Airbnb首席执行官布莱恩·切斯基谈人工智能战略与新首席技术官、微软与Anthropic的合作 | 2026年1月14日

谢谢邀请我。

Thank you for having me.

那么关于微软与Anthropic的合作，我们了解到了什么？

So what did we learn about Microsoft's work with Anthropic?

是的，我的意思是，在过去一年里，我们看到微软和Anthropic的关系变得更加紧密，尽管OpenAI仍然是微软最重要的人工智能合作伙伴和供应商。

Yeah, so I mean, over the past year, we've seen Microsoft and Anthropic get a lot closer, even though OpenAI remains Microsoft's most important AI partner and provider.

具体来说，这表现为微软成为Anthropic最大的客户之一，预计每年在Anthropic的模型上花费超过5亿美元。

And specifically, that's looked like, you know, Microsoft becoming one of anthropics biggest customers spending, you know, on track to spend more than $500,000,000 annually on anthropics models.

我们还了解到，微软现在激励其销售人员销售Azure上的Anthropic模型，就像激励他们销售OpenAI的模型一样。

And we've also learned that Microsoft is now incentivizing its salespeople to sell anthropics models on Azure, just as much as it incentivizes them to sell OpenAI's models.

除此之外，我认为Anthropic甚至开始更多地参与帮助微软构建使用其模型的产品。

Beyond that, I think Anthropix is also starting to even get more involved in helping Microsoft build products that use its models.

所以总的来说，这两家公司正变得越来越紧密。

So across the board, these companies are getting a lot closer.

现在，你报道称Anthropic预计每年与微软的业务额将达到5亿美元，这将使其成为Anthropic最大的客户之一。

Now, you reported that Anthropic is on pace to spend $500,000,000 annually with Microsoft, that would make it one of Anthropic's biggest customer.

我说对了吗？

Did I have that right?

是的，Anthropic在微软上花钱。

Yeah, Anthropic is spending the money on Microsoft.

不，是微软在Anthropic上花钱。

No, Microsoft is spending the money on Anthropic.

是的，微软每年在Anthropic的模型上花费5亿美元。

Yes, Microsoft is spending $500,000,000 annually on Anthropic's models.

同时，Anthropic确实在去年晚些时候承诺将在Azure云服务上花费数十亿美元。

At the same time, Anthropic did commit earlier, or late last year to spend billions of dollars on Azure cloud services.

所以资金是双向流动的。

So the money is going both ways.

但确实，微软在Anthropic上的支出对这家初创公司来说相当重要——Anthropic预计去年收入在70到90亿美元之间。

But yeah, I mean, the money that Microsoft is spending on anthropic is pretty significant for the startup anthropic projected that it was, you know, going to make in the realm of 7 to $9,000,000,000 in revenue last year.

因此，能从微软获得如此巨额的资金对其意义重大。

So this is pretty significant that it's getting this much from Microsoft.

现在，我想弄清楚的一个问题是，这些是什么类型的模型？

Now, one of the questions I wanted to understand here is what types of models is this?

因为微软显然拥有广泛的产品系列。

Because Microsoft obviously has an expansive suite of products.

这些是编码模型吗？

Are these coding models?

这些是大型语言模型吗？

Are these large language models?

他们从Anthropic购买的是什么？

What are they buying from Anthropic?

是的，截至每年5亿美元的数额，大部分资金都用于编码模型，微软用这些模型为其GitHub Copilot软件提供AI编程功能支持。

Yeah, so you know, as of the $500,000,000 annualized figure, most of that was coming from coding models, which Microsoft was using to power AI agent coding features in its GitHub Copilot software.

自那以后，微软的支出实际上可能大幅增加，因为他们也开始将Anthropic的模型整合到其他AI产品中，比如Office 365 Copilot和安全Copilot。

And since then, the the amount that Microsoft is spending has actually probably grown a lot because they've started to also add anthropic models to other AI products like Office three sixty five Copilot, and security Copilot.

所以目前来看，微软正在广泛地将不同类型的Anthropic模型应用于各种AI产品中。

So at this point, it's really, you know, across the board that Microsoft is using different Anthropic models for different types of AI products.

Anthropic有回报吗？

Is Anthropic reciprocating?

他们有在微软身上花钱吗？

Are they spending with Microsoft?

有的。

They are.

所以你知道，微软在九月份向Anthropic投资了50亿美元。

So they, you know, Microsoft also invested 5,000,000,000 in Anthropic in September.

就在那时，Anthropic表示，他们承诺将花费数十亿美元向微软租赁云服务器。

And around that time, anthropic said that they would commit to spend several billion dollars on renting cloud servers from Microsoft.

他们还表示，未来几年可能会再租用数吉瓦的云服务器空间，尽管具体这意味着什么很难确定。

And they also said, you know, they might contract several more gigawatts of cloud server space in the coming years, although it's hard to know exactly what that means.

但总的来说，这两家公司现在都已成为彼此的重要客户。

But yeah, basically, both of the companies have now become large customers of one another.

你认为在这段关系中，谁更有话语权？

And who do you think has the leverage in this relationship?

是的。

Yeah.

我的意思是，考虑到资金来回流动的种种方式，很难说谁从中获益更多。

I mean, with with all of the ways that money is changing hands back and forth, it's a little hard to say who's benefiting more.

