OpenAI’s Multi-Billion Ad Strategy, Modal Labs CEO on AI Infrastructure, Twilio’s Voice AI Surge

欢迎各位来到《The Information》的TITV节目。

Welcome everyone to The Information's TITV.

我是阿卡什·帕什里卡。

My name is Akash Pasricha.

今天是5月26日，星期二。

It is Tuesday, May 26.

希望你们度过了一个愉快的长周末。

Hope you had a great long weekend.

在开始今天的节目之前，提醒大家，《The Information》年度AI大会的门票现在已经开售了。

Before we get into today's show, tickets are now on sale for the Information's annual AI Agenda Live Conference.

活动将于9月23日在旧金山现代艺术博物馆（SFMOMA）举行。

It's taking place in San Francisco on September 23 at SFMOMA, the Museum of Modern Art.

目前可以抢购早鸟票了。

You can now lock in early bird tickets.

更多详情请访问我们的活动页面：theinformation.com/events。

We have more details on our events page at theinformation.com/events.

欢迎大家去了解一下。

I encourage you to check it out.

今天节目的第一件事，是《The Information》独家报道，关于OpenAI如何在备受期待的IPO之前扩大其广告业务。

First up on the show today, The Information has exclusive reporting about how OpenAI is looking to scale its advertising business ahead of its eagerly anticipated IPO.

我们的电子商务记者将做客节目，和大家分享她了解到的情况。

Our e commerce reporter will join the show to share with us what she knows.

我还会采访Modal Labs的首席执行官，聊聊他们最近完成融资后的战略规划。

I'm also speaking with the CEO of Modal Labs about the company's roadmap after its recent funding round.

这家公司估值达到了46.5亿美元，稍后我们会详细聊。

It was valued at $4,650,000,000 We'll get to that shortly.

我们还会重点关注一家软件股，它逆势而上，没有跟整个板块一起下跌——那就是Twilio。

We're also zeroing in on one software stock that has defied the broader declines in the sector, Twilio.

但我们的财经分析专栏作家指出，尽管近期受AI概念推动股价大涨，投资者可能忽略了更关键的长期问题。

But our financial analysis columnist says that despite the recent AI driven surge, eager investors might be missing the bigger picture.

今天这期节目内容很精彩，咱们马上开始。

It's going to be a fun show, so let's get right on into it.

OpenAI 正在加紧拓展广告业务，为即将到来的IPO做准备。

OpenAI is quickly trying to scale its advertising business as it looks ahead to its IPO.

我的同事安·盖汉今天发表了独家报道，详细介绍了这项工作的现状以及公司做出的一些调整。

My colleague Ann Gehan published exclusive reporting today on the current state of that effort and some changes that the company has been making.

我想请安来跟我们多聊聊她了解到的情况。

I want to bring on Ann to tell more with us about what she knows.

安，欢迎来到节目。

Ann, welcome to the show.

很高兴你能来。

It's great to have you here.

嗨，阿卡什。

Hey, Akash.

好的。

Okay.

所以，OpenAI 现在全力推广告业务。

So OpenAI is on this ads push.

我们知道，今年公司对广告业务寄予了很高的期望。

We know it has big expectations for what it's going to do for the business this year.

你深入调研了广告和电商领域，摸清了目前的进展。

You went out into your orbit of advertising and e commerce, and you found out how it's all going.

你发现了什么？

What did you find?

是的。

Yeah.

今年早些时候，OpenAI正式推出了广告业务。

Mean, so OpenAI launched advertising earlier this year.

这段时间发生了不少新动向。

There have been a lot of developments.

我每次跟消息源聊天，甚至从日到周，事情都在飞快变化。

Basically, every time I talk to my sources, even day to day, week to week, things are changing really quickly.

所以我一直在琢磨，OpenAI这一连串的调整和发布，到底意味着什么。

So I was trying to figure out what all these changes and launches from OpenAI all add up to.

所以你看，OpenAI 正在全力加快步伐，打造一个更成熟的广告业务。

So I mean, you see that OpenAI is really taking steps to build a more mature advertising business basically as quickly as it can.

他们明显在紧追 Meta 和 Google，这两家是数字广告领域的绝对巨头。

They're really trying to compete closely with Meta and Google, who are the two dominant players as far as digital ads go.

单看 OpenAI 推出新功能的速度，简直是在不遗余力地快速上线各种东西。

And just I think the pace that OpenAI is rolling out these new features, they're basically launching things as as quickly as they can.

这周他们推出了首个以转化效果为导向的广告测试，衡量标准是用户是否真的采取了行动——比如下单、注册账号、下载应用之类的。

And so this week, they're launching the first test of conversion oriented ads, which measure success and bill advertisers based on whether or not, users actually take an action, whether that's making a purchase, creating an account, downloading an app, things like that.

你可以明显看到，OpenAI 已经从几个月前那种只做品牌曝光的初期广告，转向了更具体、可衡量的成果，而这正是广告主一直强烈要求他们加上的功能。

You really see OpenAI moving from just general brand awareness with the first ads that they rolled out a couple of months ago to some more concrete, more measurable outcomes, which is something that advertisers have really been wanting and really been pushing OpenAI to add.

这挺有意思的，我记得你上次上节目时，我们聊过 OpenAI 怎么把自己定位成高端广告产品，当时他们的定价策略就很明显体现了这一点。

So this is interesting because I remember last time when you were on the show, and we've talked a lot about how OpenAI is positioning its product as a premium advertising product, and that certainly came through with the pricing, I believe, that it was offering advertisers.

