AWS CEO Garman：AI 需求有望推动未来两年增长逾 40%

简要说明Brief Description

AWS 首席执行官 Matt Garman 讨论了云业务的预期增长、客户承诺、企业级 AI 的应用、基础设施限制、Trainium，以及基础设施扩建的融资问题。AWS CEO Matt Garman discusses projected cloud growth, customer commitments, enterprise AI adoption, infrastructure constraints, Trainium and financing the buildout.

目录Table of Contents

• 增长预测与客户承诺
  Growth forecasts and customer commitments
• 企业 AI 采用与客户多样性
  Enterprise AI adoption and customer diversity
• 基础设施与 Trainium
  Infrastructure and Trainium
• 为建设扩张提供融资
  Financing the buildout

增长预测与客户承诺Growth forecasts and customer commitments

我们来和亚马逊云科技（AWS）首席执行官马特·加曼聊聊。马特，很高兴你再次来到《蓝鸟》节目。谢谢邀请。安迪刚才谈到，AWS这条业务线本身在五到十年内就有望成为一家营收达一万亿美元的企业？Let's talk through it with the CEO of Amazon Web Services, Matt Garman. Matt, it's great to have you back with us, on Blue Bird. Thanks for having me. Andy was talking about this path to AWS in and of itself being a $1,000,000,000,000 revenue business in five to ten years?

我知道有人问过他这个问题，但我还是想请你带我们梳理一下你是怎么得出这个结论的，以及接下来有什么路线图。嗯，我想……你看。I know that he was asked, but I'd invite you to walk us through how you get there, what the road map is from here. Yeah. Well, I think... Look.

如果看看我们今天的业务，我们的年化营收规模已达1840亿美元，增长率为39%，而且过去六个季度，每个季度的增长速度都在加快。所以，这确实非常罕见。事实上，对于一家规模如此庞大的公司或业务来说，还能以39%的速度增长，这简直闻所未闻。展望未来，结合我们看到的需求以及客户反馈，我们预计未来两年增长率也将达到40%以上。If you look at the business that we have today, we're a $184,000,000,000 run rate business that's growing 39%, and we've accelerated every single quarter for the last six quarters. And so it's it's really unusual. In fact, it's it's it's very unheard of for a company or a business of that size to be growing at 39%. And as we look forward to the demand we're seeing and what we hear from customers, we see 40% plus growth rates for the next two years as well.

随着你们开始在这个规模已经相当庞大的业务上持续累积增长，它很快就会变得非常庞大。这是我们非常期待的事情，对亚马逊、AWS 以及我们的客户来说，也蕴含着巨大的潜力。我觉得刚才提到的内容特别有意思的是，据我所知，亚马逊过去一直不太愿意预测 AWS 更长远的增长。你们现在却表示，2027 年和 2028 年都会增长 40%。是什么让你们有信心做出这样的预测？And so as you start to compound on that really large business, that that that gets really big really quickly, and it's it's something that we're quite excited about and, and a really big potential for, for Amazon and for AWS and for our customers. I think what was so interesting about what was said is, in my experience, Amazon's been reluctant to forecast AWS growth, you know, that further out and further afield. So you're saying 40% growth in both 2027 and 2028. What gave you confidence to say that?

你最密切关注的是哪些数据点？是的，这是个好问题。我们看到，越来越多大型企业基于 AWS 构建业务，我们也就能更清楚地了解这些企业未来一到五年的业务情况。What are the data points that you're looking at most closely? Yeah. It's it's a good question. And what we're seeing is as we get more and more really large enterprises building on top of AWS, we get better visibility into what that business looks like for the next year, two, three, four, five.

随着这项业务和客户基础开始增长，我们获得的承诺也越来越多。正如你在我们的积压订单中看到的那样，其规模已达到 5800 亿美元。这意味着客户承诺在 AWS 上支出这么多，而这个数字每个季度都在快速增长。因此，我们正在投入大量产能和资本来发展这项业务，以便为所有客户提供足够的产能。我们的目标是管理好这些产能，确保我们确实拥有所需的产能。And so as that business and that base started to grow and we get more and more commitments, if you see in our backlog, it's grown to be $580,000,000,000. So we have that much committed to spend on AWS, and it's growing really rapidly every single quarter. And so for us, we're investing lots of capacity, lots of, capital to grow that business so that we have capacity for all of our customers. And so our goal is to manage that capacity and make sure that we have that.