但我觉得，目前Anthropic从中获益更多，因为它们真的希望拓展企业业务，而获得微软Azure云服务的客户资源，能极大地扩展它们潜在的客户群体。

But I I think my take is probably that Anthropic is getting the most out of this right now because they, you know, really want to grow their enterprise business and getting access to customers of Microsoft's Azure cloud service is a way to massively expand, you know, their potential universe of customers.

与此同时，微软已经可以免费使用OpenAI的模型。

At the same time, Microsoft does already have access to reuse OpenAI's models for free.

所以，也许微软对Anthropic的依赖，没有Anthropic对微软的依赖那么大。

So it maybe doesn't need Anthropic quite as much as Anthropic needs Microsoft.

但不管怎样，很明显，这两家公司都从这一合作关系中获得了巨大收益。

But you know, regardless, it is obvious that both companies are benefiting a lot from this partnership.

我想用一张图表来结束今天的讨论，我鼓励大家去阅读文章中的这张图。

And I want to end off with a chart that I would encourage people to look at in the story.

你基本上梳理了AI模型公司、云服务商，以及像英伟达和Meta这样的公司之间相互合作的所有方式，这是一张非常棒的矩阵图。

You basically mapped all the ways in which the AI model companies and the cloud providers and companies like Nvidia and Meta, the ways in which they are doing deals with each other, and it's a great matrix.

我从中学到的结论是，每个人都在和每个人合作。

My takeaway from that is everyone's working with everybody.

我想问你的是，这是否不可避免？

My question for you is, is that inevitable?

你认为这有风险吗？

Do you see it as a risk?

你怎么看这一点？

What do you make of that?

是的，我的意思是，在某种程度上，这确实是不可避免的，尤其是当你看到微软、谷歌和AWS这样的云服务商，他们都希望客户在自己的平台上使用各种模型，而不希望客户为了使用不同的模型而不得不切换到其他云服务商。

Yeah, I mean, I think in some ways, this was inevitable, especially when you look at cloud providers like Microsoft, Google, and AWS who want to have customers on their platforms using all models, and they don't want, you know, customers to have to use a different cloud provider to seek out different models.

而且我们还看到微软CEO萨提亚·纳德拉谈到，他认为模型最终会变得商品化，也就是说，未来所有不同的模型提供商都将能够提供大致相同的智能。

And we've also, you know, seen Microsoft CEO, Satya Nadella, talking about how he thinks models will become commoditized, essentially, that in the future, you know, all all different model providers will essentially be able to offer the same type of intelligence.

至于具体使用哪个模型作为某个工具的基础，将不再那么重要。

And it won't matter as much which model is underlying a specific tool.

这还没有发生，但我认为，从宏观上看，这正是云服务商正在推行的策略。

That hasn't happened yet, but I think that that is sort of the strategy that we're seeing from the cloud providers in a macro way.

很好。

Great.

好的，Aaron，感谢你参与节目。

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

这位是我们The Information的微软记者Aaron Holmes。

That is Aaron Holmes, our Microsoft reporter here at The Information.

好的。

Okay.

人工智能热潮催生了一类新型公司——推理服务提供商。

The AI boom has given rise to a new class of companies called inference providers.

随着人工智能应用依赖模型和算力，推理提供商在这个故事中变得越来越重要。

As AI applications rely on models and on compute, inference providers are becoming more and more important to the story.

这个领域的一条大鱼是Base10，该公司去年估值达到21.5亿美元。

One big fish in that arena is Base10, a company that was last valued at $2,150,000,000 last year.

现在有请Base10联合创始人兼首席执行官Tuhin Srivastava，来谈谈他公司的未来。

I want to bring on co founder and CEO of Base10, Tuhin Srivastava, to talk about the future of his business.

图欣，欢迎来到节目。

Tuhin, welcome to the show.

很高兴你能来。

It's great to have you here.

嗨，阿卡什。

Hi, Akash.

你怎么样？

How are you?

谢谢邀请我。

Thanks for having me.

我很好。

I'm doing great.

感谢你加入我们。

Thanks for joining us.

我期待接下来的对话。

I'm excited for the conversation here.

所以，我想明确一下。

So, look, I just want to make this clear.

你并不是一家云服务商，也不在开发模型。

So you are not a cloud provider, and you're not building the models.

你只是把两者连接起来。

You connect the two.

对吗？

Is that right?

没错。

That's right.

我们位于其上层。

We sit on top of it.

我们对世界的看法是，未来会有大量的模型。

We're like, Al, the way that we think about the world is that there's going be a lot of models in the world.

它们都需要在某个地方运行，并且需要一些专门的软件来实现生产环境中的运行。

They're all going to need to run somewhere, and they're gonna need some specialized software to be able to run-in production.

所以，你知道，我们就像是那种中间件，位于我们合作的所有这些不同云平台之间。

So, you know, we're kind of that middleware that sits between, you know, all these different clouds that we work with.

目前，我们与大约10个不同的云平台合作，以及那些构建模型的人们。

Today, we sit around 10 different clouds and the folks building models.

如今，我们为一些全球增长最快的公司提供支持，比如像Bridge、OpenEvent、Clay、Cursor这样的公司。

And today, we power some of the fastest growing companies in the world, think about like a bridge, OpenEvent, Clay, Cursor.

