[BidClub_]
Sharp Tech · · 15 分钟

AWS、Apple与顺风期转向的挑战|Ben Thompson 的 Sharp Tech

Andrew SharpBen Thompson

YouTube
TL;DR
  • AWS的看空逻辑不是廉价AI基础设施会失败,而是智能体让能力而非成本成为瓶颈。 Andrew Sharp将Amazon的基准情景概括为:把AI视作存储或算力,即“又一个基础原语”。Ben Thompson则认为,能够替代工作的智能体真正要回答的是AI能否“真正把事情做好”,这更有利于 Nvidia/OpenAI,或Google与芯片一体化的技术栈,而不是更便宜的 Trainium 和开源模型。
  • AWS未能满足的需求,可能暴露的是战略弱点,而不只是上行空间。 Thompson称,积压订单中的客户要的是 Nvidia,而不是 Trainium;Amazon采购 Nvidia 的占比远低于其云业务份额。他怀疑,相比一个潜在竞争对手,Nvidia可能更愿意优先满足Microsoft,并“扶一把这个世界上的 CoreWeave”。
  • AWS更可能沿着“变成Azure”而非“变成Nokia”的路径恶化。 既有工作负载仍被锁定,但创业公司会选择Microsoft/OpenAI或Google;Amazon的目标则是把现有客户带入AI时代,因为“他们的数据已经在这里”——这与7或8年前的云计算格局“完全反转”。押注客户惯性让Microsoft获益良多,但并不保证Amazon的战略也能奏效。
  • Amazon和Apple可能陷入赢家诅咒:过去的优势反过来成为拒绝转向的理由。 Sharp指出,Satya Nadella直到经历多年Ballmer时代的羞辱后,Microsoft才得以转向;而前AWS负责人Andy Jassy仍在沿用同一套打法,这意味着文化与技术选项同样重要。
  • Apple强劲的季度表现,并没有消除其中长期AI风险,因为当下的成功让牺牲变得更难。 Thompson提醒,业绩中有特朗普关税下的提前拉货,以及中国政府补贴带来的价格影响;他认为,在管理层更迭前,Apple不会改变方向,任何现在指出问题的人“至少未来5年都会看起来像个傻瓜”。
  • 反方逻辑依然成立:在一批“把自己烧到毁灭”的竞争者中,Apple可能成为Sony,而这些巨头押注惯性的策略也可能奏效。 两位嘉宾都保留了“最终可能证明他们是对的”的可能性;对Apple而言,尚未解决的问题是,它究竟需要调整方向,还是只需要“执行得更好”。
摘要 · 为研究而整理的核心内容

1. 智能体让能力而非成本成为瓶颈

  • Sharp先给出Amazon的基准情景:AWS客户会像消费存储或算力一样消费AI,把它视为“又一个基础原语”。但这建立在一个前提上:AI仍只是现有范式中的一项功能,而这正是Amazon已经知道如何变现的范式。
  • 智能体意味着“替代工作岗位”,此时约束因素不是价格,而是AI能否“真正把事情做好”。Thompson认为,Amazon因此正在优先优化廉价功能,但这些功能还没有强到足以胜任工作。
  • Thompson质疑,能力是否会出现在运行开源模型的、更便宜但“没那么好”的 Trainium 芯片上,尤其是在他认为Amazon自有模型“很糟”的情况下。他更倾向于认为,能力会先在 Nvidia 搭配 OpenAI 的组合上出现,或出现在针对Google芯片调优的Google模型上。Google可以同时做到更便宜、更好,而且从一开始就是一体化的;Amazon则“有点像在等它自己找上门”。

2. Nvidia短缺暴露AWS的战略冲突

  • AWS云业务增长令人失望,外界将其辩护为需求受到产能约束;但Thompson称,积压订单要的是 Nvidia,而不是 Trainium。Amazon采购 Nvidia 的份额远低于其云业务份额,部分原因正是它偏爱自研芯片。
  • Thompson猜测,相比Microsoft,Nvidia没有太多理由优先满足一个潜在竞争对手,也没有理由去“扶一把这个世界上的 CoreWeave”。Nvidia仍在向Amazon和Google供货,但在供给稀缺的情况下,AWS只能等待,并希望产能最终充足。

