为智能体世界构建云 | AWS 首席执行官马特·加曼

而且因为我们有丰富的服务组合，AWS 正是人们真正来启动生产工作负载的地方。

But also because we have that rich set of services, AWS is where people really are coming to launch their production workloads.

所以目前我们大部分的使用量，实际上要么是核心计算、存储，要么是推理，这些都是应用的一部分。

And so the majority of our usage today actually is either core compute and storage and inference, which is part of that application.

这些工作负载，我觉得是不会消失的。

And so those are the workloads that I think just aren't going to go away.

因为我们看到企业获得了正的投资回报。

Because we see enterprises getting positive ROI.

你去跟客户聊，问他们：“以目前的能力和成本，你们看到对业务有正回报吗？”

You go talk to the customers, and you say, "At the capability today and the cost today, are you seeing positive returns to your business?"

几乎每个人都会说：“哦，是的。”

And almost to a person, they'll say, "Oh, yeah."

所以你就会想：“那这肯定不会消失。”

And so you're like, "Well, that's not going to go away."

不存在他们会停止在这方面花钱的泡沫。

There's no bubble in which they stop spending on that.

对。

Yep.

你也知道，风投这个模式，是不是每个十亿美元的初创公司都能成功？

And you know this, the VC model, is every billion-dollar startup going to make it?

不，它们不会。

No, they won't.

但这就是游戏规则。

But that's kind of the game.

这就是游戏规则。

That is the game.

而且这个情况已经持续了50年。

And that's been true for 50 years.

它们并不一直都是十亿美元的初创公司。

They haven't always been billion-dollar startups.

数字变了，但原则没变。

The numbers have changed, but the principles haven't changed.

但原则是一样的，对吧？

But the principles are the same, right?

你投十个，只要有一个成功，就能把其他十个的损失都赚回来，或者差不多这个比例。

You bet on 10, and one makes it and pays for the other 10 or whatever the percentage is.

希望成功率能高于十分之一。

Hopefully it's higher than one out of 10.

这就是著名的幂律分布。

Famous power law.

所以你在互联网时代也看到了这种情况，当时有泡沫，很多互联网公司没撑下来，但互联网本身依然存在。

And so you saw that with the internet where there was a bubble and a bunch of internet companies didn't make it, and the internet's still a thing.

而且很多拥有可持续业务的公司——

And a lot of the companies that had durable businesses-

像谷歌、亚马逊这些，它们都做得很好。

... the Google, the Amazons, the others, they did pretty well.

所以对我们来说，我们对那笔投资以及未来的持续投资感到非常乐观。

And so for us, we feel really good about that investment and the continued investment going forward.

你们对需求有独到的见解——

You guys have a view of the demand-

——这是无与伦比的，对吧？

... that's unparalleled, right?

因为你们能看到全球的情况，覆盖每一个细分市场，从企业到——

Because you're seeing across the globe, you're seeing across every segment, enterprise and-

——大型实验室、AI原生公司等等。

... the big labs and the AI native companies and so on and so forth.

如果谁能最先判断出来，那一定是你们。

If anybody should call it, you should be able to call it first.

希望如此。

I would hope so.

这正是我们的计划。

That's our plan.

说实话，我们花了很多时间思考这个问题。

And honestly, we spend a lot of time thinking about it.

我们在使用股东资金方面非常谨慎，而且我们认为我们正在做很棒的投资。

We're very intentional about how we spend our shareholder capital, and we think that we're making great investments.

我们有很多好的保障措施，确保我们有意地投资这笔钱。

We have a lot of good protections of how we intentionally invest that money.

但是，是的，我们非常看好，而且我觉得安迪也公开说过这一点。

But yeah, we're very bullish on, and I think Andy's been public about saying this.

AWS的潜力真的非常大。

The potential for AWS is really, really large.

在未来十年里，潜力是存在的。

And over the next decade, the potential is there.

而且任何时候，只要机会这么大——

And anytime you have an opportunity that's that big-

你就得投资

You just got to invest

——你就想投资去抓住它。

... you want to invest to go after it.

随着这些数字越来越大，对吧？

As the scale of these numbers go larger and larger, right?

你们的规划和流程有没有发生巨大变化？比如你们不再写——

Has your planning and process dramatically changed in terms of-- You're not writing-

……十亿美元的支票了。

... billion-dollar checks.

你们现在写的是五百亿的支票，或者——

You're writing $50 billion checks or-

……两百亿的支票，或者类似的规模。

... $20 billion checks or whatever it is.

既是也不是。我觉得，很多时候我们仍然非常看好这些投资，并且积极进取，但流程上确实有很多东西完全变了。

Yes and no. I think, look, a lot of times we're still very bullish about the investments and lean forward, but the process, so there's a lot of things that have completely changed.

你们怎么——

How do you even-

如果我想想——

If I think about-

怎么估算2028年或2029年的需求？

... estimate 2028 demand or '29 demand?

嗯，现在有些问题是我们以前根本不用考虑的。

Well, and there's things that we have to think about now that we just never had to think about.

所以，如果回到15年前，我们需要更多电力时，就找电力公司要——

And so if you go back 15 years, if we needed more power, we asked the power company to give us-

对。

Yep

几兆瓦或者什么的，对吧？

... couple of megawatts or whatever it was, right?

几十兆瓦。

Tens of megawatts.

他们就直接给我们，因为10兆瓦不算多。

And they would just give it to us, because 10 megawatts wasn't that much.

而且那对我们来说足够继续发展了。

And that was plenty for us to keep growing.

现在，我们自己带电力，所以我们为电力项目付钱，也为可再生能源项目付钱。

Now, we bring our own power, and so we pay for power projects, we pay for renewable projects.

过去十年，我们每年都是最大的可再生能源购买方之一。

We're one of the biggest renewable power purchasers each year for the last 10 years.

所以我们经常在启动新的太阳能项目、新的核能项目、新的——

And so we're regularly bringing on new solar projects, new nuclear projects, new-

那你说的是在电表后面，还是说——

So are you talking about behind the meter, or are you talking about-

两者都有。

Both

……和运营商合作？

... working with the operator?

我们两种都会做。

We'll do both.

好的。

Okay.

所以这两种方式都有。

And so both of those things.

而且很多时候，我们上这些电力项目，会先投入资金，把项目建好，然后并入电网，之后我们就能获得相应的额度。

And so oftentimes with these power projects that we bring on, we'll pay the capital and pay for the project, and then it'll go into the grid, and then we'll get credit for that.

明白了。

I see.

好的。

Okay.

所以我们会把这些项目做起来。

So we'll bring those on.

有时候我们也会做表后项目。

Sometimes we'll do behind the meter, too.

是混合的。

It's a mix.

以我们目前的规模，这些方面都得考虑到。

At the scale that we're doing, you have to think about all of those things.