它们都使用Base 10运行生产工作负载。

They would all run production workloads with base 10.

是的，这，这正是我们的定位。

And yeah, that's, that's exactly where we stand.

而且

And

所以帮我理解一下。

so help me understand this.

我的意思是，如果我在构建一个AI应用，并且想要访问模型，一种方式是直接去找像OpenAI和Anthropic这样的公司，然后说，嘿，我就用你们的API来获取模型访问权限。

Mean, if I am building an AI application and I want access to models, I mean, one way to do it is I could just go straight to the companies like OpenAir and Anthropic, and I could say, Hey, I'm just going to use your API and get access to the models that way.

但另一种方式是通过你们。

But the other way to do it is you go through you guys.

第三种方式，我只是想弄清楚你们是如何与之区别的。

The third way and I'm just trying to understand how you differentiate yourself here.

第三种方式是我可以直接去找云服务商，他们也会给我提供这些模型的访问权限，对吧？

The third way is I could just go to the cloud providers and they give me access to all these models as well, right?

是的。

Yeah.

所以，我是这么理解的：我们实际上与一类特殊的模型合作，即开源模型和自定义模型。

So look, the way I think about it is we actually work with a special class of models, which are open source and custom models.

很多客户都在使用OpenAI和Anthropic的模型，但他们后来发现了一些特定的使用场景，或者想使用开源模型，就需要能够运行这些模型。

So a lot of a lot of a lot of customers are using open and Anthropic, and then they find some specialized use case or they wanna use an open source model, and they need to be able to run those models.

因此，他们有三种选择：可以去找云服务商，可以自己搭建，或者选择像BaseTent这样的推理服务商。

So they've got two three choices where can go to a cloud provider, you can build it yourself, or you can go to an inference provider like BaseTent.

我们努力提供的是运行这些模型所需的所有抽象层，以便在生产环境中大规模部署。

What we try to provide is all that abstraction you need to be able to run this in production at scale.

这就是性能。

That's performance.

所以这非常快，而且可扩展。

So that's very fast, that's scalable.

当你流量增加10倍、100倍、甚至1000倍时，它也能随之扩展。

So, you know, as you scale your traffic, you know, up 10 x, 100 x, thousand x, it kinda scales with you.

同时，我们为你提供出色的开发者体验，让你能够轻松实现这一点。

And then we give you a great developer experience so you can do that.

是的，云服务商确实也提供这些功能。

And, yes, the clouds do provide that.

它们也是我们的合作伙伴，但我们试图打造更专业、更用户友好、更快速且可扩展的产品。

They are our partners too, but we're trying to make something a bit more specialized, a bit more user friendly, and a bit more fast and scalable.

但这些全部都是开源模型。

But but these are all open source models.

所有这些基础设施提供商

All all these infrastructure providers

我将其描述为开源模型，或者经过微调的、基于强化学习的模型。

open source or they're fine tuned or RL models is how I describe it.

所以，这就像你从货架上直接取用的模型，或者是根据你的使用场景进行定制的模型。

So it's like, you know, models that you are taking off the shelf or models you are customizing to your use case.

好的。

Okay.

那么，Base Ten 的商业模式是什么？

And then what is the business model for Base Ten?

你们如何盈利？

How do you make money?

是的。

Yeah.

我们基本上是根据使用量来盈利的。

So we we basically make money based on usage.

也就是说，你使用这些模型越多，我们的收入就越多，而且理想情况下，你的收益也会增加。

So, like, the more you use these models, the more revenue we make, and, you know, ideally, the more our customers make too.

所以我们与客户的盈利方式非常、非常一致。

And so we're very, very aligned with how our customers make money.

你们上个月收购了这家名为Parsed的公司。

You bought this company, Parsed, last month.

我不知道

I don't know

是Parsed还是Parsed，但那是你们收购的一家公司。

if it's Parsed or Parsed, but it was a company you bought.

我们在The Information上报道了这件事。

We reported it in the information.

你们为什么决定进行那次收购？

Why did you decide to do that acquisition?

是的。

Yeah.

所以你看，我们认为未来不仅仅只有闭源模型。

So look, what we think is that the future is not just gonna be closed models.

未来将是开源模型和定制模型。

It's gonna be open source models and custom models.

而PaaS是一家强化学习公司，它基本上允许企业带入自己的数据，采用一些基础模型，并为特定客户用例开发出更好、更快的定制模型。

And then PaaS is a reinforcement learning company that basically allows companies to bring their own data, take some base models, and come up with a custom model that is better, faster, better and faster for their particular use case for those customers.

所以我们认为这非常符合未来的趋势，也是这些模型的发展方向。

So we think that is very much the future or where these models are going.

而且，你知道，过去的团队，他们是一支来自英国和旧金山的小团队。

And, you know, the past team, you know, they're a small team based out of The UK and San Francisco.

他们曾是我们的客户，我们在帮助客户将推理工作负载迁移到Base 10方面看到了巨大的协同效应。

You know, they were a customer of ours, and we saw a massive synergy in terms of enabling customers to be able to bring inference workloads to base 10.

当你思考当前AI领域需要完成的工作时，让我印象深刻的是，我们都是通过ChatGPT和这些聊天界面接触到AI的。

As you think about the work that needs to be done in AI now, the thing that stands out to me is that we all got exposed to AI through ChatGPT and through these chat interfaces.