我们投入了大量资源，也为客户提供了充足的产能。事实上，我们与那些都在 AWS 上构建业务的大型企业和初创公司交流时，希望确保并让他们确信，随着他们看到眼前这个巨大的机遇，他们有足够的能力实现增长。因此，这些因素结合起来，让我们对业务增长充满信心。We're investing a lot, and we have lots of capacity for our customers. In fact, as we talk to our enterprises and startups that are all building on AWS, we wanna be sure and give them assurance that they have capacity to grow as they see this big opportunity in front of them. And so the combination of those really gives us confidence in the business growing.

另外，Matt，澄清一下，你刚才用了“已承诺”这个词，但你的意思是，这反映的是积压订单，也就是已签约的需求，而不是预期需求吗？其中很大一部分是已签约的。当然不是全部。我们看到很多初创公司不断涌入。And, Matt, just to be clear, you used the word committed, but is the idea that this is reflective of the backlog, this is contracted demand rather than anticipated demand? A lot of it is contracted. Not all of it, of course. We see lots of startups coming in.

我们看到客户在不断增长，同时也看到了这批客户。因此，我们通常会看到这样的情况：客户会加入我们，并对我们作出重大承诺，说我们就是他们的云平台。客户可能是 ADP，也可能是 Roblox——他们刚才还和你一起参加了节目——他们正在了解自己的需求情况，以及 AI 正在推动什么。We see customers growing, but we see this base of customers. And so the typical pattern that we see is a customer will come in and make a big commitment to us and say, we're their cloud platform. And it might be customers like ADP. It might be customers like actually Roblox who was just on earlier with you that are seeing what their demand looks like and seeing what AI is driving.

这样我们就能真正获得飞轮效应，既了解他们的基线增长，也了解他们的潜在增长。然后我们会把所有这些客户都纳入其中，无论他们是企业、初创公司，还是 AI 实验室。And we get really that flywheel effect where we understand what their baseline growth is as well as what their potential growth is. And we kinda layer in all of those customers, whether they're enterprises, whether they're startups, whether they're AI labs.

然后，我们确保所有这些方面都具备相应的产能并能持续增长，这样就能清楚地了解未来几年会发生什么。And then we make sure that we have capacity and growth for all of those, and that gives us good visibility into what's happening for the next couple of years.

企业 AI 采用与客户多样性Enterprise AI adoption and customer diversity

Matt，Andy 刚才这么说过，我也听你这么说过，而且他特别强调，现在还处于 AI 发展的早期阶段。我们才刚刚起步。这还只是早期。Matt, Andy was saying, and I've heard you say, and he really emphasized, this is the early days of AI. We're in the early innings. This is early.

嗯。而且这多少有点悖论：到底还有什么事尚未发生，才能让我们一直说现在还处于 AI 发展的早期？是啊，这特别有意思，因为我们从未见过一家公司发展得这么快。Mhmm. And while it's sort of a bit of a paradox, what is it that hasn't happened yet that we can keep saying we're in the early days of AI? Yeah. It's it's super interesting because we've never seen a business grow this fast.

我们从未见过一种技术发展得如此迅速。对于我们这些身处其中、能够看到这项技术所能做到什么的人来说，它确实正在改变客户所做的每一件事。而我们的许多客户才刚刚开始探索如何从中获得价值。We've never seen a technology grow this fast. And for those of us that are on the inside and can see what the technology is capable of, it really is transformative across every single thing that our customers are doing. And many of our customers are just in the early stages of really figuring out how they get value out of this.

我每天都会花时间与客户、大型企业和初创公司交流，他们正看到 AI 如何彻底改变这些公司的业务方式、运营方式以及产品的运作方式。不过，他们目前还处于非常早期的阶段，即使只是转型了 1%、2%、3%、4% 或 5%，也已经从中获得了巨大的价值。因此，我们看到客户正踏上这样的转型之旅：业务中转型的部分将从 3% 增长到 10%，再增长到 50%。这为所有客户提升效率、加速增长带来了巨大的机遇，也进一步转化为对 AI 和云服务的更多需求。And I spend time with customers and enterprises and startups every single day, and they are seeing how they do business, how their company runs, how their products work, completely be transformed by AI inside of those companies. But they're at the very early stages, and they're getting so much value even out of the one, two, three, four, 5% that they're transforming. And, and so we see customers on that journey where 3% is gonna grow to 10% is gonna grow to 50% of their business that's transformed. And so that is a massive opportunity for better efficiency, faster growth for all of our customers, and manifests into more demand for, AI and cloud services.