但那是规划，你得考虑未来20年怎么处理电力、怎么处理输电、怎么处理那笔资金，这些是我们以前从没想过的事。

But that's planning where you're thinking 20 years out of how you're going to think about power, how you're going to think about transmission, how you're going to think about that capital, and that's stuff we'd never had to think about before.

因为这种规划我们以前根本不用做。

Because that's planning we just never had to do.

但另一方面，以前我们考虑服务器需求时，会有多个季度的需求计划，我们会和我们的——

But the other things is when we used to think about server demand that we needed, we'd have multiple quarter demand things, and we'd talk with our-

……供应商那些人沟通。

... suppliers and things.

现在我们要提前好几年规划，因为规模太大了，我们得考虑26年需要什么、27年需要什么、28年需要什么。

Now we work multiple years out, just because the size is so much that we have to think about what do we need for '26, what do we need for '27, what do we need for '28?

但这也是我们给客户带来的价值之一，对吧？

But it's also one of the value that we bring to customers, right?

这确实是客户自己做不到的事情。

That's a thing that legitimately customers can't do themselves.

他们不是——

They're not-

不是。

No

……去做电力规划。

... going to do power planning.

挺难的。

Pretty hard.

他们不会去规划2028年的内存占用。

They're not going to plan their memory footprint in 2028.

他们做不到。

They can't do that.

所以，这正是我们为众多客户带来的真正价值之一——就是有一大堆事情你根本不用操心，而我们却花了大量时间去考虑。

And so that's one of the real values that we bring to our broad set of customers, is just that's just a whole set of things that you don't have to worry about, and that we spend a huge amount of time thinking about.

你们一般来说，我想说，实际上是几乎所有服务器组件的最大买家，对吧？

You guys are generally, I think for the record, the largest buyers of practically every component of a server, correct?

这个我不清楚。

I don't know that.

我确定我们是市场上比较大的组件采购方之一，这是肯定的。

I am sure that we're one of the bigger purchasers of components out there, for sure.

至于其他所有人，谁知道呢。

And who knows about everyone.

这取决于你怎么看待他们，以及你怎么衡量。

And it kind of depends on how you think about them and how you measure.

但是——

But-

所以我想说的是，你觉得在哪些方面限制最严重，在——

So what I was leading to is where do you see the constraints being most severe in-

……我们就说‘27年、‘28年吧，你觉得哪些方面的限制会缓解？

... let's call it '27, '28, and where do you see the constraints easing up?

这挺有意思的。

It's funny.

所以我用个迂回的方式来回答，我记得在大学本科的时候，我们其实……

So I'll answer this in a roundabout way, but I remember that in undergrad, we actually...

这是很久以前的事了。

So this is a long time ago.

我从没想过我读的这本书会这么有用。

And I never thought this would be a useful book that I read.

但我们读了《目标》这本书。

But we read "The Goal."

哦，当然。

Oh, of course.

你读过《目标》，对吧？

You read "The Goal," right?

结果发现，从来都不只有一个瓶颈。

And so it turns out there's never one constraint.

总是只有最新的那个瓶颈。

There's always just the latest constraint.

所以你得把所有瓶颈都考虑进去，而且一旦你解决了一个，又会出现另一个，对吧？

And so you have to think about all of them, and as soon as you hit one, there's another one, right?

那么，20年和28年会出现什么瓶颈呢？

And so what is the constraint that's going to happen in '20 and '28?

其实我觉得不会有一个单一的瓶颈。

I actually don't think there will be one.

我觉得每个月都会有。

I think it's every month.

对我们来说，就是电力够不够。

For us, it's do you have enough power?

一旦电力不再是瓶颈，可能就轮到内存、台积电的产能、HBM，或者网络组件。

And then as soon as power is no longer the constraint, it might be memory, it might be TSMC capacity, it might be HBM, it might be networking components.

也可能是供应链某个环节出问题，比如连接器之类的，对吧？

It could be there's a blip somewhere in the supply chain and connectors or whatever it is, right?

到了某个阶段，你得把所有环节都考虑进去。

At some point, you have to think about all of those pieces.

而且，这些瓶颈发生在哪里也很重要。

And it's not also the kind of where they happen matters, too.

所以可能会是，嘿，我们在印度尼西亚有大量的电力。

So it may be like, "Hey, we have a ton of power in Indonesia-

但我们在德国却不够用。

... but we don't have enough in Germany."

所以你也得考虑你想把产能放在世界的哪个地方，因为事实证明，并不是所有东西都是完全可以互换的。

And so you think about where in the world you want that capacity, too, because it turns out that not everything is totally fungible.

有些可以，有些不行。

Some is and some's not.

但可能是硬盘，也可能是固态硬盘。

But it might be disk drives, it might be SSDs.

我们会考虑每一个零部件，我们有成千上万个零部件，都在我们的跟踪和考虑范围内。

We think about every single component, and we have tens, hundreds of thousands of components all that we track and think about.

有些我们依赖供应商来管理。

And some we rely on our suppliers to manage.

大部分我们直接管理。

Much of it, we directly manage.

是的，我们有一个专门的团队负责这个，他们做得非常出色。

And yeah, we have a whole team that does that, and they're fantastic.

我觉得他们是行业领先的，大概十年前我们就预见到了这个问题，并且不只是考虑我们需要追踪多少台服务器，而是深入思考整个供应链，一直到第四层、第五层。

I think they're industry-leading, and we saw this problem coming probably a decade ago and really started not just thinking about, okay, how many servers do we need to track, but just thinking all the way through the supply chain, four tiers, five tiers down.

哪些组件可能会给我们带来问题，并确保我们在这些组件上有稳定的供应保障。

What is the component that could cause an issue for us, and making sure that we had guaranteed supply on that.

如果你还记得，天哪，我不记得这是什么时候的事了。

If you remember, gosh, I don't remember when this was.

那是十多年前，泰国发生洪水的时候，

This was over a decade ago, when there was the floods in Thailand and-

……大家都没有硬盘了。

... no one had disk drives anymore.

对。

Yep.

所以那就是……

And so that was-

当时有硬盘危机——

There was disk drive crises-

……然后又有内存危机。

... and then there was memory crisis.

没错。

Exactly.

所以我觉得我们把这些都考虑进去了，同时也在考虑制造业的多元化布局。

And so I think we think through all of those things, and so we also think about where is there diversification in manufacturing.

所有这些方面我们都尽量去处理，虽然不可能做到完美，但总会有不同的供应瓶颈。

All of those kind of pieces we try to work through, and we're never going to be perfect on it, but there's always a different supply constraint.

所以很明显，关于数据中心有很多广泛的讨论。

So obviously, there's a lot of wide-ranging debate about data centers.

而且像我们这样的人立场很明确，但你觉得作为行业，我们是不是没有很好地解释——

And it's clear that folks like us, where we stand, but do you think as an industry, we have not done a good job-

……为什么数据中心对美国有好处——

... of explaining why data centers are good for America-

以及对整个世界来说，但是——

... and generally the world, but-

那在内部讨论中——

... and what's the internal talk-

安迪的团队是怎么讨论应对这个问题的？

... amongst Andy's team on how do we deal with this?