所以现在看来，他们明显更看重用户实际的行为结果了。

And so it sounds like they are very much leaning then towards an action oriented result from the users essentially.

光有屏幕上的展示量可不够。

Mean, it's not enough just to have impressions on the screen.

他们是按实际效果收费的。

They're charging based on an outcome.

没错。

Right.

对，就是这样。

Yes, that's correct.

我觉得他们最初推出的那些广告，你说得对，价格很高，主要是为了提升品牌认知。

I think you see with the original ads that they launched, which you're right, were these very high priced, just more general awareness oriented ads.

我感觉， advertisers 对在 ChatGPT 上投放广告的兴趣其实很大。

I think, you know, there's obviously a lot of advertiser interest in running ads on ChatGPT.

但尽管需求旺盛、热度很高，广告主很快就说：好吧，这不错。

But I think despite all of that demand and all of that hype, I think very quickly advertisers said, okay, this is great.

你知道，ChatGPT 每周有九亿活跃用户，但我们也得看到实际成果。

You know, ChatGPT has 900,000,000, weekly active users, but, you know, we wanna see results.

我们得知道，花这么高的价钱，到底能换来什么。

We wanna know what we're getting for this premium price that we're spending.

所以我觉得，一些广告主已经开始施加压力了。

So I think that a little bit of that pressure is already kind of starting to build from some of the advertisers.

我想这就是为什么OpenAI先推出了点击类广告，最近又推出了更注重行动效果的广告系列。

I think that is why you see OpenAI moving to roll out, know, first these click oriented campaigns a couple of weeks ago, and now the the more action oriented campaigns as well.

你提到的另一个有趣点让我很感兴趣，就是OpenAI也在关注中小企业。

Another interesting part of your story that fascinated me is you said that OpenAI is also focusing on smaller businesses.

你举了干洗店的例子。

You talked about dry cleaners as one example.

是的，这确实挺有意思。

Yeah, it's interesting.

如果从整体来看，他们开发和推出的所有工具，其实都是为那些希望广告投入能带来明确回报的小企业设计的。

I think if you take a step back and look at the sum of all of the tools that they're building and that they're rolling out, they are really oriented more towards small businesses that you know want a very concrete return on their advertising spending.

他们追求的是可衡量的效果，而不是像Adobe、Target或福特这样的大品牌。

They want very measurable outcomes as opposed to a bigger name brand or a bigger name advertiser like Adobe or Target or Ford.

OpenAI最早合作的广告主就是这些大品牌。

Those were some of the very first advertisers that OpenAI worked with.

他们更在意的，还是那种笼统的品牌曝光。

They're more interested in, again, just kind of that general brand awareness.

就是一般性的品牌投放，而像一些本地的小企业或者很小的电商品牌就不一样了。

You know general kind of brand spending, whereas yeah, a smaller local business or a really small ecommerce brand.

他们希望自己的每一分钱广告投入都能看到明确的回报。

They want very measurable results from their spending.

所以我觉得，OpenAI 现在推出的这些工具，明显是更针对后一类小型企业设计的。

So I think that thinking and catering, I think the tools that OpenAI is launching now definitely cater more to that second group of smaller businesses.

你能再解释一下吗？

Help me understand.

如果我们把

I mean, if we take

所有

the

小企业的广告总投入和拿下像塔吉特这样的大客户相比，我本来以为你们会更看重大客户，毕竟小企业客户太分散了。

aggregate advertising spend of small businesses and we compare that to what it means to get a large advertising deal from Target or someone like that, I mean, I would have thought that you really would have wanted to focus on the large businesses and that the small businesses, I mean, it's a much more fragmented segment of customers.

对他们来说，专注于小型商户和个体经营者是一个战略重点吗？

Is this a strategic imperative for them to focus on the try cleaners and the mom and pop shops?

是的。

Yeah.

这不仅仅是那些小商户的问题。

It's not just the mom and pops of the world.

我觉得，小型电商品牌、中小型企业，这群广告主确实非常有潜力，竞争也特别激烈，谁都想拿下他们。

I think, smaller e commerce brands, small and medium sized businesses, that's definitely a super, super lucrative group of advertisers that there definitely is very fierce competition to own.

多年来，Meta 和 Google 一直主导着面向中小企业的广告市场。

Meta and Google have dominated the advertising business for small and medium sized advertisers for years.

他们每年光靠这些广告主的投放，就能赚几百亿美元。

They make hundreds of billions of dollars in ad revenue, from those those advertisers spending, you know, every year.

这些企业对 Meta 和 Google 的依赖程度非常高。

And those businesses are really dependent on Meta and Google.

很多人都在抱怨，这些广告主对 Meta 和 Google 太依赖了，所以任何新的广告渠道或展示方式，对他们来说都特别有吸引力。

Know there's a there's a lot of kind of complaints about how how beholden some of these advertisers are to Meta and Google and any kind of new ad or new place to show ads that they can incorporate is is really appealing.

这里有很大的机会。

There's there's a big opportunity there.

需求非常大。

There's a lot of demand.

如果你看看过去几年 TikTok 广告的增长，就能清楚看到，市场对非 Meta 和 Google 的广告形式有多大的需求。

I think if you look at TikTok's advertising growth over the past couple of years, that's a great example of just how much demand there is for new kinds of ads that aren't on Meta and Google.

我认为 OpenAI 正在努力抓住这部分需求。

I think OpenAI is really seeking to capitalize on some of that demand.

是的，这些广告主的预算确实比大公司小得多。

Yes, these advertisers definitely do have the smaller budgets than the targets of the world.