3. AWS有变成旧Azure的风险

  • 下行情景更像Windows,而不是Nokia:AWS可以保住被锁定的工作负载,却变得不再重要,增长也越来越慢。新公司可能选择Microsoft/OpenAI或Google;Amazon的目标则是把现有客户带入AI时代,因为“他们的数据已经在这里”。
  • Thompson称,这与7或8年前的云计算格局“完全反转”。即使Anthropic在Amazon平台上表现不错,从A100的视角看,它也“非常昂贵”。
  • Sharp指出,押注客户惯性让Microsoft受益匪浅——“情况可能更糟”——但Amazon押注的是,其庞大的存量客户基础会把它带入下一个范式。

4. 赢家诅咒变成固化于文化中的战略

  • Sharp将这种动态称为“一半是战略,一半是心理”:Apple和Amazon假设过去的优势会延续,而陌生的自我重塑代价高昂,甚至可能超出自身能力边界。Microsoft也是在经历多年Ballmer时代的羞辱后,才有条件让Satya Nadella推动转向。
  • Amazon与后Ballmer时代的断裂不同:其CEO Andy Jassy长期担任AWS负责人,这意味着“还是同一个人在执行同一套战略”。Thompson对文化的定义是:“那些最终成为公司心理组成部分的良好决策残余”,只有真正的障碍出现后,它们才会受到挑战。
  • Thompson提到Siri事件作为证据:他说,Apple仍在试图“收拾John Gruber”,因为后者称这一结果既尴尬又糟糕;这种行为听起来既不谦逊,也不像是在积极学习。

5. Apple会先更用力执行,再考虑调整

  • 在Thompson看来,Apple在财报电话会上的信息是:“我们什么都不改。”Apple Intelligence的方向没错,只是需要更好的执行。他用NBA作比喻——“我们是调整,还是执行得更好?”——这让Sharp重新提起Thompson在第1场比赛后,将Tim Cook刻薄地比作Mike Budenholzer的说法。
  • Sharp补充称,消费者并没有因为Apple Intelligence而更换手机。Thompson指出,季度业绩依然非常强劲,但部分销售来自关税前的提前拉货,以及中国政府补贴下的定价,因此这些因素未必会重现。
  • 反方逻辑仍然成立:Apple可能是“一群把自己烧到毁灭的公司中的Sony”,两家巨头“最终可能证明自己是对的”。但Thompson认为,在Apple更换领导层之前,战略不会改变;一位中期看空者可能“至少未来5年都会看起来像个傻瓜”。
  • Thompson称,如果Apple最终陷入困境,最恰当的类比是Intel:一家公司的主导地位可以按季度、按年份延续,但长期恶化却很难被及时看见。Sharp的个人反例是,软件体验恶化后,Samsung和Android在约10年后重新成为可考虑的选择——只是“那些甜美的蓝色气泡”除外。
Andrew Sharp

And all this, it's not that it's not going to work. Yes, that makes total sense. Infrastructure is going to dominate. All these companies with workloads on AWS, you can see them sort of adopting it, just using what's there. Just like you can get storage and compute, you can get AI from Amazon. It's just another primitive.

But implicit in that talk is the assumption that AI is slotting into the current paradigm. It's a feature you add, and we're going to provide that feature at a lower cost to everybody else. Again, they might be right. It is notable, though, as you're pointing out, that it happens to perfectly align with the way Amazon functioned in the last paradigm. And what happens if the paradigm changes?

The change in the paradigm is agents. I hate using the word “agents” because it's so overused right now. Everyone's talking about it. But the idea is that instead of a person doing something, the AI is doing it. It's replacement. Job replacement is the easiest way to think about it. And what's the limiting factor in job replacement? It's not cost. It's performance.

Ben Thompson

Yes, can you actually do it well? And so the question is, Amazon is laser-focused on cost for a functionality that is not yet capable enough. So where is it more likely that the capability emerges? Is it going to emerge on Amazon's cheaper, not-as-good Trainium chips running some random open-source model because Amazon's models stink? Is it going to come out on an Nvidia stack with OpenAI, or Google with its model tuned to its chips?

