。

Okay, a new startup is tackling a big challenge to reduce the frequency and severity of wildfires.

Seneca最近以6000万美元的融资正式启动。

Seneca recently launched with a $60,000,000 financing ground.

它正在开发一种自主灭火系统，利用人工智能来阻止火灾并将响应时间缩短至十分钟以内。

It is building an autonomous fire suppression system that uses AI to stop fires and cut response times to under ten minutes.

现在邀请我一起讨论此事的是该公司创始人兼首席执行官斯图尔特·兰德斯伯格。

Joining me now to discuss this is the company's founder and CEO, Stuart Landesberg.

斯图尔特，欢迎来到节目。

Stuart, welcome to the show.

很高兴有你加入。

It's great to have you.

非常感谢邀请我参加。

Thanks so much for having me.

谈谈你是如何对这项使命产生如此深厚感情的，以及你们公司到底做什么。

Talk to me about how you got so attached to the mission here, and what your company actually does.

Seneca 制造自主早期响应系统，我们的核心目标是专注于那些如今依靠现有技术对消防员来说效率低下、不安全或根本不可能应对的场景。

Seneca makes autonomous early response systems, and fundamentally what we try to do is focus on use cases that today are inefficient, unsafe, or impossible for firefighters with existing technology.

我们的首款实际产品是一支由五架自主灭火无人机组成的机队。

Our first product in practice is a fleet of five autonomous suppression copters.

你可以称它们为无人机，但它们体积很大，每架重达二百五十磅。

You can call them drones, but they're very large, weigh two fifty pounds each.

每架无人机携带超过一百磅的A类泡沫。

Each carries over 100 pounds of class A foam.

它们能够比目前更快地做出响应，有望在五分钟或十分钟内抵达火场。

And they can respond earlier than is possible today, hopefully getting to fires within five or ten minutes.

这可以防止火势蔓延。

And that can keep them from spreading.

它们能够进入难以到达、对消防员而言极其危险的区域，并能在当前消防员操作效率低下的情况下工作，为前线人员提供类似兵力倍增的效果。

They can access places that are hard to reach, are really dangerous for firefighters, and they can operate in situations that are really inefficient for firefighters today, giving sort of a force multiplication effect to folks on the front lines.

所以这些是大型无人机。

And so these are large drones.

我见过它们。

I've seen them.

它们通常被放在皮卡的车尾，比我们在Instagram上偶尔看到的遥控无人机要大得多。

They kind of put them on a back of a pickup truck, and they're bigger than sort of the remote control ones that we've seen sometimes on Instagram.

你们已经卖出了多少台？

How many of these have you sold?

我们不具体谈论数字，但我们确实有一些消防机构已同意在明年部署它们。

So we're not talking specifically about numbers, but we do have a number of fire agencies who've agreed to deploy them next year.

我们把它们视为一组五台来考虑。

And we think about them as going in a group of five.

五台一组的神奇之处在于，这五架飞行器——没错，它们的大小介于普通玩具无人机和大型直升机之间——一组五台所能提供的灭火能力，大致相当于一辆野外消防车。

And the magic of a group of five is that across five aircraft, and you're right, they're sort of like the average of your little hobby drone and a large helicopter, across a group of five, you can get about as much suppression power as a wildland fire engine.

而且，如果能在探测后最初的五到十分钟内将它们部署到位，我们的模型显示，你甚至可以阻止一场95%风险的火灾，或者至少为地面人员争取足够的时间赶到现场，防止真正的大规模破坏。

And there's something too, if you can get that there in the first five to ten minutes post detection, our modeling shows that you can stop even a 95% risk fire or at least give the ground troops enough time to get there to prevent real devastation.

所以，如果我们不谈你们卖出了多少，那现在有多少个机构在使用它呢？

So so if we're not talking about how many have you sold, then how many how many agencies are using it?

给我一个规模上的概念。

Give me a sense of scale.

你们的客户是谁？

Who are the customers?

我们现在的客户主要分为三大类。

So our customers today fall into three major groups.