但我觉得当你仔细看的话

But I think when you you look

最终这会成为他们营销支出中非常重要的一部分。

It at it ends up being a huge, huge component of their marketing spend.

我的意思是，我记得你之前来过，跟我们聊过，这其实就是他们的营销预算。

I mean, is the you know, I remember you coming on and and talking to us about how I mean, this is their marketing spend.

你可以真正建立一个商业模式，基于这样的计算：如果我在Facebook广告或OpenAI上投入这么多，就能获得预期的投资回报。

You can actually build a business based on the calculation that if I spend this much in Facebook ads or in OpenAI, for example, I can expect this ROI.

在某些情况下，这简直成了一道可以算清楚的题。

It kind of becomes a very solvable equation in some cases.

对。

Right.

而且说到广告投放，你知道，渠道多得数不清，但这些小企业其实只在少数几个平台上投广告。

And I mean, and targets advertising, you know, all different, you know, dozens of different channels and these small businesses, they're really only spending on a handful of advertising platforms.

所以我觉得，只要OpenAI能争取到原本会花在Meta或Google身上的预算，哪怕是一点点，对他们来说都极其重要，尤其是现在。

So I think any dollars that OpenAI feels like they can win that would have been spent on Meta or Google, you know, those are hugely important to them, especially right now.

毕竟他们正朝着上市迈进。

You know, as they look towards an IPO.

能证明广告收入的增长，更别说还能从Meta和Google手里抢走预算，这意义重大。

Think being able to show that ad revenue, you know, let alone the fact that they're they're potentially taking dollars away from Meta and Google.

我觉得这对他们现在的业务来说至关重要。

I think that's really important to their business right now.

总的来说，安，从这些讨论中我理解，OpenAI 在2026年的广告收入目标是24亿美元。

So overall, Ann, you came away from these discussions, I think it was what, dollars 2,400,000,000 was the advertising revenue goal for OpenAI in 2026.

今年早些时候我们有报道说，他们当时的收入是一亿美元，也就是说还有很长的路要走。

I think we had reporting earlier this year that they were at $100,000,000 which meant they still had a ways to go.

我不确定现在有没有更新的数据，但你整体怎么看？

I'm not sure if we have figures to update that, but what is your overall conclusion here?

进展顺利吗？

Is it going well?

还是不太理想？

Is it going not so well?

OpenAI 确实设定了非常宏大的广告收入增长目标。

OpenAI definitely, they have very big goals for growing their ad revenue.

我们之前报道过，他们告诉投资者，预计到2030年广告收入将超过一千亿美元，占公司总收入的35%以上，可见广告业务对他们来说至关重要。

We've reported that they've told investors expect over 100,000,000,000 in advertising revenue by 2030, which would be more than 35% of the company's total revenue, which, you know, so advertising is clearly a very significant piece of their business.

但我觉得要实现这个目标，OpenAI 还得持续完善他们的广告工具。

But I think for that to happen, I think OpenAI will continue they'll have to continue to build out the advertising tools.

你看他们推出新功能的速度，明显是想尽快把这套广告体系铺开，哪怕产品还很早期，也先上线，然后马上开始向广告主收费。

I think you see just in the pace at which they're releasing new offerings, they definitely are trying to scale this up as quickly as possible and just kind of get, know, even if it's very early versions of something, get it released as quickly as possible and start charging advertisers for it basically.

不过跟广告主和帮他们做广告的代理机构聊下来，大家还是觉得这事儿更像一个实验。

But I think just in talking to advertisers and talking to agencies that are helping advertisers set up these ads, it's definitely they view it more as an experiment still.

目前还没人给ChatGPT广告单独划预算。

It's, No one has a dedicated budget for chat GPT ads yet.

很多广告主还在摸索，到底什么样的效果才管用。

I think a lot of advertisers are still just trying to figure out what works.

OpenAI甚至在不断推出和琢磨新的广告形式。

OpenAI has been even releasing and thinking about new kinds of ad formats.

变化太快了，这东西又太新，目前谁都没法拿出一套靠谱的打法。

I think there's so much change that you know, and it's so new that no one really has a reliable strategy for this yet.

所以广告主们还是把它当实验在做，现在还看不到广告预算从Meta或Google大规模转向ChatGPT的迹象。

So I think advertisers are still mainly treating this as an experiment and we're not seeing any kind of long term shifts of spending move from Meta or Google to ChatGPT just yet.

但OpenAI正在竭尽全力向广告主证明，这些广告真的有价值，能带来他们想要的效果。

But I think OpenAI is trying to do everything they can to prove to advertisers that these ads are valuable and that they're getting they're able to provide the results that advertisers are looking for.

很好。

Great.

非常感谢你今天来做客。

Well, I want to thank you for coming on.

这位是安·吉汉，我们《信息》杂志的电子商务记者。

That is Ann Gehan, our e commerce reporter here at The Information.

Modal Labs最近完成了3.55亿美元的融资，估值达到46.5亿美元。这笔融资的细节是我同事斯蒂芬妮·帕拉佐洛本月早些时候率先报道的。

Modal Labs raised $355,000,000 at a valuation of $4,650,000,000 This was a funding round that my colleague Stephanie Palazzolo first published details about earlier this month.

今天，我们请到了Modal Labs的首席执行官埃里克·伯恩哈德森，来和我们聊聊公司未来的发展规划。

Today, we have Modal Labs CEO Erik Bernhardsson on the show to share with us more about his company's roadmap.

埃里克，欢迎来到TITV。

Erik, welcome to TITV.

很高兴你能来。

It's great to have you here.