接着AI生成图像出现了，然后演变成音频，再发展到视频。

Then AI generated images came along that turned into audio, which turned into video.

现在你们拥有的模型最终也将为机器人和物理世界提供动力。

Now you have the models that eventually will have to power robots and the physical world as well.

但我想请你帮我理解的是，当我们谈到推理时，这些应用中的推理是否都相同？

But the thing that I am hoping you can help me understand here is when we think about inference, is inference the same across all of those applications?

还是说，比如视频推理需要更多的计算资源？

Or might it be that video, for example, requires more inference?

这是一种不同的推理类型吗？

Is it a different type of inference?

解释一下，是的。

Explain Yeah.

那到

That to

所以这在很大程度上取决于模型的架构以及它们需要运行的位置。

So it's all somewhat dependent on the architectures of the model and where they need to run.

因此，所有模型的架构都略有不同。

And so all the architectures of the model are slightly different.

例如，视频模型主要基于扩散变换器构建，它们比当今的变换器需要更多的计算资源。

So for example, video models are predominantly built on diffusion transformers, which are a bit more compute intensive than transformers today.

因此，它们需要不同的框架和不同的运行时环境。

So they need different frameworks and different runtimes.

当你谈到机器人时，很多机器人需要在边缘端运行。

When you talk about robots, a lot of those need to run on edge.

这意味着，比如，直接在机器人本身上运行。

So what that means is, like, you know, running on the robot itself.

因此，这与云推理服务截然不同。

And so, again, that's very different than a cloud inference service.

我们非常专注于 Club One 的云推理，并抽象掉任何模态的差异。

You know, we are very, very focused on Club One cloud inference and to abstract away from any modality.

因此，我们处理音频工作负载。

So, you know, we work with audio workloads.

我们处理图像工作负载、视频工作负载、语言工作负载，现在还包括视觉语言模型。

We work with image workloads, video workloads, language workloads, VLMs now.

因此，我们对这些模态持中立态度，但它们都需要略有不同的软件和运行时环境，以确保为最终客户高效、可靠、高性能地运行。

And so, you know, we're kind of agnostic to that, but they all require slightly different, I'd say, software and runtimes to make them work efficiently, reliably, performantly for the end customer.

但我很好奇，从这个领域来看，两三年后，你是否认为会出现某种细分，比如Base 10成为某种特定类型应用或特定内容生成的推理提供商？

But do you I'm curious, looking at this space maybe two, three years from now, do you see there being sort of a segmentation in maybe Base 10 is the inference provider for a specific type of application or a specific type of content that needs to be generated?

你知道的？

You know?

讨厌这个。

Hate that.

这会是最终的发展方式吗？

Might that be the way it plays out?

有可能。

It could be.

听我说，我真希望事情会这样发展。

Look, I'd love for that to be the way it plays out.

这么说其实挺自我服务的。

Know, that's very self serving of me.

我认为推理是一个广泛的类别。

I think inference is a broad category.

推理是一个非常大的市场。

Inference is a very large market.

你知道吗？

You know?

正如你所听到的，你现在听过多少位CEO说推理是最大的市场？

As you you've heard, how many how many CEOs have you now heard say Inference is the largest market?

所以，你知道，我确实认为

And so, you know, I do think there

我听过多少位CEO说过？

How many CEOs have I heard?

我不知道。

I don't know.

我的意思是，我认为Jensen说过。

I mean, I I think Jensen said it.

我认为Satya也说过。

Think Satya said said it.

好的。

Okay.

我想萨提亚也说过。

I think Satya said it too.

但关键是，你知道，这是一个足够大的市场，会有专门的推理用例。

So but the point being that, you know, it is a big enough market that there will be specialized use cases of inference.

我们希望为尽可能多的用例提供支持。

We want to power as many of them as possible.

目前，我们非常专注于实时生产推理，有点抽象，无论是什么模式。

Today, we are very, very focused on real time production inference, kind of abstract, regardless of modality.

这可能会随着时间改变，但你知道，我们确实认为这是一个巨大的市场，巨大的机会。

That might change over time, but, you know, we we do think it's a massive market, massive opportunity.

而且，会有专门的参与者，但通用公共推理平台也有很大的空间。

And, there will be specialized players, but there is a a big place for a a general public platform for inference.

你们自己购买芯片吗？

Do you guys buy your own chips?

不。

No.

我们不买。

We don't.

所以我们与10家云服务商在底层合作。

So we work with 10 clouds kind of underneath the hood.

因此，我认为我们创立Base Tenet的核心洞见之一是：推理是一个多云问题。

So and, you know, we we we provide one of the core insights here, I think, in starting base Tenet that Inference is a multi cloud problem.

你需要应对云服务中断的弹性能力。

You need resiliency for outages on clouds.

所以，我们的客户能够从这种弹性中受益，因为我们基本上建立在10个不同的云平台之上？

And so, you know, our customers are able to from that resiliency because we kind of sit on top of 10 different clouds?

我之所以这么问，是因为我想问你的问题之一是关于这些新推出的芯片。

The reason I'm asking is because one of the questions I wanted to ask you is about these newer chips that are coming out.