我们正在彭博电视和彭博电台进行现场直播，并采访 AWS 首席执行官马特·加曼。AWS 对未来的增长前景描绘得相当乐观。We're live on Bloomberg Television and on Bloomberg Radio and speaking to the CEO of AWS, Matt Garman, and AWS gave a really strong picture of the future in terms of growth.

你刚才提出的观点确实很有意思，因为这些情况既出现在初创公司，也出现在大型企业中。对吧？大家真正想知道的是，目前从整体来看，究竟具体是谁在推动增长？所以，推动增长的只是那两家大型 AI 实验室，还是这种趋势其实更为普遍？The point that you just made is really interesting because those are across the start up and the enterprise. Right? The thing that people want to know is who specifically is driving the growth at the high level right now. And so is it just those two big AI labs, or is it more widespread than that?

是的。说实话，正是这种广泛的增长让我们充满信心。我认为，尤其对我们的业务来说，我们的客户群非常多元。因此，当然，我们与 Anthropic 和 OpenAI 这两家大型 AI 实验室都有深入合作。Yeah. And that's that's honestly what gives us confidence is that widespread of growth. And I think for our business in particular, we have a really large diversity of customers. And so we work deeply, of course, with the large AI labs, Anthropic, and OpenAI.

我们与大型科技公司合作得非常好，无论是 Meta、Salesforce、Apple，还是其他大型企业。不过，我们的客户遍布各行各业，既有银行、医疗保健公司，也有刚刚起步、在某个公寓里办公的两人初创公司。We work really well with the large technology companies, whether it's a Meta or a Salesforce or Apple or, you know, the large companies. But we have customers across the whole portfolio, whether they're banks, whether they're health care companies, whether it's a two person startup that's just getting started in a apartment somewhere.

纵观整个行业，这就是我们看到的需求。因此，我们有数万乃至数十万客户，每年在我们这里的支出都越来越多。今年，承诺在我们这里投入数十亿美元的客户，比以往任何时候都多。Matt，这类客户群体有多大？Across that whole industry, that's the demand that we see. And so we have tens, hundreds of thousands of customers that are spending more and more every year. We have more customers that have committed to billion dollar spends with us this year than we ever have before. How big is that cohort, Matt?

他们队列里的那个有多大？这是个……但它确实相当显著，而且我们看到它在不断增长，一次又一次地持续增长。所以，正是这种增长让我们真正确信它具有多样性。How big is the one they in cohort? It's a... But it's it's it's pretty significant, and we see it growing, regularly over and over again. And so that growth is what really gives us confidence that it's a diversity.

这不只是……你知道，我觉得看看其他一些正在增长的公司，它们的业务高度集中在少数几家这样的实验室上。AWS则不是这样。我们的业务遍及几乎所有你能想到的行业。It's not just... You know, I think you look at some of the others that are growing, they have really high concentration with just a couple of these labs. For AWS, that's not the case. We're spread across almost every single industry that you can imagine.

随着这些行业开始迁移到云端，并希望利用 AI，AWS 就是他们想要启程的地方。And as those industries start to move to the cloud, they wanna take advantage of AI, AWS is the place that they wanna launch.

这是因为他们想要发展。他们信任我们，信任我们的企业级安全性和可用性，也信任我们所提供的一系列技术。And it's because they wanna grow. They trust us. They trust us for enterprise security. They trust us for availability, and they trust us for the set of technologies that we have available.

这正是推动持续增长、并让我们对广泛客户群体充满信心的原因。能否针对 Anthropic 单独量化？比如说，目前这 40% 的增长中，有多少来自 Anthropic 在 AWS 上的计算支出？有意思的是……And that's really what's leading to that sustained growth and gives us confidence across a wide set of customers. Is it quantifiable specifically on Anthropic? Is this case, like, how much of that 40% growth right now is Anthropic spending on compute from AWS? The interesting thing is...