谢谢。

Thanks.

能来这里真的很棒。

It's great to be here.

你能用最简单的话告诉我们，Modal Labs 到底是做什么的吗？

Tell me, in the simplest possible terms, can you explain to us what it is that Modal Labs does?

哎呀。

Oh, boy.

我平时总试着跟我孩子解释，简单说，Modal 就是帮企业运行人工智能的。

I always try to explain to my kids, but Modal basically helps companies run AI.

你想啊，现在有这么多酷公司都在开发各种 AI 应用，比如做 AI 音乐的 Suno，还有支持实时编程的 Lovable，它们都离不开底层基础设施。

So if you think about all these cool companies out there building a lot of cool AI applications, whether that's Suno, which does AI generated music or Lovable, which powers a lot of live coding, five coding, they all need infrastructure.

我们就是为它们提供这层底层支持。

So we help them with the underlying infrastructure layer.

之所以需要我们，是因为传统的基础设施——

And part of why that's needed is a lot of traditional infrastructure.

如果你去用那些大云服务商，它们根本不是为 AI 设计的。

If you go to the clouds, the hyperscalers, etcetera, they're just not built for AI.

这些大型模型很难操作，想快速迭代、调配GPU资源，不管是训练、推理还是沙盒环境，都特别麻烦。

It's quite hard to work with these large models, iterate quickly, work with GPU capacity, whether that's training or inference or sandboxes, a bunch of other stuff.

所以我们搭建了这一层底层架构，让工程师能跑得更快，把那些底层的容量、路由、负载均衡之类的细节全都抽象掉，你根本不用操心。

So we built this underlying layer that helps engineers move much faster and helps you kind of abstract the way in a way you don't have to think about that underlying capacity, routing, load balancing, all that stuff.

你说云服务商，我想到那些巨头，还有那些我们叫新锐云的公司。

So when you say that the cloud providers and I'm thinking about the hyperscalers, and then I'm also thinking about the neo cloud companies who we call them neo clouds.

他们自称是专为AI打造的云平台。

They call themselves AI specialized clouds.

但他们主打的卖点就是：我们让你轻松在我们的云上运行应用。

But I mean, their their whole pitch is that we make it easy for you to run your applications on our cloud.

可你刚才的意思是，他们其实并不适合运行这类应用？

So I hear you saying that they're not actually well suited then to run the application.

你能说说为什么你觉得必须要有我们这一层吗？为什么应用不能直接上云？

Help me understand why you think it is that your layer needs to exist and that an application can't just go straight to a cloud.

是的，原因有几个。

Yeah, there's a couple of reasons.

顺便说一句，这些新云公司确实不错。

The NeoClouds are great, by the way.

我们大量使用新云服务。

We use a lot of NeoClouds.

我们也和很多超大规模云服务商合作。

We work with a lot of hyperscalers.

我们把他们当作合作伙伴，很多都是。

We consider them partners, a lot of them.

他们提供的是原始的GPU或者硬件资源。

What they give you is raw GPUs or raw hardware.

所以你还是得自己去配置。

So you have to still set it up.

通常你得一次性租用一年、三年，或者固定数量的算力。

And typically the way you get access is you rent for a year or three years or something like that, like kind of fixed number of capacity.

所以你依然得面对固定的算力配额。

So you still have to work with like sort of fixed capacity.

一个大问题是，当你在生产环境中运行大量应用时，需求其实波动很大。

A big problem is when you have a lot of these applications in production, your demand is actually very variable.

会有明显的高峰和低谷，这完全取决于你的用户行为。

You have this peaks and troughs and that depends on your users.

所以，很多人在部署基于GPU的生产系统时，面临的最大挑战就是这种需求的不稳定性。

So a big part of the challenge that a lot of people face when they deploy things into production running on GPUs is this variable demand.

因此，我们帮他们实现高度灵活的算力分配。

So we help them basically get very flexible capacity.

我们采用完全按使用量计费的模式。

We have a fully usage based model.

你只用为GPU实际运行的时间付费。

You only pay for the time the GPU is actually running.

你可以随时扩展到数千张GPU。

You can scale up to thousands of GPUs.

也可以缩减到零。

You can scale down to zero.

你只用为这段使用时间付费。

You only pay for that time.

我们之所以能提供这种弹性，是因为我们整合了来自全球众多云服务商和数百个地区的计算资源。

And we offer that flexibility because we pull together capacity from a lot of different neo clouds and hundreds of regions all over the world.

另一个重点是开发者的使用体验。

The other is the developer experience.

当你从这些云平台获取硬件后，还得部署应用、配置 Kubernetes 和 Docker，管理所有这些组件。

So when you get the hardware from these clouds, you still have to deploy the apps, set up maybe Kubernetes and Docker and manage all of that stuff.

这就需要专门的基础设施团队。

So you need an infrastructure team.

而使用 Modal，我们提供了好得多的开发体验。

Whereas with Modal, we offer a much better developer experience.

只需几行代码，就能部署应用，自动根据需求扩缩容，流量来了就扩容，没了就缩容。

A few lines of code, can just deploy something audit scales completely, asset demand comes in, scales up, scales down.

你可以快速修改和迭代。

You can edit very quickly.

研究团队可以直接把项目部署到生产环境。

You can have research team that can deploy things to production themselves.

你可以构建非常复杂的应用程序。

You can build very complex applications.

没错。

Right.

所以你现在的竞争对手包括Fireworks、Base10、Together AI，还有Modal Labs。

So you compete with the fireworks, the base 10s, the Together AI, and now Modal Labs.