By the way, Google's cheap, too. They already have the advantages Amazon has, except they're fully integrated from the get-go, and they're just better at AI than Amazon is. The problem is, the lane Amazon's going for is the lane Google is super primed to dominate—not just by being cheaper, but also by being better. And in a world where the limitation on this new paradigm is being better, Amazon's just sort of hoping it comes to them. They're not really pursuing that at all. Again, it might work out for them. But they're a lot more like Apple than it might seem.

Andrew Sharp

Exactly. I mean, they're definitely on the back foot here. Is there anything that Amazon could do, or are they potentially on the back foot if we enter a world where performance is the differentiating factor for various enterprises?

Ben Thompson

The defense people have for Amazon's earnings, which were disappointing as far as cloud growth, is that they have a huge amount of demand that's not being filled because they don't have capacity, right? The problem is: why is that demand not being filled? That demand is not for Trainium chips. That demand's for Nvidia chips. The problem is AWS can't get enough Nvidia chips.

And you have to ask, when you look at the market share of who buys from Nvidia relative to cloud market share, Amazon's is way lower compared to their cloud market share. Some of that is a function of Amazon prioritizing its own chip. I suspect a lot of it is Nvidia sitting there and saying, “Everyone wants our chips. Why should we prioritize you when your entire goal is to come compete with us?”

It's like, “We'll sell to Microsoft all the chips that they want. We'll definitely prop up the CoreWeaves of the world, who do nothing but run Nvidia. And they'll resell their capacity to Microsoft or wherever else it might be.” Amazon's also sort of hoping, betting, that Nvidia gets sufficient supply to meet all the demand, so that Nvidia's like, “Well, I guess we'll sell it to you, Amazon, after all.”

Again, Nvidia is selling to Amazon. It's not like they're completely shutting them out. They are selling to Google. But if you're in a situation where there is limited supply of this resource, which is Nvidia chips, it's no surprise Amazon has a backlog of customers who want Nvidia chips and they can't serve them.

Andrew Sharp

Yeah. Well, I would say, looking at it all from a big-picture standpoint, what I enjoyed about the article is, one, it was a nice coda to our conversation about continued big-tech dominance from the last episode we recorded. And it dovetails with some of what I said on that episode, where it seems unlikely that things are going to remain completely unchanged when we get 10 years down the road.

Granted, AWS is not going to become Nokia. Apple is not going to become Nokia. But there are ways things could really start to slide.

Ben Thompson

Right. Well, it's kind of like the Windows thing, right? Windows didn't go anywhere, but it stopped being super relevant, and it wasn't growing like it used to back in the day. Maybe that's an AWS story. People are locked in for the workloads that they have there. But in the long run, in the future, if you're a new company, AWS has become Azure.

Where it used to be that all the new companies were on AWS, Microsoft grew a nice Azure business by basically carrying along all the enterprises who realized the benefits of the cloud and trusted Microsoft. They already had their stuff, so Microsoft sort of carried them along to the cloud, and they built it up together. Microsoft had a very nice Azure cloud business, basically serving all the people that weren't hip enough to be in AWS—these old enterprises.

That's Amazon's goal for AI. It's like, “Look, if you're a new company, you're probably going with Microsoft or OpenAI, which is running on Microsoft, or you're going with Google. Maybe you're going with Anthropic, who's running on Amazon, but even there, you're getting performance and it's very expensive from an A100 perspective.”

Amazon's real goal in the whole Trainium thing is: “We have all these companies on our platform. They're going to come along to AI. Their data's already here. They'll go with us.” It's really a total role reversal from the cloud 7 or 8 years ago. Look, betting on inertia has served Microsoft fairly well.

Andrew Sharp

That's right. It's worked out very well for Microsoft. Could be worse. But it's interesting, because what you were writing about with the winner's curse is half strategy and half psychology. I understand why these companies would look backward and think, “All right, everything that served us well in the current era is going to continue serving us well.”

Also, in Apple's case, it's far more affordable for us to continue to bet on this strategy than to bet on a complete unknown that we may not be competent enough to execute on. And so it clarifies the risks that these companies are taking as they move forward.

And also, it is like you trace how companies end up going in this direction. In Microsoft's case, Microsoft was an example that bet on its strengths and lost, and ultimately had to pivot. But would Microsoft have been able to pivot its strategy under Satya Nadella without the years-long humbling that they experienced under Ballmer, trying to double down on all the Windows advantages? I would guess not.