就在本月早些时候，我们看到了英伟达公布了其Rubin系列芯片的更多细节。

We just saw earlier this month, NVIDIA unveiled more details about its Rubin family of chips.

我一直在考虑客户在从一个芯片转换到另一个芯片时可能面临的切换成本，更不用说在同一家公司内部了。

And I've been thinking about the switching costs that customers may undergo when going from one chip to another, let alone within one company.

我的意思是，如果你想从英伟达切换到另一个芯片，那又是另一项切换成本。

I mean, if you want to switch from NVIDIA to another chip, that's another switching cost.

我的意思是，升级芯片、更换芯片有多困难？

I mean, hard is it to upgrade a chip, to switch out a chip?

客户在这方面是否遇到任何困难，

Are customers having any difficulty with it,

从什么

from what

你听到的？

you're hearing?

嗯，理想情况下，推理服务提供商可以帮助你处理这类问题。

Well, ideally, inference provider can help you with things like that.

你知道，什么

You know, what

我们我

we I

比如，当你从一个NVIDIA系列切换到另一个NVIDIA系列，比如从Hopper到Blackwell，现在再切换到Rubin，我们做了大量工作，以确保这些模型能在这些新芯片上良好运行。

and, like, you know, when you go from one NVIDIA family to another NVIDIA family, so from Hopper to Blackwell, say now you go to Rubin, you know, we are doing a lot of that legwork to make sure these models can run very well on those new chips.

是的，这确实具有挑战性，而且确实会带来一些问题。

And and so, yes, it is challenging, and, yes, there are issues that come up with that.

但这也正是专门的软件层——Imprint存在的原因。

But also, you know, that is why the specialized software layer for imprint exists.

在切换过程中，有哪些具体的挑战呢？

What what are what are some of the challenges there switching?

你看，它们只是不同，各有不同的架构，是不同的硬件。

Look, they just have different, you know, they have their the different heart it's different hardware.

而且有些模型非常敏感。

And like some of these models are very, very finicky.

要让它们以正确的吞吐量运行，并针对该芯片类型进行优化，充分利用硬件性能，这很有挑战性。

And getting them running with the right throughput and optimized for that chip type, so you're taking advantage of that hardware is challenging.

我们已经看到，每一代芯片发布后，通常需要一到两个月，有时甚至六个月，才能将芯片针对特定工作负载调优到位。

And we've seen that with every kind of family where a chip comes out and it takes maybe, you know, a month, two months, six months at times to basically get that chip dialed in for a workload.

对。

Right.

图欣，感谢你参加我们的节目。

Well, Tuhin, I want to thank you for joining us on the show.

这是一次非常精彩的对话。

It was a great conversation.

这位是塔欣·斯里瓦斯塔瓦，Base Ten的联合创始人兼首席执行官，欢迎来到TI TV。

That is Tuhin Srivastava, the co founder and CEO of Base Ten here on TI TV.

好的。

Okay.

一家新的数字银行正在吸引金融界的关注，因为它在去年获得华盛顿监管机构的有条件批准后，即将开业。

A new digital bank is drawing the the focus of the financial world as it prepares to open after getting conditional approval from regulators in Washington last year.

现在，让我们邀请我们的财经编辑肯·布朗，为我们介绍这则故事，出现在我们每周的财经专栏中。

Joining me now with the story in our weekly finance column is our finance editor, Ken Brown.

肯，欢迎回到节目。

Ken, welcome back to the show.

周三快乐。

Happy Wednesday.

周三快乐，你好阿卡什。

Happy Wednesday, hi Akash.

我应该每个周三的环节都这样开场。

I should open every Wednesday segment with that.

周三快乐，因为我们只在周三见到你。

Happy Wednesday, because we only see you on Wednesdays.

那是个特别的日子。

That's a special day.

你写的那家新银行是什么，肯？

What is the new bank that you wrote about, Ken?

这家新银行叫做埃布拉多银行。

The new bank is called Ebrador Bank.

我让你说出了这个名字。

I made you say the name.

这是一家由一群知名科技人士创办的新科技银行。

It's a new tech bank started by a bunch of big name tech folks.

它在去年年底获得了联邦政府的批准，筹集了大量资金，正准备推出。

It was approved by the federal government late last year and it raised a ton of money and it is getting ready to launch.

为什么世界需要一家新银行？

And why does the world need a new bank?

谁在推出这家银行？

Who's launching this bank?

这就是问题所在。

There's the question.

所以它正在被推出。

So it's being launched.

帕尔默·拉基是主要支持者之一。

Palmer Lucky is a big backer.

乔·朗斯代尔是支持者之一。

Joe Lonsdale's a backer.

美国不需要更多银行。

America does not need more banks.

银行业务，特别是中小型银行业务，是一项艰难的生意。

The banking business, especially the mid size and small banking business is a tough business.

增长并不多。

There's not been a lot of growth.

过去十年左右，这个国家的银行数量减少了一半，主要是通过合并而非倒闭。

The number of banks in this country has fallen by half in the last decade or so, mostly through mergers, not failures.

它们就这样消失了。

They just disappear.

所以这是个问题。

So it is a question.

他们会辩称，经济中有服务不足的部分，比如科技领域，特别是加密货币领域服务不足，需要银行服务。

They would argue that there are underserved parts of the economy in tech, for example, and in particular crypto is underserved and needs banking.