我们管这种公司叫基础设施提供商。

I mean, this is what we call the infrastructure provider.

你现在所处的正是这个市场细分领域。

That's sort of the segment of the market that you're competing in right now.

大致上是这样。

Roughly, yeah.

我觉得它们和Modal的共同点是，我们都把自己看作是第二层云平台——至少我们是这么想的，至于它们怎么定位，我就不太清楚了。

I would say all of them, and Modal, what we have in common is we think of ourselves probably, or at least we think of ourselves, I don't know if they think of themselves, but as a kind of a second cloud layer.

你们有底层的云平台提供算力，而我们则在上面做一层管理，让工程师能更轻松、更快地使用这些资源。

So you have the underlying cloud with the capacity, and then we sort of manage the layer above, making that easy to access, making possible for engineers to do quickly.

我们把LM and Friends作为服务提供，但同时也做很多其他事情。

We offer LM and Friends as a service, but we also do a lot of other stuff.

所以我觉得，我们在定位上比这个领域的许多其他公司更通用一些。

So I think we're a little bit more general purpose than many other players in this space.

从长远来看，我想问的是，你们现在所处的这一层——所谓的第二层云，为什么必须是一个独立的层？

So long term, guess my question is, does the layer that you live in right now, this second cloud layer, why does it need to be an independent layer?

换句话说，长期来看，这只是一个我的推测。

In other words, long term, I mean, this is a hypothesis I have.

我觉得云厂商迟早会盯上你们、Together、Fireworks，甚至那些新兴的云公司，他们自己也在搭建软件平台。

I mean, I could kind of see the cloud companies looking to you or to together or to Fireworks, you know, even the neo cloud companies, they're all building out their own software platforms as well.

我觉得长远来看，你们这样的公司最终很可能会被某个云服务商收购。

I mean, I sort of see the long term game here being a company like you getting acquired by a cloud provider.

你难道不觉得，长期来看，这会是必然的结果吗？

Do you not see that to be the case in the long run?

我不觉得这是对的。

I don't think that's true.

我觉得从历史上看，比如Snowflake和Redshift，Redshift是AWS的产品。

I think I look at you know, historically, when you look at companies like Snowflake versus Redshift, you know, Redshift was AWS's product.

Snowflake则是一个独立的数据库。

Snowflake was a standalone database.

Snowflake表现得非常出色，尽管它运行在多个超大规模云平台上。

Snowflake is doing exceptionally well, even though they're running on a lot of hyperscalers.

我们认为这里的情况也类似。

And we think it's something similar here.

我们是在构建一个更上层的服务，而Neo Cloud和超大规模云厂商非常擅长提供全球范围的算力，价格也合理，但管理这些算力却很困难。

We were building a service higher up, Neo Clouds and the hyperscalers, they're very good and we work with a lot of them and they're very good at offering capacity all over the world at a decent price, but it's quite hard to manage that capacity.

他们真正擅长的是把这些能力封装成API，让使用变得简单。

But what they're very good at is putting that behind an API and making it possible to They're very good at running data centers, physical capacity.

但在我的看法里，他们往往不擅长提供优秀的开发者体验，也很难打造出能让开发者快速迭代的软件产品。

What they're often not that good at is, in my opinion, delivering good developer experience and delivering a software product that makes it possible to iterate quickly.

我真心觉得这是一种很棒的合作关系。

And I frankly think of it as a great partnership.

我们在上面构建这一层，同时在底层的超大规模云和Neo Cloud上投入了大量资金。

We're building that layer above and we're spending a lot of money on the underlying hyperscalers and neo clouds.

所以从这个角度看，我们并不觉得这是一种竞争关系。

And so in that sense, think we don't necessarily think of it as a competitive dynamic.

我们把它看作是一种合作。

We think of it as a partnership.

你们打算怎么用融到的钱？

What are you planning to do with the money that you raised?

我们计划快速扩张。

Well, we're planning to grow very quickly.

我们正在招聘很多人。

We are hiring a lot of people.

我们也很积极地收购，可能会买下一些公司，以便加速发展，同时

We're quite acquisitive, so we might go out and buy some companies in order to move faster, as well as

你们打算收购什么样的公司？

What do think you what are you looking to buy?

我们喜欢那些规模小但非常务实的团队，可能他们还没达到预期的商业成果。

We like small teams that are extremely tactical and maybe haven't had the commercial success they were expecting.

市面上有很多公司，工程师能力超强，正在做特别酷的东西。

There's a lot of companies out there that are exceptionally strong engineers building very cool things.

我在想，比如AI编程这种领域，现在开发者都往那儿涌。

I I mean, I'm thinking, you know, I I don't know, like like AI coding, for example, that's a place where developers have been flocking to.

也许你们正想加强这方面的兼容性。

Maybe that's an area that you're trying to build more compatibility with.

也可能是在后端，跟云服务商做更深的整合。

Maybe there's more on the back end integrating with the cloud company.

你们觉得最有可能在哪个领域出手收购？

Where do you think you could see yourself actually making an acquisition?

我觉得复杂AI领域有很多有意思的方向，比如强化学习，或者大模型推理优化这类东西。

I think there's a lot of interesting stuff in complex AI stuff, whether that's reinforcement learning or LM inference optimization, things like that.

我们还在关注另一类方向，比如文件系统、运行时环境，还有非常底层的东西，像 Linux 和基础设施的深度管理。

Another bucket of things we're looking at would be things like file systems and runtimes and very low level, you know, Linux and deep sort of management of infrastructure.

还有其他几个不同的方向。

There's a couple of different other buckets.