嗯，你看，我觉得你之后会听到我们更多关于这方面的表态，我也同意，我们需要更积极发声，更坦诚一些。

Well, look, I think you'll hear more from us over this, and I agree, I think we need to be more vocal and be more upfront.

因为我们确实做了很多非常有益的事情，无论是对我们运营所在的社区。

Because we actually do a ton that's really beneficial, both for the communities we operate in.

我们花了很多心思思考如何为这些数据中心引入可再生能源。

We think a ton about how do we bring renewable energy to these data centers?

我们如何做到水资源正效益？

How do we think about being water positive?

实际上，我们的数据中心用水量非常非常少。

Actually, our data centers use a really, really small amount of water.

我们主要用的是自然风冷却。

We mostly use free air cooling.

我们怎么考虑成为所在社区的好成员，以及怎么为我们运营的社区带来高薪工作？

How do we think about being great participants in the communities where we are, and how do we bring high-paying jobs to the communities we operate in?

并不是所有数据中心运营商都这么做。

And not all data center operators do that.

我觉得有些——

I think there are some-

而且你有一个相当——

And you have a pretty-

有些其他公司的例子被广泛报道，他们在这方面做得不好，根本不重视法规。

There are some well-chronicled examples of others out there that are not great at that, and they just don't really pay attention to regulations.

他们觉得规则不适用于自己，或者就是……不考虑这些就急着上线。

They think that the rules don't apply, or they just ... launch really quickly without thinking about those.

我觉得这给整个行业带来了问题，因为大家都被归为一类了。

And I think that it causes a problem for the whole industry because everybody kind of gets lumped into that.

所以你看，我觉得你说得对。

So look, I think you're right.

我们确实需要更主动地自我批评，同时也要更积极地宣传我们带来的好处，并思考更多方式帮助社区了解我们为他们带来的价值，无论是他们使用的服务，对吧？

Vocally self-critical, we need to be more vocal about the benefits that we do bring, and think about additional ways that we can help communities understand the benefits that we bring to them, both for the services they use, right?

如果你通常去一个社区说："那你们不想用Netflix吗？"

If you usually go to a community and say, "Well, do you not want to use Netflix?"

他们会说："不不不，我还是想要Netflix。"

And they'll be like, "No, no, no. I still want Netflix."

对我们来说，重要的是思考和强调我们带来的好处。我最近看到一份报告，在我们运营的一个社区，由于我们缴纳的税款，那个县的每个人都少交5000美元的税。

And it's important for us to think about and highlight the benefits that we bring where I recently saw a report where one of the communities that we operate in, everybody in that county pays $5,000 a year less in taxes because of the taxes that we-

对

Yep

……我们带来的。

... bring to that.

但我们没有告诉他们。

And we don't tell them.

他们甚至都不知道，对吧？

They don't even know it, right?

对。

Yep.

没错。

Exactly.

这对他们来说完全是看不见的。

It's just invisible to them.

所以我觉得我们需要把这些好处讲得更清楚一些。因为如果你告诉社区：顺便说一句，如果没有我们，你们的税款会多五千美元，他们可能就会对……

And so I think we just need to be more clear about those benefits that we bring, because I think if you told the communities, "By the way, your tax bill is $5,000 less than it would otherwise be if we weren't here," they might have a little bit of different thought about-

……那边的建筑有些不同的看法。

... the building that's over there.

但并不是每个人都这么做。

But not everyone does that.

想必在……方面还需要更多的透明度。

Presumably there's more transparency that's needed on the part-

我觉得数据中心行业里的大部分公司，不只是我们，其实都是挺守规矩的。

And I think much of the data center community, not just us, is actually pretty good actors.

我完全同意。

I completely agree.

就是有那么少数几家不太守规矩的，我觉得最近引起了一些不满，所以我们需要更好地宣传哪些是好的企业，哪些不是。

And there's just a few that aren't that kind of, I think, have caused some of the angst recently, and I think we just need to do a better job of highlighting who's being good citizens and who's not.

在结束硬件话题之前，我想聊聊Trainium。

And before we leave the hardware topic, I want to touch upon Trainium.

还有你们自己造芯片的整个历程，对吧？

And your whole history with building your own chips, right?

我们是很早以前就用Nitro的第一批合作伙伴之一。

We were one of your first partners using Nitro a long time ago.

从那以后，还有Graviton，你们取得了巨大的进步。

And since then, Graviton, and you made tremendous progress.

那么，是什么想法让你们决定说‘我们要自己搞一套’呢？

So what was the thinking that led to saying, "Look, we're going to do our own thing"?

那这个进展怎么样，你们现在到哪一步了——

And then how has that progress been, and where are you-

——关于Trainium？

... with respect to Trainium?

这其实是个很有意思的故事，我觉得它很好地体现了亚马逊AWS会不断创新、不断迭代，并且持续思考我们能做什么更大的事情，但我们会一步步证明自己，而不是……

It's actually a fascinating story, and I think it's a great example of where Amazon AWS will innovate and will iterate over time and continue to think bigger about what we can do, but kind of prove our way there as opposed to...

就拿这个举个例子吧。

And so take this as an example.

这大概是，我也说不准，10……

This is probably now, I don't know, 10...

大概13、14年前吧。

It was probably about 13, 14 years ago.

我们当时发现，虚拟化对整个资源数量来说，是一笔相当大的开销。

We were seeing that there was a pretty significant virtualization tax on the overall number of resources.

我们当时在想怎么——客户跟我们说，“我要裸机性能”，他们在比较——

And we were kind of thinking about how do we-- Our customers were telling us, "I want bare metal performance," and they're comparing to-

拥有服务器的全部资源。

... having all the resources of a server.

所以我们做的第一件事，就是拿一块网络卸载卡，把所有的网络虚拟化都放到上面去，这样网络虚拟化就更接近裸机性能了。

And so the first thing that we did is that we took a network offload card and virtualized all of our network virtualization and pulled it off onto an offload card so that network virtualization got closer to bare metal performance.

那时候，虽然还不是完全的裸机性能，但已经更接近了。

And back then, it was not quite bare metal, but it was closer.

然后我们特别兴奋，就说：“好，那如果我们把存储虚拟化也移出去呢？”

And then we got really excited about that, and we said, "Okay, what if we could move storage virtualization off as well," right?

但那些网络卸载卡都做不到这一点。

And none of the network offload cards could do that.

后来我们找到了一家公司，他们的卸载卡上有一些Arm核心，他们做这个是为了其他用途。

And then we found this one company who had some Arm cores on an offload card, and they were doing it for other reasons.

我不记得他们最初的目的了，但我们就问：“能不能用这些核心来做存储虚拟化，还有其他一些功能？”

I can't remember their original purpose, but we're like, "Could you use those to do storage virtualization and some of these other functions?"