而且，你知道，你可以看他们那份申报文件，他们会在文件中披露一部分在我们这里的支出。不过我要说的是，这些支出中的大部分实际上也分摊到了客户身上，因为其中很大一部分支出是在 Bedrock 上产生的。所以实际上，是我们的客户在 Bedrock 上为 Anthropic 付费。And, you know, you can read their one, and they share some of what they spend with us in, in that filing. But I'll say that most of that spend is actually spread across customers too because a lot of that spend is in Bedrock. And so it's actually our customers that are spending on Anthropic in Bedrock.

所以对我们来说，嗯，其中很多看起来甚至像是直接来自 Anthropic，而事实也确实如此。他们是我们非常出色的合作伙伴。这些业务分布在数万到数十万名客户之中，他们都在基于 Bedrock、基于 AWS 的应用中运行推理。对我们来说，它甚至看起来……And so for us, well, a lot of that can even look like it's coming directly to Anthropic, and it is. They're a fantastic partner of ours. It's spread across tens to hundreds of thousands of customers that are all doing inference inside of their applications on top of Bedrock, on top of AWS. And for us, it even looks...

实际上，它的多样性比这些数字所显示的还要丰富。It's actually more diverse than it even looks from those numbers.

基础设施与 TrainiumInfrastructure and Trainium

你们正尽可能快地增加产能，同时努力让客户相信，在他们需要时，资源会及时到位。Matt，就在我们现在交谈之际，最大的制约因素是什么？是芯片、数据中心建设，还是电力？是吧。You are adding capacity as quickly as you can and trying to give, you know, confidence to the customers that it will be there when they need it. What is the biggest constraint today, Matt, as you and I speak now, you know, chips, getting the data center built power? Yeah.

你看……你知道，整个行业都在非常激进地加速发展，而我们也在以极快的速度建设。所以，最大的瓶颈是什么，每天都在变化。Look. It's... You know, the whole industry is ramping up really aggressively, and we are building really, really fast. And so it's it's day to day as to what the biggest bottleneck is.

我觉得你提到的那些方面，我们都在快速推进。还有……电力和数据中心，确保我们有足够的空间为客户提供服务。有些时候是 DRAM。I think we're running really fast across all of those that you mentioned. There... It's power and data centers to make sure that we have enough space to run for our customers. It's DRAM some days.

有时候是芯片，有时候是网络组件。我们为客户做的部分工作，实际上是深入供应链三层，真正弄清楚客户需要哪些组件？我们如何获得……方面的承诺？It's chips other days. It's sometimes it's network components other days. And part of what we do for our customers is actually really understand all the way down three levels into that supply chain to really understand deeply, what are the components our customers are gonna need? How do we get, commitments for...

提前多年从供应商那里采购？然后确保我们把所有流程都整合到位，以便能够非常高效地部署这些设备。我想我们最近分享过，我们每十四秒就会安装一颗加速器芯片，无论是 GPU 还是 Trainium 芯片。这是一次规模巨大、速度极快的扩张。From our suppliers multiple years out? And then ensure that we have all of the processes together so that we can land these in a really efficient way. I think we shared recently, we're installing an accelerator chip, whether a GPU or a Tranium chip every fourteen seconds. It is a massive and rapid scale up.

而且，而且今天我们的供应仍然受限，同时我们正在非常快速地为客户扩大规模。你知道，我们希望能做得更快。明年，这项业务的下一次重大演进，就是在 AWS 之外销售 Trainium。我觉得很多人误解了这在实际中是怎么运作的，也误解了我们说这话时的意思，Matt。And, and today, we're still supply constrained, and, and we're scaling really rapidly for customers. And, you know, we wish we could do it faster. The next bigger evolution of the business is to sell Trainium outside of AWS next year. I think a lot of people misunderstand how that works in practice, what we mean when we say that, Matt.

是的。但这一定是个非常艰难的选择，因为内部对使用 Trainium 的需求也确实存在。是的。它……Yeah. But it must be a really hard choice because the demand internally is also there for the use of Trainium. Yeah. It...