但我觉得，市场上有很多优秀的团队，我们非常期待能和他们合作。

But I think there's a lot of strong teams out there that we're very excited potentially working with.

另外，我对你们所关注的这类公司另一点很感兴趣，就是你们从云厂商那里租用计算资源。

And so the other part that I have been interested about with your segment of companies is you renting cloud capacity out from the cloud companies.

你们自己有拥有任何芯片吗？

Do you own any chips yourself at all?

没有。

No.

明白了。

Okay.

这是你们未来的计划吗？

Is that a plan?

你们有计划进入这个领域吗？

Do you plan to get into that business at all?

我不会排除任何可能性。

I wouldn't rule out anything.

我觉得所有选项都还在考虑范围内。

I think everything is on the table.

我们是一家纯软件公司，只做软件让我们能快速扩张、灵活推进，而且资本效率高，更像是一家平台型业务。

We are a software company, being pure software enables us to grow very quickly and move very fast and being very capital efficient, more like a marketplace business, frankly.

我们不会主动推出这类业务。

I wouldn't roll anything out.

未来我们也许会拥有芯片，但并不是我们特别想做的事。

It's possible we might own chips in the future, but it's not something we're particularly desiring to do.

明白了。

Right.

现在，你们的核心客户是开发者，我想知道，过去六个月里，他们在使用你们平台时，有没有出现什么让你感到意外的趋势？

Now, the developers that are your core customer base here, I wonder if you think about just the last six months and the trends with respect to how they are using your platform, things that have surprised you maybe.

我们发现AI编程现在成了大家最关注的话题。

We've seen AI coding be very top of mind.

令牌上限的问题时而浮出水面，时而又退到边缘，取决于大家当时在关注什么。

We've seen token maxing come in and out of the periphery in terms of what people are focused on.

过去六个月里，开发者对你们平台的期待有哪些变化，这让你感到意外吗？

What has surprised you about what developers have changed in their expectations over the last six months?

是的，对我们来说，一个关键变化就是沙箱环境的爆发式增长。

Yeah, I mean, I think a big factor for us has been the massive explosion in sandboxes.

Modal五年前刚起步时，是把这片领域当成一个通用平台来做的。

So Modal started out five years ago thinking about this space as a very general purpose.

我们想打造一个包含多种工具的平台。

Like we wanted to build a platform with a lot of different tools.

一开始，我们在推理任务上找到了非常明确的产品市场契合点。

And so we saw a very strong product market fit initially with Inference.

我们也在做训练相关的功能，也看到了大量批处理任务。

We've been working on training as well, seen quite a lot of batch jobs.

我们三年前推出了一款叫‘沙盒’的产品，核心想法是能安全地隔离运行第三方代码，通常是大模型生成的代码，因为我们发现很多人用大模型来生成代码。

We put out a product called Sandboxes about three years ago and the idea is you can basically run third party code in an isolated safe way, typically LM generated code, because we saw that a lot of people were using LMs to generate code.

今年夏天，这个功能突然爆发了。

That exploded this summer.

我们看到强化学习、即时编码和后台代理这些领域都出现了惊人的增长和广泛采用。

So we've seen tremendous growth and enormous traction with reinforcement learning, with vibe coding, with background agents.

我觉得这反映了我们所处的新现实：工程师们越来越多地运行后台代理，编写更多代码。

And I think it sort of reflects a lot of the new world we're in where engineers are running background agents and building a lot more code.

很多时候，他们背后用的正是Modal平台。

In many cases under the hood, it's actually powered by Modal when they're doing that.

有些公司训练代码模型时，也会用Modal来做强化学习和发布部署。

Or when some companies are training coding models, they're using Modal for reinforcement learning for their rollouts.

GitHub最近老出问题，我说的是那些宕机事件，这有没有影响到你们的业务或者任何集成？

How is the issues that GitHub has been having, and I'm talking about the outages here, has that affected anything to do with your business or any integrations at least?

确实，我们也被波及了。

I mean, we were affected by it.

我们花了不少时间处理合并时出现的问题。

We actually spent a lot of time dealing with the there was some merge issues.

我们的一位工程师发了条推文，结果火了。

One of our engineers tweeted about it and it went viral.

他具体说了什么？

What did he say?

他到底说了什么？

What did he say?

他说GitHub现在出了严重问题。

That GitHub is having some severe issues.

在我看来，GitHub这个产品很久以来都没怎么好好维护。

I I I'm GitHub is a product that hasn't been well maintained in my opinion for for quite some time.

说实话，我也不太确定。

Like, I I don't know.

我觉得这个领域更需要一些新的替代方案和更多投入。

Like, I I think that the the space would benefit from some more alternatives and some more development.

你认为他们哪些地方没做好，本来可以做得更好？

What haven't they done well that you think they could have done better?

我觉得 GitHub 拥有所有顶尖的工程师。

I think GitHub has all the engineers.

它有一群很棒的开发者社区，工具也相当不错，但总是这儿那儿冒出些小问题。

It has an amazing set of community and the tools are quite good, but there's like constant small issues here and there.

我也说不清楚。

I don't know.

我真的很喜欢这个产品。

I love the product a lot.

总的来说，它把我们带到了一个非常好的状态。

Overall, it's gotten us to a very good place.

我用它大概已经有十到十五年了，但那些小毛病一直在积累。

I've used it for probably ten, fifteen years at this point, but small bugs here and there that are sort of accumulating.

对于一个不使用 GitHub 的人来说，我得先简单解释一下。

And for someone who I don't use GitHub, so I'm asking for a bit of an explainer here.