他们就说：“也许可以吧。”

And they're like, "Maybe."

所以我们真的和他们反复磨合了很久。

And so we really iterated with them.

这就是Annapurna团队。

This was the Annapurna team.

对，就是Annapurna团队。

Yeah, this is the Annapurna team.

我们非常喜欢那个团队。

And just loved that team.

他们真的很有创新精神，也很有使命感——

Like really innovative, really mission-driven-

非常棒的团队。

Fantastic team

……真的很想解决问题。

... really wanting to solve problems.

所以我们收购了他们，然后说：“你们能不能给我们造一个稍微大一点的卡，把所有的网络虚拟化都拿掉，基本上给我们一个没有虚拟化的裸机服务器，没有虚拟机虚拟化，所有东西都通过卡上的API来操作？”

And so we acquired them, and we said, "Look, could you build us a slightly bigger card that actually could take all the network virtualization off and basically give us a bare metal server that has no virtualization on it, no VM virtualization, everything is through APIs on the card?"

因为我们当时认为，第一，性能会好很多，资源利用率也会更高。

Because we had this view that, one, performance would be much better, a resource utilization would be better.

安全隔离。

The security isolation.

而且安全隔离会好得多得多。

And security isolation would be much, much better.

然后我们就可以理直气壮地告诉别人：“我们完全无法访问你们运行在那里的任何虚拟机。”

And we could then legitimately tell people, "We have no access to any of your VMs that are running there."

这在过去十年里对我们来说是个巨大的优势，说实话，我们一直处于领先地位，而且——

And this has been a huge benefit for us for the last decade, where frankly, we've been leading and-

……其他人在这方面动作比较慢，因为这不是一件人人都能做到的普遍事情。

... others have been kind of slow to do this because this is not a generalized thing that people can do.

所以我们做到了那一步，然后基本上说：“看，我们在这里取得了很大的进展。

So we got to that, and we basically said, "Look, we're making a lot of progress here.

如果我们把……”

What if we take..."

而且那块卸载卡上有一堆Arm核心，

And there's a bunch of Arm cores that were on-

然后我们说：“要是把它变成一台服务器会怎么样？”

... this offload card, and we said, "What if we turn that into a server?"

我们最初用Graviton做到了这一点。

And we did that first with Graviton.

我们推出的那台服务器性能很弱，体积也很小。

It was a very underpowered, very small server that we launched.

但客户们都很兴奋。

And customers were excited.

他们说：“哇，我太想要一台Arm服务器了。

They're like, "Ah, I'd love to have an Arm server.

这太有意思了。”

This is super interesting."

于是我们就沿着这条路走下去，Graviton……

And so we went down the path and Graviton...

我们当时做的一部分工作就是观察，发现Arm核心的性能提升速度在加快。

And part of what we did is we looked and saw that there's the slope of Arm cores were getting faster.

对。

Yep.

而且你看那个功耗利用率的图表，就是那个……

And where you saw the power utilization and the graph where the...

你知道这个架构在某个点上会变得非常适合某些应用，只是需要有人来推动生态系统，把一些环节落实到位。

And you knew the intercept was going to happen for where this architecture was going to be really good for parts, and they just needed somebody to drive the ecosystem and get some of the pieces in place.

所以我们用Graviton做了这件事，而Graviton现在已经是爆款了。

So we did that with Graviton, and Graviton's been a runaway hit at this point.

你们有没有公开过Graviton在你们整个服务器集群中的占比？

Have you been public about what percentage of your fleet is Graviton?

我们每年部署的Graviton服务器比其他任何类型的都多。

We land more Graviton ships every year than any other type.

所以它非常受欢迎。

So it's very popular.

你看，它们便宜20%，性能还高出20%——

Look, they're 20% cheaper at a 20% better performance-

而且过去五六年一直是这样。

... and have been like that for the last kind of five, six years.

所以这个价值主张很容易理解，而且我认为绝大多数，大概超过90%的顶级客户，都以某种方式在他们的整个基础设施中使用了Graviton。

So it's an easy value proposition, and I think the vast majority, something like 90-plus percent of our top 100 customers all use Graviton in some way, shape, or form across their fleets.

所以这对我们和客户来说都是一个巨大的成功。

And so that's been a huge win for us and for customers.

客户降低账单最简单的方法就是迁移到Graviton。

The single easiest way that customers lower their bill is to move to Graviton.

我们有过这样的例子，有些客户迁移了整个基础设施，服务器数量减少了一半。

We've had examples where people have moved their whole fleets and cut the number of servers they had in half.

哇。

Wow.

性能提升太明显了。

The performance is so much better.

太厉害了。

It's amazing.

所以服务器数量减半，每台服务器的成本也更低。

And so half as number of servers, each server costs less.

这真是个巨大的胜利。

It's a big win.

大概五六年前，我们看到了AI计算需求的兴起。

And so then about five or six years ago, we said we saw the rise of AI compute happening.

虽然当时没预料到今天这么火爆，但还是……看到了它会成为重要的推动力。

Not nearly expecting what it was today, but still- ... saw it was going to be a big material mover.

是的。

Yes.

于是我们投入研发，推出了第一代Trainium芯片，现在第三代Trainium3已经上市，效果非常棒。

And so we went in, built our first chip in Trainium, and we're now in market with the third generation, Trainium3, and have seen fantastic results.

所以我们的产能已经售罄，可能要到明年年底左右才能有新的供应。

So we're sold out For capacity through probably, maybe towards the end of next year or something like that.

而且我们还在想办法节省一些算力，让初创公司也能用上一些，因为我们看到了很好的效果。

And again, we're trying to figure out how we can save some capacity and get startups to be able to use some of the capacity, because we see great results.

所以Bedrock上的大部分流量都跑在Trainium上，我们和Anthropic、OpenAI都有很好的合作，让他们在Trainium上构建，还有一些较小的初创公司。

So most of the majority of traffic on Bedrock all runs on Trainium, and we have great deals with both Anthropic and OpenAI to build on top of Trainium, as well as a set of smaller startups.

我觉得现在大概有六到十二家初创公司在Trainium上开发。

And I think we have half a dozen to a dozen startups that are building on top of Trainium now, too.

而且——

And-

所以从名字上看，这是一个训练芯片。

So now the name suggests it's a training chip.

对，我们不太擅长——

Yeah, we're bad at-

但大家也都用它来做推理。

But everybody's using it for inference, too.

那在架构上，它的优势在哪里？

So where is the lean architecturally-

……那它未来的发展方向是什么？

... and where is it going?

这个问题问得好。

It's a good point.

说实话，我们对自己挺苛刻的，我们起名字真的很差，这不是我们的强项。

Look, the vocally self-critical, we're terrible at naming, and so it's not our strength.

最初，我们有一款芯片叫Inferentia，是用于推理的——

Originally, we had a chip called Inferentia for inference-

……而Trainium是用于训练的。

... and Trainium for training.