他们是消防机构，一些全国最具前瞻性的团队。

They're fire agencies, some of the most forward thinking folks in the country.

我们已经公开提及与圣贝纳迪诺县的合作，这是美国面积最大的县，管辖范围超过2万平方英里，位于南加州。

We've already talked publicly about partnerships with San Bernardino, the largest county in the country with over 20,000 square miles in their jurisdiction in Southern California.

显然，去年我们与帕利塞兹地区的朋友们关系非常密切。

Obviously very close to our friends in the Palisades last year.

我们已经与落基山脉西部的多个机构合作飞行，包括科罗拉多州的阿斯彭。

We've talked about flying with agencies across the Mountain West, including Aspen, Colorado.

我们已为数十个机构执行了数千次任务，覆盖加利福尼亚、怀俄明、蒙大拿和科罗拉多等地。

We've flown this for dozens of agencies, thousands of missions across California, Wyoming, Montana, Colorado.

因此，预计该系统将在2026年广泛部署，并在2027年实现更大规模的应用。

So expect to see it widely deployed in 2026 and even a broader scale in 2027.

接下来，我简单谈一下除消防机构之外的其他领域。

And then I'll speak quickly just beyond fire agencies.

还有公用事业公司、开发商以及各种土地所有者，他们都非常重视为自己的财产提供适当的保护。

There's also utilities, developers, and various landowners who obviously really think hard about making sure they have the right protection for their property.

这里稍微普及一下消防基础知识，因为我对这方面的了解还不够深入，但无人机体积很大，那水是从哪里来的呢？

And a bit of firefighting one zero one here because I I don't know as much as I should, but the drones are big, but where does the water come from?

我的意思是，我觉得它们不可能携带那么多水，对吧？

I mean, they can't house that much water, I think, in them.

对吧？

Right?

五的神奇之处在于，每个设备都预装了超过100磅的容量。

The magic of five is that each of them comes preloaded with a little over 100 pounds.

目前的容量约为14加仑。

The capacity today is about 14 gallons.

当你使用A类泡沫并以极高压力喷洒时，气流会吸入空气，与泡沫混合后，会产生5到15比1的膨胀比。

When you use Class A foam and you spray it at really high pressure, the airflow actually sucks air in, and with the foam, you get what's called an expansion ratio of somewhere between five and fifteen to one.

因此，500磅的载荷实际上能发挥出2500到5000磅的效果，这相当于野外消防车约500加仑的灭火能力。

So 500 pounds will end up playing more like 2,500 to 5,000 pounds, which gets you to approximately the 500 gallons of suppression power you'd see in a wildland engine.

所以起初可能不太直观，但实际上，飞机搭载的载荷就能提供相当可观的灭火能力。

So they they it may not be intuitive at first, but actually you get quite a bit of suppression power just from the payloads that's onboard the aircraft.

我漏掉了这部分。

I I and I I missed that part.

所以这其实不是水。

So it's not actually water.

而是一种特殊的泡沫，没错。

It's a special foam that Exactly.

搞定。

Getting Okay.

现在跟我谈谈这里的长期愿景吧。

And now talk to me about the long term vision here.

我的意思是，消防，我不确定这个市场有多大。

I mean, firefighting, I don't know how big a market this is.

我知道问题正在恶化，但在很多方面，你是在向公共部门销售，因为他们希望购买这项技术。

I know the problem is getting worse, but many ways, you're selling to the public sector in terms of them wanting to buy the technology.

你们也在考虑向私营部门销售更多产品吗？

Are you also looking to sell more to the private sector?

这家公司的长期愿景是什么？

What is the long term vision for the company?

有很多利益相关者关心这个问题，对吧？

There are lots of stakeholders who care about this, right?

显然，我们看到许多公用事业公司把这视为一种生存风险——防止火灾危险。

Obviously, we've seen utilities, for many of them, this is an existential risk, stopping fire danger.

但你也看到，从木材种植者到拥有位于野地-城市交界区高价值房产的主要开发商，都受到了影响。

But you also see everything from timber farmers to major developers who have got properties, high value properties in the wildland urban interface.