这些人将带着全新的银行进入市场，使用的是各种先进的技术，而不是陈旧的技术，他们会满足某种需求，从而赚取利润。

These folks are gonna come in with a new bank that has all kinds of zippy technology, not legacy technology, and they are going to fill a need and that's how they're going to make money.

所以，我只是想弄清楚，当你提到先进的技术时，广义上说，比如一家更支持加密货币、更注重科技、更适合人工智能和国防公司的银行，它们可能会提供哪些类型的服务或产品，或者有哪些工具能帮助这些公司，而传统银行却做不到？

So, I'm just trying to understand, when you say zippy technology, and broadly speaking, I mean a bank that is more crypto friendly, more tech forward, better suited for AI and defense companies, what are the types of services or products that they might offer or even tools that might help those companies that a traditional bank might not?

嗯，这很难说，对吧？

Well, so it's hard to say, right?

因为传统银行其实能做很多事情。

Because traditional banks can do a lot of things.

我的意思是，他们常说，即使传统银行涉足加密货币，它们用的也是老旧的系统。

I mean, one of the things they say is traditional banks, even when they do crypto, for example, they have these old systems.

所以它们试图把这种新技术或新货币体系强行嫁接到旧技术上，这就带来了问题。

So they're trying to paste this new technology or the new money system onto the old technology and that is causing problems.

这限制了它们能服务的客户数量。

It's limiting the amount of customers they can take.

我们已经见过这样的例子。

We've seen examples of that.

所以，如果我们能做得更好呢。

So, say we can do this better.

这是

It's

很有趣。

interesting.

我有点怀疑，因为很多银行由于这不是个太好的生意，都在尝试新东西并进行投资。

I'm a little skeptical because a lot of banks Because it's not such a great business, they're all trying new stuff and they're all investing.

因此，它们将能够处理其中一些事务，而且越来越多的银行正朝这个方向发展。

And so they are going to be able to handle some of this stuff and more and more banks are moving in this direction.

那么还会需要吗？

So will there be a need?

这还不太清楚。

It's not really clear.

就公司而言，过去十年科技发生了巨大变化。

In terms of companies, tech has changed a lot in the last decade.

我们也很清楚，但过去主要是软件公司，它们不需要太多现金。

We know that as well, but it used to be mostly software companies and they didn't need a lot of cash.

它们是相对资本密集度较低的业务。

They were a relatively capital light business.

现在，由于国防科技、人工智能和其他领域，人们需要资金。

Now between defense tech and AI and other stuff, people need capital.

当你需要资金时，就需要贷款之类的东西。

And when you need capital, you need loans, things like that.

你需要更多的银行服务。

You need more banking.

所以，银行业不再只是用于发工资之类的事情了。

So banking is not just for your payroll and stuff anymore.

这确实是一种需求。

It really is a need.

所以，如果这些公司能够进入市场，为传统银行可能持谨慎态度的行业提供贷款，那也可能是一种优势。

So if these guys can come in and make loans that serve some of these industries where the traditional banks may be wary, then that could be a benefit too.

这些贷款是否必然风险更高？

Would these loans necessarily be riskier loans?

这家银行实际上在努力成为一家安全的银行。

So this bank is really trying to be a safe bank.

如果你还记得，硅谷银行在2023年，也就是三年前倒闭了，这真的让硅谷的人们感到恐慌。

So if you remember, Silicon Valley Bank failed in 2023, three years ago, and it really freaks people out in Silicon Valley.

所以他们希望非常、非常安全。

And so they wanna be very, very safe.

因此，他们不会相对于其存款发放巨额贷款。

So they're not gonna make huge loans relative to the deposits they have.

他们将持有非常安全的资产，但他们可以做些事情。

They're gonna have very safe assets, but they can do stuff.

如果他们比某些其他银行更了解某个行业，那么他们或许可以发放一笔安全的贷款，但在别人看来却是有风险的。

If they understand an industry better than some other bank somewhere, then maybe they can make a loan that's safe, but would be considered risky by somebody else.

我的意思是，银行业就是这样，你根本无法预知。

I mean, with banking, you just don't know.

银行不会在一两年内倒闭。

Banks don't fail in a year or two.

银行会在五年后贷款无法收回时倒闭。

Banks fail five years when the loans don't get paid back.

所以你根本无法预知。

So you just don't know.

银行业内常说的一句话是：世界上最可怕的事情就是一家银行增长过快。

I mean, one of the lines in the banking industry always is, the scariest thing in the world is a very fast growing bank.

这意味着他们正在大量放贷。

It means they're making a lot of loans.

因此，三年后你才能看到他们有多成功。

And so, three years from now, you see how successful they were.

你认为这家银行可能面临哪些挑战，无论是从监管角度还是业务角度？

What do you see as some of the challenges that could face this bank, either from a regulatory perspective, from a business perspective?

你一开始说过你持怀疑态度。

You said at the outset that you were skeptical.

请详细说明一下。

Walk me through that a bit.

嗯，银行急于放贷。

Well, banks are eager to lend.

所以这是一方面。

And so that is one thing.

另一方面是，过去十年里私人债务领域已经发展成了一个庞大的世界。

The other thing is there's a whole huge world that's grown up in the past decade of private debt.

而且有大量资金在寻找投资去处。

And there is a ton of money out there looking for places to go.