后来模型越来越大，结果发现你其实需要在这些超大型系统上运行推理，结果就是——

And then as the models got bigger and bigger, and it turns out you actually want to run the inference on these really large systems, it turns out that-

而且训练也需要推理。

And training requires inference

……Trainium现在可能是市场上最好的推理芯片了——

... Trainium is actually maybe the best inference chip on the market right now-

……

...

从绝对性能和性价比的角度来看。

from a absolute performance and cost performance point of view.

所以——

And so-

是内存更好，还是内存带宽更好？

Is it a better memory, memory bandwidth?

优势在哪里？

Where is the wins coming from?

有很多方面。

A bunch.

只是架构有点不同——

It's just the architecture is a little bit different-

好的

Okay

比其他产品要好，而且便宜得多。

... than others, and it's much cheaper.

所以从性价比和绝对性能的角度来看，Trainium 非常出色。

And so from a cost performance perspective and absolute performance perspective, Trainium is great.

所以我们大量使用它进行推理，而且正如我所说，Bedrock 今天的很多推理都是由它驱动的。

And so we use it a ton for inference and as I said, Bedrock, it drives much of the Bedrock inference today.

但它也是一个很好的训练芯片，我认为我们很多广泛的客户使用它并不是为了训练模型，而是为了使用它。

But it's also a good training chip too, and I think a lot of our broad set of customers' usage is not in training models, but is in using it.

所以很多人是在幕后使用它，这也是我们对它感到兴奋的地方。

And so that's where a lot of people get to use it under the covers, and that's where we're excited about it.

但很多大客户也对它用于训练集群感兴趣，尤其是随着我们发布的 Trainium 3 和 4。

But a lot of the big customers are interested in it for training clusters as well, and particularly as you get to Trainium 3 and 4, which we've announced.

我们还没有推出 Trainium 4，但已经发布了。

We haven't launched Trainium 4 yet, but announced it.

人们看了这个架构后说：'是的，这也是我希望我的训练集群未来所在的地方。'

Folks have kind of looked at that architecture and said, "Yeah, that's the future of where I want my training clusters to be as well."

所以我们对未来非常兴奋，尤其是这些超大规模的训练集群。

So we're quite excited about the future, where that goes for these really broad scale-

训练集群也是。

... training clusters also.

对，两者都有。

But yeah, it's both.

现在，我们回到智能体的话题，但从大型企业的角度来看。

Now, let's get back to talking about agents, but from a perspective of large enterprises-

或者大型中型企业。

... or large medium enterprises.

他们在采用方面处于什么阶段？

Where are they in their adoption?

你已经看到他们获得了哪些好处？

And what sort of benefits are you seeing them reap already?

从你的角度来看，他们的路线图是什么？

And what is the roadmap for them as far as you can tell from your vantage point?

这个问题问得真好。

It's a really good question.

我觉得这个问题我们花了很多时间思考。

I think it's one that we've spent a lot of time thinking about.

而且我跟客户们聊天的时候，他们现在从已经做的事情里收获了很多价值。

And when I talk to customers all out there, today, they're getting a lot of value out of what they've done today.

我想说，大多数企业构建的代理相对简单直接，他们开始考虑——而且这些代理大多不是自主的，对吧？

And I would say the agents that most enterprises have built are relatively simple and straightforward, and they're starting to think about-- And they're mostly non-autonomous, right?

它们还是需要人在里面参与，可以这么说。

They're still kind of people in the loop, if you will.

顺便说一句，客户现在仍然从中获得很多价值。

And by the way, customers are still getting lots of value out of that today.

所以他们真正在思考的是，怎么让这些代理自主运行，但又要安全。

And so they're really thinking about, how do I have these be autonomous, but in a safe way?

我觉得有两件事现在阻碍了客户继续扩大规模。

And I think there's two things that I think hold customers back today from just continuing to scale.

现在这已经是一门不小的业务了，但我认为它还有巨大的机会，真正改变每一个客户和每一个工作流程，这确实值得深思。

And it's already a pretty big business today, but I think it has a massive opportunity to really change every single customer out there and every single workflow and really thinking about it.

所以第一点就是怎么去思考这个问题。

And so number one is just how to think about it.

我觉得我们最初看到的是，企业有一个工作流程，然后他们说：‘太好了，我可以让一个代理去执行同样的流程。’

I think what we originally saw was that enterprises had a workflow, and they're saying, "Great, I would have an agent go do the same workflow."

而我们真正鼓励他们做的是，不要只是复制鲍勃做第一步、第二步、第三步、第四步、第五步，然后代理也做第一步到第五步，最后鲍勃再检查一遍。

And what we encourage them to really do is think, not just replicate Bob does step one, two, three, four, five, so agent is going to do step one, two, three, four, five, and then Bob's going to check it at the end.

这其实不是你想要的模式。

That's not really the model you want.

你应该退一步想：‘如果我想完成某件事，代理怎么能用不同的方式去做？’

You want to actually step back and say, "If I want to accomplish something, how can an agent do it differently?"

它可以大规模并行处理。

It can do it in a massively parallelized way.

它可以尝试50种不同的方法，然后达到目标。

It can try 50 different things and get to that.

那你如何帮助它达到那个正确的结果，并且重新思考计算机解决问题的方式，而不是像人类那样解决问题呢？

And how do you help it get to that right outcome and rethink how a computer would solve a problem versus a human solving a problem?

所以其中一件事就是我们帮助客户理解如何思考这个问题，真正地拥有那种空白画布，因为那才是你真正获得价值的地方——不是仅仅复制你目前在做的事情，而是从全新的角度思考如何以完全不同的方式解决问题。

And so one of the things is us just helping customers understand how to think about that and really kind of have that blank slate, because that's where you really get value, is not just replicating what you're doing today, but thinking from a greenfield approach about how you go solve a problem completely differently.

我相信你们很多初创公司也是这么想的。

And I'm sure that's how many of your startups are thinking about this, too.

你们如何帮助客户从零开始解决问题，而不是复制现有的流程？

How do you help customers greenfield solve a problem, not replicate the thing that happens today?

所以这是第一点。

So that is number one.

那些运营着大量代理和代理集群的人，不管你怎么称呼它们——

People running fleets of agents and swarms or whatever you want to call them-

……现在非常普遍。

... very common these days.

你只需要思考一下。

And you just want to think about it.

所以企业其实没那么前卫，这又是一个从初创公司学习的地方，你得想办法把它应用到企业环境中，比如保险公司不一定那么前卫，但他们很想知道怎么优化审批流程之类的。

And so enterprises are not as, again, this is one where you learn from the startups and you try to, how do you apply that to an enterprise world where an insurance company is not necessarily as forward-leaning, but they would love to figure out how they can have a better approval workflow or something like that.

这是第一点。

So that's number one.

但第二点是，怎么把这些变成完全自动化的流程，以及怎么真正信任这些智能体。

But then the second one is this, how do you turn those into fully autonomous workflows, and how do you actually trust the agents?