但如果你从长远来看，如果我们无法阻止美国西部的房屋因野火而被烧毁，那么美国西部的房屋保险将不复存在，对吧？

But if you look at the long term fundamentally, if we can't stop homes from burning down as a result of wildfire across the American West, we are going to lose home insurance across the American West, right?

我们不能通过公平计划和其他方式来分担野火造成的损失。

We're not going to socialize with the fair plan and other things, the losses from wildfire.

因此，我们必须解决这个问题。

So we have to stop this problem.

因此，从长远来看，我认为每一个位于高风险地区的社区——不幸的是，美国西部的许多地区都属于此类——都需要采取积极措施，因为这只是时间问题，一旦我们开始失去保险，就会接着失去抵押贷款，这将威胁到全国一半地区的美国生活方式。

And so, when I look out over the long term, I think of this as something that every single community that in a high risk place, which unfortunately, much of the Western United States needs to be taking a proactive look at, because it is only a matter of time before we all start losing our insurance, which means we're going to start losing our mortgages, which is going to threaten the American way of life across half our country.

因此，我们将此视为美国西部的生存危机，而技术可以在解决这一危机中发挥切实作用。

So we look at this as an existential crisis for the American West, and one that technology can play a real role in solving.

因此，这令人兴奋且充满激情，我们非常感激能在公共和私营部门拥有如此出色的合作伙伴。

So it's exciting and passionate, we're grateful to have terrific partners across public and private sectors.

很好。

Great.

斯图尔特，我非常感谢你所做的工作，也非常感谢你参加我们的节目。

Well, Stuart, I really appreciate the work you're doing, and I really appreciate you coming on the show.

谢谢你的到来。

Thanks for coming.

很高兴能来。

A pleasure.

随时欢迎你再来。

Back anytime.

好的。

Okay.

我们的下一个环节将由我们的赞助伙伴亚马逊云科技带来。

Our next segment is with our presenting partner, Amazon Web Services.

我们已经写了很多关于企业采用AI代理所面临的挑战的信息，其实并没有单一的因素导致这种困难。

We have written a lot at the information about how challenging it is for businesses to adopt AI agents, and there's no one thing that makes it difficult, really.

企业每天都会面临众多问题。

There are a multitude of issues that companies face every day.

因此，我想邀请AWS的总监肖恩·纳迪，让我们从一线视角了解这些障碍是什么样子，以及如何克服它们。

And so I wanna bring on Shown Nandy, a director at AWS, to help bring us a view on from the ground on what these obstacles look like and how they can overcome them.

肖恩，欢迎再次做客节目。

Shown, welcome back to the show.

很高兴

It's great

见到你。

to have you.

博施，很高兴看到

Bosch, it's good to see

你和这样一个有趣的话题。

you and such a fun topic.

让我们谈谈智能代理。

Let's talk about agents.

你知道吗，我今天早上看到了这份报告。

You know, I I saw the report this morning.

麦肯锡本周末发布了一份关于人工智能采用的报告，报告的标题指出，许多公司已经采用了人工智能。

McKinsey actually put out this weekend talking about AI adoption, and the the headline of the report really was that a lot of companies have adopted AI.

但人工智能的规模化应用则是另一回事，实际实现规模化的企业比例要小得多。

Scaling AI is a much different story, and it was a much smaller proportion there.

因此，我想和你聊聊在扩展智能代理时面临的挑战。

And so I want to talk to you about what the challenges are scaling agents.

客户向你反馈，在将这项技术全面推广到整个公司时，他们遇到了哪些具体困难？

What are the challenges that customers are telling you they're facing with respect to actually rolling this out across their entire companies?

是的，绝对如此。

Yeah, absolutely.

听我说，这是一个极具现实意义的话题，因为每个人都在谈论它。

Look, this is such a topically relevant concept because everyone's talking about it.

我们看到客户已经取得了首批智能代理的成功案例，有些客户甚至已经实现了大规模部署。

We're seeing customers deploy these first agentic wins, and some customers are at a massive scale.