所以银行的核心业务是贷款，这是一个竞争相当激烈的领域。

So the bread and butter business of banks is lending and it's a pretty competitive space.

另一个风险是，当你处理加密货币时，我的意思是，需要连接到美国金融体系，特别是稳定币，去年夏天稳定币法案通过后获得了巨大推动。

The other risk is, yeah, when you're handling crypto, I mean, needs to be connected to The US financial system, especially stable coins, which have gotten a huge boost following the stable coin law last summer.

问题是稳定币会让你面临各种关于制裁和了解客户规则的问题。

The problem is stable coins open you up to all kinds of issues around sanctions and around know your customer rules.

银行业对这些东西非常担忧。

The banking industry is very nervous about that stuff.

因此，他们可能会承担其他银行不愿承担的风险，仅仅是因为监管风险。

And so they may be taking on a risk that other banks won't take just because of regulatory risk.

从监管角度来看，我们目前处于非常友好的环境，尤其是加密货币领域。

And then we're in a very friendly environment from a regulatory perspective, especially crypto.

这种情况可能不会持续下去。

That may not stay true.

情况可能会改变。

That may change.

如果出现问题，这一届政府可能会改变政策。

It may change in this administration if there are problems.

下一届政府也可能改变政策。

It may change in the next administration.

因此，这就是他们所承担的风险。

And so, that's a risk that they take.

三年后，他们可能会被视为异类，并在监管问题上受到严厉处罚。

Three years from now, they may be seen as an outlier and really get nailed on the regulatory stuff.

是的。

Yeah.

嗯，我的意思是，这就是布莱恩·切斯基的观点，他今天虽然没有重申这一点，但在我为今天的对话做准备，观看他过去的一些采访时，他提出了这个观点。

Well, mean, and this is the point that Brian Chesky, he didn't echo the point today, but as I was watching some of his past interviews in preparation for the conversation today, he made this point.

他说，看吧，技术人员会随着政府更迭而改变方向。

He said, Look, tech people go one way, another way, depending on the administration.

情况可能在两年内发生变化，到时候怎么办？

Stuff could change in two years, and then what?

你打算怎么做？

What are you going to do?

你可以改变路线。

You can reverse course.

我认为这肯定会很有意思。

I think it's definitely going be interesting to play out.

肯，这次通话很棒，我想感谢你前来与我们讨论这个话题。

Ken, it was a great call, and I want to thank you for coming on and discussing it with us.

这位是肯·布朗，我们The Information的财经编辑。

That is Ken Brown, our finance editor here at The Information.

好的。

Okay.

在科技界，加利福尼亚仍然是中心地带。

In the tech world, California is still the center of gravity.

它既是硅谷的所在地，也是行业巨头的家园，但也一直被评为生活成本和税收最高的州之一。

It is home to Silicon Valley and the industry's biggest players, but it also consistently ranks as one of the states with the highest costs of living and taxes.

随着一项针对亿万富翁的新财富税提案，紧张局势正在加剧。

Tensions are rising with a new proposed wealth tax on billionaires.

该提案将对净资产超过十亿美元的加州居民征收一次性5%的税。

It would place a one time 5% tax on Californians worth more than a billion dollars.

虽然仍处于早期阶段，但该提案已引起广泛关注，尤其是一些人认为它会适得其反，将财富驱离加州并阻碍新创业者的出现。

It is still in its early stages, but the proposal has attracted a lot of attention, especially from people who say that it will backfire by driving wealth out of California and discouraging new founders.

为了了解我们的加州读者对这一问题的看法，我们进行了一项调查，询问他们是否考虑离开，以及如果离开，打算去往哪里。

To get a pulse on where our readers in California stand on this issue, we ran a survey to ask them if they were considering leaving, and if so, to where.

现在加入我讨论调查结果的是我们《信息》杂志的高级数据记者谢恩·伯克。

Joining me now to discuss the results is Shane Burke, our senior data journalist here at The Information.

谢恩，欢迎来到节目。

Shane, welcome to the show.

很高兴你能来。

It's great to have you here.

嘿，

Hey,

科什。

Kosh.

最近怎么样？

How's it going?

这是你第一次上节目，我们会邀请你更多次回来的。

First time on the show, and we're going to bring you back more and more.

我决定了。

I've decided.

是的，我知道。

Yeah, I know.

这就像抓到一只稀有宝可梦一样。

It's like catching a rare Pokemon or something.

你比任何宝可梦都酷多了，虽然我小时候是宝可梦迷，所以有些宝可梦确实很酷。

Well, you're much cooler than any of the Pokemons, although I was a Pokemon fan growing up, so some of them are pretty cool.

我们先放一放。

Let's put it aside.

好吧。

All right.

我们来谈谈这些数据。

Let's talk about the data here.

那么，我们最初为什么决定发布这个故事呢？

So why did we decide to run the story in the first place?

嗯。

Yeah.

正如你所说，我们只是想了解人们对加利福尼亚的感觉，特别是我们的加利福尼亚读者，以及他们是否有搬家的意愿。

So as you said, we just wanted to get a pulse on how people were feeling about California, our readers specifically in California, and if they were interested in moving.

我们还想知道他们对亿万富翁税的看法。

We also wanted to know their thoughts on that billionaire's tax.