所以我们花了很多时间思考，怎么构建服务，让企业觉得他们的系统是安全的，能信任智能体做决策，有合适的护栏，对数据有正确的权限，不会删掉生产系统，不会犯灾难性的错误。

And so we're spending a lot of time thinking about how do we build services to help enterprises feel like their systems are secure, and that they can trust an agent to make a decision, that they can have the right guardrails, that it can have the right permissions on their data, that it's not going to delete production systems, that it's not going to make tragic mistakes.

现在，我觉得这种担忧可能正在阻碍人们前进。

And right now, I think that nervousness is probably holding people back.

对，这影响很大。

Yeah, it's huge.

顺便说一句，这种担忧可能也是合理的，它阻止了企业直接说：“好吧，放手去干吧。”

Maybe appropriately, by the way, is holding enterprises back from just saying, "Okay, go nuts."

你当然不希望智能体乱来，不小心删掉生产数据库。

You don't actually want an agent to just go crazy and accidentally delete a production database.

那可就麻烦了。

That's going to be pretty bad.

所以这就是我们目前和客户一起积极应对的方式，既要帮他们设计架构，坦白说也要发明新的技术和能力，来帮他们解决这个问题。

And so that's how we're kind of actively working through this with customers on how do we both help them architect and frankly invent new technologies and capabilities that are going to help them solve that problem.

这也是我认为我们会持续创新的领域之一，我们一定能做到。

And so that is one of the areas where I think we'll continue to innovate, and we'll get there.

我觉得我们有一些非常好的想法和技术正在酝酿中，真的能帮上忙。

I think we have some really good ideas and some good technologies brewing that I think can really help.

那企业是不是也在学习怎么做评估系统之类的，来确保智能体不跑偏？

So are enterprises learning how to do eval systems and so on and so forth to keep the agents on the straight and narrow?

他们需要帮助。

They need help.

说实话，评估是一方面，怎么建立一个持续的测试循环？

Honestly, both evals, how do you have a constant loop of testing?

怎么考虑目标导向——

How do you think about goal seeking-

这是一个新的软件初创公司。

This is a new software startup

……以一种合理的方式？

... in a reasonable way?

你怎么给数据打标签，才能让评估系统真正合理地模拟你在生产环境中的操作？

How do you have your data labeled in such a way that it actually even makes sense that the eval can actually kind of approximate what you're going to be doing in production?

你怎么在生产环境中进行测量和回测，以确保没有出现漂移？

How do you measure in production and back test it so you're not seeing drift?

所有这些问题都是企业目前不知道如何解决的。

All of those things are problems that enterprise don't know how to solve today.

实际上——

Actually-

当然。

Sure

……

...

我不知道现在是不是真的有人能把这些事情做得很好。

I don't know if anyone really is great at solving these today.

这就是为什么你会看到那么多FDE团队冒出来，AWS也是，我们的合作伙伴也在大力推动FDE，去帮助客户，而这是客户最需要帮助的领域。

It's why you've seen so many FDE teams kind of spin up, and AWS is, and our partners are really leaning into the FDE motion to go and help, and this is the single biggest area where customers need help.

当我们考虑如何……

And when we think about how you...

我的看法是，我们希望培训我们的客户，让他们能够自己去做这些事情。

And my view is we want to train our customers to be able to go and do this themselves.

这不是传统的模式，不是那种靠人力驱动的业务，永远持续下去，你不停地付钱给顾问——

This is not the traditional motion where I want to have a people-driven business that goes on forever, where you just keep paying consultants-

一次又一次地付钱。

... over and over and over again.

我们对FDE应该怎么运作的看法是，我们希望进入一个准备好真正接受在项目结束后自己接手这些工作的客户，在45天内，我们教他们如何做评估，教他们如何以有标签的方式整理数据，和他们一起工作，然后在45天结束时，我们离开，那个客户已经准备好，训练有素，可以自己继续了。

Our view on how FDE should work and is, we want to go into a customer who's ready to really kind of accept owning this when we're done, and in 45 days, do work where we can teach them how to make an eval, teach them how to get their data in a labeled way, do work alongside with them, and then at the end of 45 days, you leave, and that customer is good and ready to go and trained up.

而这就是我们的客户告诉我们的，他们想要的。

And that's what our customers tell us they want.

他们不想在未来五年里依赖外部劳动力。

They don't want to be beholden to an external workforce for the next five years.

但他们现在就需要帮助。

But they need help today.

你在FDE上投入了巨资。

You've made a massive investment in FDEs.

再进一步说，你的一些同行说过：“你看，你不能把所有数据都放进一个大型前沿模型里。

So taking that even one step further, some of your industry peers have said, "Look, you can't have all of your data going into a big frontier model.

企业真正应该做的是采用开源模型，然后用自己的数据、工作流程和术语等进行后训练。”

What enterprises should really do is to take an open source model and then post train on your own data and workflows and phrases and whatnot."

你对此怎么看？

Where do you stand on that?

你看到客户真的在尝试这样做吗？

Are you seeing customers actually trying to do that?

是的。

Yep.

或者你们——

Or do you guys-

嗯。

Well.

你们怎么看这个问题？

How do you think about that?

第一点，我完全同意。

The first point, I wholeheartedly agree on that first point.

企业数据是他们最宝贵的资产。

Enterprise data is their most valuable asset.

所以从一开始，我们就是这样构建Bedrock的。

And so from the very beginning, it's why we built Bedrock like we did.

我们保证你的数据永远不会离开你的VPC。

We have a guarantee that your data never leaves your VPC.

所以如果你在Bedrock内部运行，你的数据不会回到模型提供商那里。

And so if you're running inside of Bedrock, your data doesn't go back to the model providers.

不会。

No.

他们根本看不到你的提示词。

They never see your prompts.

这些信息都保留在你自己的可信环境里。

That stays inside of your own trusted environment.

所以企业才倾向于在 Bedrock 上运行，这也是为什么你看到它的业务增长得那么快，规模达到了数百甚至更多。

And so that is why enterprises kind of run, they prefer to run on top of Bedrock, and it's why you see that business growing massively, like hundreds and hundreds.

每个月我们都能看到业务增长，几乎每周都能感受到业务在爆发式扩张。

Every month, we see that business, just every week, we see that business exploding.

这也是为什么你会看到 OpenAI 的工作负载迁移到 Bedrock 上。

And it's why you see open AI workloads migrating to Bedrock.

这也是 Anthropic 之所以能迅速崛起的原因。

It's why you see Anthropic really growing really rapidly.

所以不管你是用开源模型还是闭源的前沿模型，据我们的客户反馈，Bedrock 都是一个很棒的解决方案。

And so whether you're using open models or closed frontier models, I think Bedrock is a great solution that our customers told us.

顺便说一句，你还记得吗，三年前我承受了很多压力。

By the way, if you remember, three years ago, I got a lot of heat.