Just from a cultural standpoint, it's really hard for them to make that commitment until they bang their head against the wall over and over again and lose. And so that's some of what I see happening at Apple. And I know less about AWS, but it seems like a similar story in that respect as well. Does that resonate with you at all?

Ben Thompson

Andy Jassy at Amazon, their CEO, was the longtime former head of AWS. For sure, it's the same guy doing the same strategy. You do see it much more starkly with Apple. The real thing with Apple's earnings call was that it was very clear: “We're not changing anything. We're just going to do better.”

I analogized it on Dithering to an NBA team—the favorite losing Game 1 of a series. It's like, “Do we adjust, or do we execute better?”

Andrew Sharp

Of all the Tim Cook criticism that you've put forth over the last year and a half, nothing was more withering and damning than comparing him to Mike Budenholzer after Game 1 of the Milwaukee Bucks series.

Ben Thompson

It was so infuriating. And there would be obvious adjustments on the board that would not be explored. So perhaps Apple will do better than the Milwaukee Bucks at some point over the next couple of years. But yeah, that's sort of what is happening.

They're saying, “Look, we had the right idea with Apple Intelligence. Now we just need to execute on it.” And perhaps they'll be right. That's the thing: we could get 5 or 10 years down the line and look back, and Apple could be the Sony amidst a bunch of different companies that spent themselves into oblivion here. That's on the table as well.

Well, no, that's the thing. That's why it's hard to write about. Both companies may end up being right, right? It's definitely possible. So we'll see what happens and how it goes. But, yeah, Tim Cook's not changing crap.

Andrew Sharp

Mhm. And maybe at this point it's for the best that he seems determined to stick around for a while. Everyone talks about how he wants to stay until they launch their glasses or whatever it is. Your Satya Nadella point is a very, very good one. It can be very clear from the outside what needs to happen. But for a very successful company to change track, it just can't be done in the midst of success.

Ben Thompson

There is a real psychological and cultural component to it. And all the things that got you to where you are are going to rule this. What is culture? Culture is the remnants of good decisions that become part of the company's psyche.

This is how we do it. You don't ask why you do it. It's just the way it's done. And those things don't actually get challenged or questioned until you hit real obstacles.

And to your point, the Siri thing was super embarrassing. At the end of the day, they're still trying to screw John Gruber because he had the gall to say it was embarrassing and bad. That's not a company that sounds very humble or is learning from what they did, right?

Andrew Sharp

Well, also, customers aren't changing phones in the wake of the Apple Intelligence thing.

Ben Thompson

Huge quarters, right? And some aspect of this is a pull-forward from the tariff stuff. They cut prices, usually in China, because of a government subsidy program. There's a bit going into this that's going to show up in future quarters as sales that don't happen.

People shouldn't get too excited about the numbers. By the way, they're still really good numbers, right? To the point, as I said before, there's the short-term, medium-term, and long-term.

The risk is, I just think you can't disagree that they face very significant medium-term and long-term questions. But is it possible for them to address those if addressing them means sacrificing the short-term when the short-term is going so well?

And it's hard to talk about because, by definition, if you say they're in trouble, you're going to look stupid for the next 5 years at least, right? When does the medium term kick in?

But, yeah, they're not going to change anything—not until there's a change in leadership. I think that's pretty clear. No, sir. And I think you nailed it a couple months ago. The Intel example would be what's apt if Apple is going to run into trouble 10 years down the line.

You're not going to be able to see it in terms of the quarter-to-quarter, year-to-year performance for a little while here because they're going to remain dominant. I will just say, as an iPhone user, I'm more open than ever to shifting, in large part because of what I talked about earlier with the core competencies. There's just different software that doesn't work as well as it used to, and I've always heard great things about Samsung and some of what's happening on Android.

So, I'm open to it, which is not where I've been for about 10 years.

Andrew Sharp

Pod content. We need you to switch stand right now.

Ben Thompson

I think I might do it just for the pod and just to stick it to Apple, which I feel has gotten complacent over the last 5 years. I want to make them pay somehow, but I don't want to give up those sweet, sweet blue bubbles. So it will still be scary, but perhaps we'll explore that possibility in the fall. For now

AWS、Apple与顺风期转向的挑战|Ben Thompson 的 Sharp Tech — 文字稿与摘要 | BidClub