但我认为，我们正看到一些共同的挑战线索。

But I think we're seeing a common thread of challenges.

也许我先告诉你它能有多大。

And maybe I'll first tell you how big it can be.

我曾与WPP同台，这是一家知名的广告、等等，如今也是科技公司，他们已经部署了超过七万个代理到生产环境。

I was with WPP on stage, a well known advertising, etcetera, technology now company, And they have deployed over 70,000 agents to production.

所以人们确实可以在大规模上实现这一点。

So people can do this at massive scale.

我认为挑战在于如何安全、可靠、快速且经济高效地实现规模化。

I think the challenge has been doing it at scale safely, securely, quickly, cost effectively.

我来给你一个非常简单的类比，阿卡什，我相信这会让人们产生共鸣，几乎是一个人类的类比。

And I'll just give you a really simple analogy, Akash, that I think will resonate with people, a human analogy almost.

如果你有幸拥有一支人类团队在生活中的各个领域帮助你——这正是AgenTiC所要呈现的场景——你会拥有一位法律顾问、一位财务顾问，甚至一位旅行顾问、一位规划顾问，对吧？

If you had the luxury of having a team of humans help you in life, which is what AgenTiC's going look like, you'd get a legal advisor, a financial advisor, maybe even a travel adviser, right, a planning adviser.

你拥有这么多顾问。

You have all these advisers.

拥有一位顾问相对容易。

And having one adviser is pretty easy.

他们把笔记记在笔记本里。

They keep their notes in their notebook.

他们会给你打电话。

They call you.

他们对你相当了解。

They know you pretty well.

他们可以访问你的资料。

They have access to your stuff.

但当你拥有一支团队时，你就得考虑每个人分别拥有什么权限？

But when you have a team, you've got to think about like, what access does everyone have?

我的旅行顾问能访问我的投资策略吗？

Does my travel adviser have access to my investment strategy?

不能。

No.

但他们可能需要一些基本的财务信息。

But they probably need some basic financial info.

他们如何共享笔记？

How do they share notes?

你如何限制这些顾问？

And how do you constrain all these advisers?

你如何确保他们把时间花在正确的事情上？

How do you make sure they're spending the right amount of time on things?

这就是扩展AgenTic时的一些挑战，即如何处理所有这些问题，我可以为你非技术性地分解这些部分，是的。

And that's some of the challenge with scaling AgenTic is looking at how to deal with all and I can break down those pieces for you in not technical, Yeah.

但在

But in

我的意思是，从我们这边在实地听到的情况来看，问题的一个类别是产品本身需要一些打磨。

mean, look, what we've been hearing on the ground from our end here is one category of the problem is that the product itself needs a little bit of polishing.

另一个类别是成本的可预测性。

Another category is the predictability of of costs.

我的意思是，你如何判断在这上面需要花多少钱？

I mean, how do you even tell how much you need to spend on this stuff?

我很好奇，在这两个类别中，目前哪个问题更突出？

I'm curious, in those two categories, which of those seems like a bigger problem right now?

是的。

Yeah.

让我从几个角度来说一下。

Look, I'll put it in a couple ways.

大多数组织在构建其数据、基础设施和应用基础时，目标是商业智能和机器学习，甚至是生成式AI，而不是代理式系统。

Most organizations built their data and their infrastructure and application foundations targeting maybe business intelligence and machine learning, even generative AI, not agentic.

因此，你提出的这些问题，通常并不是特别技术性的。

And so the problems you laid out, the problems don't tend to be super technical in nature.

我们之前的一期节目中讨论过关于人员方面的问题，比如首席AI官。

We talked on a prior show about the people side of it and chief AI officers.

但在这方面，一开始的成本并不是问题，成本在你扩展时才会变得重要。

But on this side, it's not cost on day one, cost matters when you scale.

成本是你无法真正做大时的阻碍，但不会阻止你起步。

Cost is what holds you back from getting really big, but not getting started.