我们提问的方式是，先问他们是否有搬家的意愿，然后再问亿万富翁税的问题，这样就不会让他们先想到税收而影响回答。

We asked it in a way so that we would have their answer to whether they would they were interested in moving or not before asking about the billionaire's tax just so we wouldn't leave them with that and get them thinking about taxes first.

但没错，我们只是想看看他们在这样一个特别的时期有何感受。

But, yeah, we just wanted to see how they're feeling in this really interesting time.

那么结果如何？

And so what were the results?

我们发现了什么？

What what did we find?

嗯。

Yeah.

所以大约三分之一的人有兴趣离开加利福尼亚，但这真的是他们会搬走吗？

So about a third about a third of them would be interested in leaving in California, but that's really would they move?

所以并不一定意味着他们会真的搬走。

So it's not necessarily that they will move.

在这些人中，很明显他们正在关注那些免税目的地，比如没有所得税和资本利得税的地方。

Of those people, it's pretty clear that they're eyeing, like, tax free destinations, like income tax and capital gains tax free destinations.

德克萨斯州排名第一，奥斯汀被提到几次，佛罗里达州排在第二位。

Texas ranks first with Austin mentioned a few times, as well as Florida that's second.

迈阿密也被提及。

Miami gets mentioned.

我想还有内华达州。

And I think Nevada.

是的。

Yeah.

内华达州排在第三位，也属于那一类。

Nevada's third, which also falls in that bucket.

嗯。

Yeah.

有多少人回复了这项调查？

And how many how many people responded to the survey?

共有374名受访者。

It was 374 respondents.

这些受访者都是我们的读者，因此需要指出，他们是一个特定的人群样本。

And these respondents are our readers, so it's worth mentioning that they're a specific sample of people.

他们是愿意付费阅读深度科技报道的人。

They're people who pay to read in-depth coverage on tech.

因此，这反映的是科技专业人士、科技高管这类群体的倾向。

So it gets at, like, a direction of tech professionals, tech executives, you know, that sort of group.

你能跟我讲讲，当你分析这些数据时，得出了哪些结论吗？

And tell me a little bit about what your conclusions were as you looked at this data.

嗯。

Yeah.

我之前没提到，但大约有一半的人不支持亿万富翁税。

So I didn't mention, but about half of them don't support the billionaires tax.

大约三分之一的人支持。

About a third of them do.

其余的人还不确定。

The rest aren't sure.

而且，是的，这里有相关的图表。

And, yeah, there's the the chart of that.

所以，关于这些数据的结论，我想说，首先，很明显我们的读者对加利福尼亚非常忠诚，这很合理，尤其是考虑到他们的群体特征。

But I so my conclusions on the data, I would say, first, it's clear that they're very loyal to California, our readers, which makes sense, especially given, like, what, like, group they are.

但那些考虑搬离的人，主要是出于税务原因，或者更低的生活成本。

But the ones who are considering going somewhere else, it would be for tax reasons, maybe a lower cost of living.

但这些地方的生活节奏也都更悠闲。

But these places all have, like, a chiller pace of life too.

所以，这是所有这些因素的综合。

So, you know, it's a mix of all of that.

尽管他们并不打算离开，但他们并不一定支持这项亿万富翁税。

And even though they're not planning on leaving, they don't necessarily support this billionaire's tax.

大约一半的人不支持，但也有三分之一的人支持。

It's about, you know, half of them don't, but also a third of them do support it.

所以这个数字也不小。

So that's not a small number either.

是的。

Yeah.

这让你感到惊讶吗？

Did it surprise you at all?

当你看到这些数据时，你有没有感到困惑？我是说，我不知道该怎么理解它。

Like, when when you saw the data, were you, like, you know, on a I'm just I'm I'm trying to understand, you you know, what what to make of it, I guess.

是的。

Yeah.

所以我想，我对他们如此忠诚感到有点惊讶。

So I guess I was kinda surprised at how loyal.

我的意思是，问题是，你会搬走吗？

I mean, the question was, would you move?

我觉得很多人都愿意搬家。

And I think a lot of people are down to move.

但我也惊讶于那些州竟然与那些免税州保持一致。

But I also think that I was surprised at the states falling in line with those tax those income tax free states.

作为一个纽约人，我以为会有更多人提到我们，你知道吗？

As a New Yorker, I thought more of them would mention us, you know?

对。

Right.

很好。

Great.

嗯，Shane，感谢你来做客。

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

这些调查结果很有趣，我们以后会再请你来讨论更多你的报道。

It was some interesting survey results, and we'll have you back on to discuss more of your reporting in the future.

这位是肖恩·伯克，我们《The Information》的高级数据记者。

That is Shane Burke, our senior data journalist here at The Information.

今天的节目就到这里。

That does it for today's show.

提醒一下，我们的直播时间为太平洋时间周一至周五上午10点，东部时间下午1点。

A reminder that we are on this stream Monday through Friday at 10AM Pacific, 1PM Eastern.

感谢大家的收看。

I want to thank you all for tuning in.

我们非常珍视各位的观看。

We really do appreciate your viewership.

我已经对明天的节目充满期待了。

I am already excited for our next show tomorrow.

祝大家周三剩余时光愉快。

Have a great rest of your Wednesday.

暂时先到这里，再见。

Bye bye for now.