说到名字不好听，不过这个名字其实挺好的。

Speaking of bad names, that's a good name, though.

对，Bedrock是个好名字。

Yeah, Bedrock is a good name.

确实不错。

That's good.

但当时我们确实因为进入AI领域比较慢而受到了很多批评——

But we got a lot of heat, actually, for being slow to the AI world-

是的

Yes

……因为我们当时是在打基础，我们说：“听着，我们不会急着推出一个服务。

... because we actually built the foundations of this where we said, "Look, we're not just going to rush out a service.

我们真的想好好考虑如何保护客户的数据，并且打造一个我们认为能够持久满足已知应用场景的服务。”

We really want to think about how do we make sure that we protect our customers' data and build a service that we think is going to be durable for the use cases that we knew about."

你还记得吗，我们当时承受了很多压力，我们说：'我们要去构建正确的东西。'

And if you remember, we got a lot of heat, and we said, "Look, we're going to go build the right thing."

现在，当人们从概念验证转向生产环境时，绝大多数都落在了AWS的Bedrock上。

And now, as people move from proof of concepts to production, vast majority of them are landing in AWS on Bedrock.

其中一个原因就是这个。

One of the reasons is because of this.

还有就是因为我们的服务组合。

It's also because of the set of services that we have.

我们也提供开源模型。

We also offer open models.

我们也提供专有模型。

We offer proprietary models.

我们围绕这些模型提供了一整套能力。

We offer a whole set of capabilities around those.

Agent core，我们构建了这些积木，这样你就可以更容易地用你想要的任何模型来构建代理，不管它们是在Bedrock内还是Bedrock外。

Agent core, we build these building blocks, so it's easier to build agents with any of the models that you want, whether they're in Bedrock or out of Bedrock-

……就这一点来说。

... for that matter.

你可以用Gemini或者其他工具来做这件事。

You can use Gemini or other things for it.

但我觉得这对我们来说是一个差异化优势，而且这是一个非常重要的问题，因为如果数据回流到模型提供商那里，我认为那是一件危险的事情。

But I think it's a differentiating piece for us, and it's a super important thing to think about because having that data go back into the model provider, I think is a dangerous thing.

不过你提到了开放权重模型。

You talk about open weights models, though.

我确实认为有一个让我很兴奋的场景，那就是我们正在大力加强对开放权重模型的支持，努力打造一个良好的环境。

I do think that there's a scenario that I'm excited about, where we're really ramping up our support of open weights models and trying to build a good environment.

而且说实话，现在很多客户都认为他们拥有重要的专有数据，如果他们……

And frankly, this is where today, and I think it's true, a lot of customers believe that they have meaningful proprietary data, that if they-

……能够把这些数据混合进去，进行一些后训练，对开放权重模型进行微调，然后进行蒸馏，他们实际上可以获得性能更好、价格更低的模型。

... could mix in, do some post-training, do some fine-tuning to an open weights model that they could distill down, they can actually get a better performing model at a lower price.

这里面有很多环节需要顺利推进。

There's a bunch of pieces in here that have to work out well.

他们确实有很好的评估来证明这一点。

They actually have a good eval to actually prove that that's true.

现在大多数人都在SageMaker上做这个。

Most people are doing that in SageMaker today.

我觉得我们还可以做更多，让这个过程更简单。

I think there's more that we can do to make that easier.

但如果你去看看人们在哪里做这个，他们现在确实是在AWS的SageMaker上做的。

But actually, if you go look at where are people doing that, they actually do it in SageMaker on AWS today.

哦，真的吗？

Oh, really?

好的。

Okay.

而且他们实际上是通过SageMaker来托管推理的。

And they actually host the inference via SageMaker.

SageMaker作为……正在获得新的生命力。

SageMaker is getting a new lease of life as a-

它一直是个模型构建平台。

It was always kind of a model-building platform.

现在，想想企业在做什么，他们基本上就是在构建自己的定制模型。

And now, if you think about what enterprises are doing in this world, that's what they're doing, is they're basically effectively building their own custom models.

所以，这就是……

And so that is-

……SageMaker 是个很好的地方来做这件事。

... SageMaker is a great place of doing that.

我觉得我们还需要继续开发一些功能，让它变得越来越容易，并且能在不同的开放权重模型上进行测试，等等。

And I think there's some things that we need to keep building on to make that easier and easier to do, and to test across different open weights models, things like that.

但这是一个不断发展的领域，我觉得非常有意思，我们想确保客户如果有合适的数据和专业知识，能够拥有所有必要的工具去走这条路。

But it's an evolving space, and I think it's a super interesting one, and it's one we want to make sure that we have all the right things for customers to be able to do if they have the right data and expertise to actually go down that path.

而且，关于 AI 扩展风险的各种争论，还有安全问题……

And with all the debate around AI extension risks and this, that, and the other, and the security-

……

...

漏洞什么的，还有Hugging Face被攻击的事，企业高管们都在问你些什么？

vulnerabilities and so on and so forth, and the Hugging Face attack, how are enterprise, what are CEOs asking you about all these things?

问的挺多的。

There's a bunch.

你看，他们主要想说的是——

And look, they mostly want to say-

当然，我该——

And of course, what should I-

这又回到那个问题，比如我部署智能体的时候，怎么才能相信它们会按我的要求去做？

And it goes back to this, like when I launch agents, how can I trust that they're going to do what I want them to do?

所以我们越来越——其实我们一直在做——

And so we're increasingly, we have been for a while.

我们基本上在大力投资，开发一些功能，让人们能安全地把智能体部署到他们的环境里，并且考虑那些控制措施。

We're basically Heavily investing in building capabilities that allow people to deploy agents safely into their environment and think about those controls.

其中一些就是，怎么确保智能体有正确的权限？

And some of those are, how do you make sure the agents have the right permissions?

他们怎么才能有正确的沙箱环境？

How do they have the right sandboxing?

你怎么确保有合适的防护措施？

How do you make sure that you have the right set of guardrails?

你怎么真正有意识地思考你希望代理做什么、不做什么？

How do you really intentionally think about what you want the agents to do and not do?

有没有人在流程中参与，还是完全自动化？

Is there a human in the loop or not?

所以我们花了很多时间和客户一起思考，如何安全地部署代理，并随着时间不断改进，以及我们还需要开发哪些能力来帮助人们安全地在他们的环境中部署代理。

And so we spend a lot of time with our customers thinking about how do you think about safe agent deployment and get better over time and, what other capabilities do we need to go build to help people deploy agents safely into their environment?

所以这方面有很多内容。

So there's a lot there.

另一个很多人担心的角度是，这些非常强大的模型会不会成为攻击面，以及某种意义上的威胁领域。

The other angle on that, which a lot of people are worried about, which is just, are some of these really powerful models going to be, attack surfaces and kind of mythosphere.

所以我的看法是——是的，我认为这对客户环境来说确实是一个风险，但也是一个真正的机会。

And so I kind of have a view on they're-- Yes, that is a real risk, I think, to customer environments, but it's also a real opportunity.