这个产品就是没什么创新吗？

The product just hasn't innovated?

你是这个意思吗？

Is that the idea?

具体哪些地方进展太慢了？

What exactly has been slow about it?

他们本来可以怎么做出创新呢，我

How could they I have innovated, I

我觉得 GitHub 过去本可以做很多事。

think there's many things that GitHub could have done in the past.

他们其实很早就涉足 AI 编程了。

Mean, they were very early to AI coding.

他们很早就推出了 Copilot。

I mean, had the co pilot very early.

我觉得他们本可以在这个领域占据更主导的地位。

I think they could have actually been a more dominant player in that space.

我觉得这个产品跟现在的开发进度脱节了。

I I think the product is just not caught up with all the development.

我不太确定。

I don't know.

说实话，我还是挺喜欢 GitHub 的。

I mean I mean, again, like, I I do like GitHub.

我可不是想否定什么。

I don't wanna, you know No.

这事儿挺有意思的。

It's it's interesting.

你明白吗？

You know?

因为我们观察到。

Because we look.

我本来想问你另一件事：几周前我们独家报道了 OpenAI 可能在开发自己的 GitHub 替代品。我们一直对这个方向很感兴趣——就是有些公司觉得 GitHub 越来越不够用了，于是开始考虑自己打造类似的工具。尤其是那些 AI 原生的公司，有没有可能做出一个更适合处理 AI 编码需求的更好版本？

We the other story I was gonna ask you about is we published a scoop a couple weeks ago about OpenAI possibly developing their own alternative to GitHub, and this is something that we've been kind of interested in, is companies finding GitHub less effective in some ways, and then maybe looking to build their own similar tools, the extent to which an AI native company could develop a better version of GitHub that could very much handle AI coding volumes better.

我的意思是，你觉得这不可避免吗？

I mean, do you think that is inevitable?

如果OpenAI推出一款产品，你会考虑购买吗？

If OpenAI came out with a product, would you consider buying that?

我觉得社区和整个行业肯定需要更多选择。

I think the community or the world would benefit from more choices for sure.

我想提一点，这跟Modal有点关系：随着越来越多的智能体写代码，我们肯定需要更多专门为AI设计的工具。

One thing I'll say, just to call out, since it's somewhat related to Modal, I do think with more agents writing code, we will need more sort of AI suited tools.

我们需要能让智能体和工程师都更快推进的工具。

We will need tools that enable agents to move faster, as well as engineers of course too.

这方面还有很多可以补充的。

But so there are some to add.

我们对持续集成特别看好，比如大规模运行测试的能力，这方面GitHub一直占主导地位。

We're very bullish on, for instance, continuous integration, like the ability to scale up and run tests at very large scale, which is a place where GitHub has been a dominant player.

我觉得未来几年会涌现出很多很棒的工具，帮助工程师和智能体都大幅提升效率。

I think there's a lot of cool stuff you can build over the next few years that will enable both engineers and agents to move much faster.

希望其中一些功能能由 Modal 来驱动。

And hopefully, some of those things will be powered by Modal.

我们对这个领域非常看好。

We're very excited about this space.

很好。

Great.

埃里克，非常感谢你今天来做客。

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

这位是 Modal Labs 的创始人兼首席执行官埃里克·伯恩哈德森，欢迎来到 TITV。

That is Erik Bernhardsson, the founder and CEO of Modal Labs here on TITV.

谢谢。

Thanks.

能来这儿真的很开心。

It's great to be here.

今年推特的股价一路飙升，涨幅超过 30%，而其他大型软件股却普遍下跌了 30%。

Shares of Twilio have been on a tear this year, up more than 30% in contrast to the 30% declines that we've seen from other big software stocks.

这只股票大涨，主要是因为市场对它日益壮大的AI语音业务充满热情，但我的同事安妮塔·拉马斯瓦米在她最新的《真实价值》专栏中提出，投资者可能忽略了更大的图景。

The stock has surged largely because enthusiasm for its growing AI voice business, but my colleague Anita Ramaswamy argues in her latest true value column that investors might be missing the bigger picture.

现在安妮塔加入我们，和我们分享她的发现。

Anita joins me now to share with us what she has learned.

安妮塔，欢迎再次做客节目。

Anita, welcome back to the show.

很高兴你能来。

It's great to have you here.

你好，

Hi there,

阿卡什。

Akash.

我们先来回顾一下，Twilio到底做的是什么业务？

Remind us, what exactly does Twilio even do?

我们就从这儿说起。

Let's start there.

是啊，没错。

Yeah, right.

Twilio 是一个为企业提供客户沟通基础设施的平台，覆盖多种通信渠道。

So Twilio is a platform that provides the infrastructure for companies to communicate with their customers, and that's across a number of different channels.

他们主要以电话和短信服务著称。

Primarily they've been known for phone calls and text messages.

比如你收到的那些促销通知，或者提醒你购物车里有东西该结账的短信，很可能就是通过 Twilio 或类似平台发送的。

So when you get those push notifications, for example, or text messages from a company that tell you that they're having a sale or that something is in your cart and you should check out, chances are that that message would be going through Twilio or a platform like Twilio.

明白了。

Okay.

也就是说，我最讨厌的那些短信，全都是 Twilio 发的。

All the text messages that I hate receiving are coming from Twilio, in other words.

差不多吧。

More or less, yeah.

还有那些根本退不掉的短信，真是服了。

And the ones that I can never unsubscribe from, for God's sakes.