确实如此。这实际上也是我们内部讨论的一个话题，但我认为这对我们来说是一项重要的战略举措，对我们和客户都有利。我们与客户交流时，他们会说：“听着，我想把大部分工作负载都运行在 AWS 里面。”It's true. And that's actually a debate that we have internally, but I think it's an important strategic move for us, both for us and for our customers. When we talk to customers, they say, look. I wanna run the majority of my workloads, inside of AWS.

而 Trainium 是一个非常强大且极具成本效益的平台，能够让客户以顶尖性能运行 AI 工作负载。And Trainium is a really powerful and great cost effective platform, at the very highest performance level for customers to run their AI workloads.

但也有一些客户告诉我们，他们希望在本地部署这些东西。所以……你知道，短期内不可能所有东西都上云。有时，客户希望拥有自己的数据中心。But certain sets of customers also tell us that they wanna run these things on premise. And so there is... You know, not everything is gonna be in the cloud in the near term. Sometimes, customers wanna have their own data centers.

他们想要本地部署，而且确实想在两个地方都运行 Trainium，这样就能在 AWS、自己的数据中心或其他地方运行。所以对我们来说，你知道，我们必须恰当地权衡具体该怎么做，不过我们对这项业务相当兴奋。我觉得它会非常有意思，也能让那些真正想基于这个平台构建产品的客户如愿以偿。他们很喜欢这个训练平台。They wanna run on prem, and they actually wanna run Trainium in both locations so they can run-in AWS and in their own data centers or in other places. And so, for us, you know, we have to balance exactly what that looks like, but we're quite excited about that business. I think it's gonna be really interesting, and it's gonna enable our customers who really want to build on that platform. They love the training platform.

他们看到了它在性价比方面的优势，也希望能在许多不同的地方运行它，而我们想让这成为可能。They see the cost performance benefits of it, and they wanna run across a bunch of different places, and we wanna make that possible for them.

为建设扩张提供融资Financing the buildout

Matt，在结束我们的对话之前，10-Q报告刚刚发布，我也像大家应该做的那样通读了一遍。亚马逊安排了约800亿美元的供应商融资，用于从明年开始的技术基础设施建设。这笔钱是做什么用的？为什么需要采用这种机制？Matt, to end our conversation, the 10 q hit, and I was reading through it as one must. Amazon's arranged about $80,000,000,000 of supplier financing for technology infrastructure starting next year. What is that for? Why do you need to do that mechanism?

嗯，你看。当我们看到眼前有如此巨大的机遇时，也会寻求负责任地为其提供资金支持，以便尽可能地为客户做大规模投入。因此，我们会考虑如何规划资本支出，使其与收入相平衡。如果你看看 AWS 的商业模式，就会发现一直以来，前期都需要投入大量资本支出，然后我们再在一段时间内将其变现——服务器的周期可能是五六年，数据中心和电力设施则可能是二三十年。Well, look. When we see such a really large opportunity in front of us, we're looking for ways to also, responsibly fund that so that we can go as big as possible for our customers. And so we look to how do we kind of plan the CapEx spend, to balance with revenue. And if you look at the AWS business model, it's always been the case that there's a very large amount of CapEx that goes up front, and then we monetize that over periods of time, whether it's five or six years for servers, whether it's twenty or thirty years for data centers and power.

你知道，这就是我们一直以来采用的模式。现在，我们非常喜欢的这项业务正处于快速增长阶段，对吧？这项业务的增长率达到39%，利润率则处于30%多的高位。You know, that's the model that we've always had. And we're in such a rapid ramp up right now with a business that we really love. Right? We have a business that's growing at 39 with margins in the high 30%.

那是一项了不起的业务，我们认为在那里不断加大投资有着巨大的机会。因此，我们会审视整个合作伙伴生态以及其他方面，设法平衡部分现金流，同时我们正大举加速发展，抓住我们认为是一个巨大机遇的机会。正如你在节目开头提到的，我们认为，随着时间推移，这有望发展成一家市值达万亿美元的企业。That is an incredible business, and we see such an opportunity to invest more and more there. And so we look across our partner ecosystem and others to figure out how we can balance some of those cash flows as we're really ramping up aggressively and taking advantage of what we think is a massive opportunity that, as you mentioned at the beginning of the show, we think can lead to a trillion dollar business over time.