所以我们最近推出了一项叫 Continuum 的服务，利用这些强大的模型来帮助客户保护他们的环境安全。

And so we recently launched a service called Continuum that uses these powerful models to help customers go secure their environment.

我们会全面检查他们的环境，和他们一起寻找漏洞。

And so we'll look across their environment, look for vulnerabilities with them.

我们会利用这些强大的模型，帮助客户发现他们之前没有发现的漏洞。

We'll use some of these powerful models and help customers find vulnerabilities they haven't found before.

而且最重要的是，我们会帮他们确定哪些漏洞需要优先处理，因为我们了解他们环境的上下文，知道环境是怎么设置的，权限在哪里，以及哪些地方可能有补偿性控制措施，这些都会影响修复的难易程度。

And most importantly, by the way, prioritize which ones, because we know context about their environment, how it's set up, where their permissions are, where they may have compensating controls that make it harder or easier to do.

所以 Continuum 在客户中非常受欢迎。

And so Continuum is incredibly popular with customers.

实际上，我们对 AI 在安全领域的应用前景非常乐观。

Actually, we're really bullish about what's possible from AI to help with AI-powered security.

因为你看，总有一天，客户需要的是机器速度的安全响应，而不是人类速度，也不是像警报那样——

Because look, at some point, customers are going to need security at machine speed, not at human speed, not at like an alarm-

……有人进去查看一下。

... someone goes in, look at it.

所以我们正在加紧努力，帮客户构建这样的解决方案，保护他们的环境。

And so that, we're running fast to go help build that for customers to help them protect their environments.

而且我对Continuum团队在这方面的工作也感到非常兴奋。

And I'm very excited about what the Continuum team is building on that front too.

那么在AWS内部——

So within AWS itself-

……

...

代理在更广泛的——

what's the state of usage of agents in the broader-

……模型中，使用情况是怎样的？

... model?

Continuum基本上就是我们尝试把内部的做法对外展示出来。

Continuum is basically us trying to expose what we do internally.

好的。

Okay.

所以我们广泛使用AI来保障我们自己的安全。

And so we use AI extensively for our own security.

我们也广泛使用AI来辅助我们自己的软件开发。

We use AI extensively for our own software development.

我们使用智能体。

We use agents.

实际上，有一件非常酷的事情是，我们在整个业务中都使用了智能体。

Actually, one of the things that's really cool is we use agents across our entire business.

我们把Amazon Quick推广给了每一位亚马逊员工，现在我看到人力资源团队在构建智能体，用来处理那些过去需要团队花几周才能完成的工作，现在一个人几小时就能搞定，比如团队规划和资源管理。

And so we rolled out Amazon Quick to every single Amazon employee, and now I see HR teams building agents to help drive what used to take teams of people weeks to do that a single person can now do in a couple of hours to think about team planning and-

……以及资源管理。

... and resource management.

我的财务团队也在构建智能体，用来从各处收集税务规则，确保我们在很多方面都合规。看到过去被软件开发人员卡住的事情，现在业务人员能够自己解决并更快地创新，这真的很棒。

I have finance teams that are building agents to go think about how do they go pull tax rules from everywhere and ensure that we have compliance on a bunch of different pieces and super cool to see that things that used to be blocked by software developers-

……实际上，业务人员能够自己解决问题，更快地创新。

... actually the line of business folks are able to go and unblock themselves and innovate more quickly.

所以Quick取得了巨大的成功。

And so Quick has been an enormous blow.

而且它像野火一样迅速蔓延。

And that has grown like wildfire.

我们看到从小型初创公司到全球最大的企业，都在使用它，并将其推广到整个客户群，以便能够访问所有企业数据，轻松应用代理和功能来加速工作。

We see customers like small startups using it all the way to the largest enterprises in the world, rolling it out to their entire customer base to get the benefit of kind of being able to access all of your enterprise data and easily apply agents and capabilities to help you accelerate your jobs.

所以，就像我说的，我们在从软件开发到安全，再到人力资源政策等各个方面都在使用它。

And so, like I said, we use it across everything from software development to security to driving HR policies.

你们以衡量一切而闻名，

You guys are notorious for measuring everything

关于你们的运营。

... about your operation.

你们在哪些方面看到了最大的收益？

Where have you seen the biggest gains?

是的，这个——

Yeah, it's-

显然，软件开发方面——

Obviously software development-

你知道答案的，就是软件开发。

You know the answer, software development.

真正的答案就是——

That's the real answer is-

但只是它的衍生方面。

But just the derivatives of that.

对，没错。

But yeah.

软件开发的速度，实际上整个产品开发，不仅仅是编码，而是——

The speed of software development has been, and really product development as a whole, not just coding, but just-

所以你看到了实际的——

So you've seen actual-

当然。

Absolutely.

新产品的推出速度是不是真的加快了？

... pick up in the velocity of new products?

我们现在推出新产品的速度，跟过去相比，简直是天壤之别。

The pace at which we're deploying new products is massively different than it has been in the past.

我觉得这一点从AWS就能看出来，它一直以快速推出新功能而闻名。

I think you can see this where AWS has always been known for rolling out features really quickly.

对，那当然。

Yeah, of course.

而且最近一年左右，我们更是加足了马力，因为我们组建了所谓的“前沿团队”，他们专注于智能体开发，而不是传统的开发方式。

And we've seen a turbo boost on that in the last year or so as we've get, we call them frontier teams, as they think about agentic development as opposed to kind of traditional development.

这不只是代码补全。

And it's not code completion.

这真的是智能体优先。

It really is agent first.

所有代码都是由智能体来写的。

The agents write all of the code.

你只需要管理一个代理团队，推动他们前进。

You're just managing a team of agents and driving that.

看到为客户带来的创新速度如此之快，真是令人兴奋，而且必须如此，因为这就是——

And it's been fun to see the pace at which innovations for customers has been massive, and it has to be, because that's the-

这就是承诺。

That's the promise

我们的客户对新功能几乎有着无止境的渴望，而且——

... our customers out there have an almost insatiable appetite for new capabilities, and-

这就是我们必须做到的。

... that's what we've got to do.

那么从组织层面来看，你对组织应该如何变革有什么见解吗？

So organizationally, do you have any insights on how organizations should change and how-

我没有。

I don't.

我想说的是——

I'll say that-

是智能体管人，还是人管智能体？

Should agents manage people, people manage agents?

会有——

There's going to be-

两种都有？

Both ways?

在很长一段时间内，会有很多人参与其中。

... there's going to be lots of people for a long period of time.

我确实认为组织会发生变化。

I do think organizations will change.

我还没有找到完美的答案。

I don't know the magic answer yet.

但我们正在积极思考这个问题。

But we're actively thinking about it.

那你们内部是在尝试各种实验，还是怎么做的？

But inside you're just trying various experiments, or what are you doing?