我是真按了无数次停止，哎，Twilio，你太厉害了。

I mean, I press stop so many times, and ah, Twilio, man, you're too good.

每次都能把我抓到。

You get me every time.

但我知道不是他们在发，

But I know that it's not them sending in.

我知道是那些商家在发，所以我们也没怪他们。

I know that it's the businesses, and so, you know, we're not throwing them under the bus.

不过你这周写了一篇文章分析Twilio的业务，他们的股价至少表现得不错。

But you wrote a column this week looking at Twilio's business, and their business, the share price at least has done very well.

你能不能说说，最近他们的股价为啥涨得这么猛？之后我们再聊聊，你觉得这涨得有没有道理。

Talk to me a little bit about why their share price has skyrocketed as of late, and then we'll get to your ultimate opinion on whether or not you think that is warranted.

当然可以。

Yeah, of course.

Twilio现在其实挺有意思的，因为它看起来像是个例外，在我们这个节目里一直聊的SaaS寒冬里，它居然没受影响。

So Twilio is kind of an interesting company to look at right now because it looks like an exception from this whole SaaS pocalypse that we've spoken lot about on this show and been hearing about.

他们是一家软件公司，但今年到目前为止，股价已经上涨了超过30%。

They are a software company, but they grew, or their share price appreciated more than 30% so far this year.

相比之下，像ServiceNow这样的大型软件公司，今年股价却下跌了大约30%。

That's in contrast if you look at like the ServiceNow or other big software companies of the world that are down around 30% year to date.

因此，对很多投资者来说，他们成了一个例外，因为他们拥有语音业务板块。

So they have really emerged as an exception to a lot of investors because they have this voice unit.

这个语音业务板块做的事情很多，但核心就是帮企业协调打给客户的电话。

And this voice unit, they do a lot of things, but essentially what they're doing is they're coordinating phone calls from businesses to customers.

通常这些电话都是客户支持或客户服务类的通话。

So oftentimes that will be like a customer support call or customer service calls.

而在语音业务里，他们推出了一套AI平台，让企业能用Twilio的工具搭建AI语音助手。

And what they're doing within that voice unit is they have an AI tool or an AI platform really, that allows companies to build AI voice agents using Twilio's stack.

现在大家都在谈智能体AI和语音助手的兴起，而Twilio至少在股价上受益于他们在这个领域推出了产品。

So when we're hearing a lot about this rise of agentic AI and the rise of, you know, voice agents, Twilio seems to be benefiting at least in terms of their share price from the fact that they have launched a product in that space.

所以你的意思是，比如在客户支持这方面——我们之前节目里聊过Decagon和Sierra这些公司，都是做自动化客服的。

So the idea being if, and maybe I'm thinking of customer support here, we've covered companies like Decagon and Sierra on the show, these automated customer service agents.

Twilio 的产品，据我理解，是提供给企业的，当你接到客服电话或者营销电话时，企业可以用它来搭建一个全自动的对话系统。

Twilio, if I understand it correctly, they have this offering that they sell to customers where if I'm talking to a customer service agent or if I get a marketing call or something, the business can actually use Twilio just to basically create an automated exchange with an automated agent type of thing.

对，没错。

Yeah, exactly.

有意思的是，Sierra 和 Twilio 有合作关系。

And it's interesting because Sierra has a partnership with Twilio.

他们互相推荐客户，但同时在某些领域又存在竞争。

They sort of cross sell some of their but then at the same time they are also competing in some spaces.

这个语音代理领域竞争非常激烈，但你可以把 Twilio 看作是端到端的基础设施提供商。

There's a lot of overlap and this voice agent space is really crowded, but you can think of Twilio as the end to end infrastructure provider.

如果你是一家公司，想用 OpenAI 或其他大模型，再加一个自定义界面，让系统用特定方式和客户对话，Twilio 都能提供所有必要的工具。

So if you're a company, you want to use, let's say, a model from OpenAI or a model from a different large language model provider and you want to put your custom user interface on top of that and have it speak in a certain way to your customer, all the toggles will be there on Twilio.

你可以自己搭建，然后通过 Twilio 平台来协调和管理整个流程。

You can sort of build it and then orchestrate and manage it from the Twilio platform.

你的分析提到，股价的上涨可能高估了这家公司。

Now, your columns suggest that perhaps the stock price surge may be a little bit overestimating the company.

你为什么这么说？

Why do you say that?

我这么说，阿卡什，是因为有几个原因。

I say that, Akash, because there are a number of reasons.

首先，AI到现在还没明显给Twilio的业绩带来巨大影响，至少还没看到。

The first is that AI is not really clearly making a super huge difference to Twilio's results, at least not yet.

就算只看语音收入，这也是大家特别兴奋的领域。

If you look at even just voice revenue, right, this is the thing that everyone's super excited about.

我跟投资者、分析师聊的时候，他们都把这当成最关键的增长动力，也是现在该买Twilio股票的理由——连那些看涨的投资者也是这么说的。

When I talk to investors, when I talk to analysts, they all described it as the number one key catalyst and the reason why you should own Twilio's stock today, even like the bullish investors I spoke to.

但这个语音AI业务，我们并不清楚它在语音板块里占多大比例，而语音板块整体在2025年只占他们收入的16%左右。

And that voice AI category, we don't know what percent of their voice segment it constitutes, but their voice segment overall is only around 16% of their revenue in 2025.

所以目前为止，语音助手似乎还没对他们的业绩产生太大影响。

So, so far it doesn't seem like the voice agents are really making a big difference to them.

另一个值得注意的点是增长方面。

The other point I think that's important to note is about growth.