人工智能投资热潮：何时才能见效？

你看。

Look.

迟早你得赚钱。

At some point, you gotta make money.

你投资企业，是为了获得回报、赚到钱，但过去几年里，我们反而离这个目标越来越远了。

You make investments in a business so that you can generate returns and make money, and we've gotten further away from that over the last couple of years instead of closer to it.

这并不意味着它永远不会发生。

That doesn't mean it's never gonna happen.

只是意味着风险更高了。

It just means the stakes are higher.

欢迎收听高盛交流会的又一期节目。

Welcome to another episode of Goldman Sachs Exchanges.

我是艾莉森·纳撒恩，今天和我一起的是高盛全球研究院联席主管李乔治。

I'm Alison Nathan, and I'm here with George Lee, co head of the Goldman Sachs Global Institute.

我们两人将共同主持这一系列节目，探讨人工智能的崛起，以及它对企业、投资者和经济可能带来的影响。

Together, we're cohosting a series of episodes exploring the rise of AI and everything it could mean for companies, investors, and economies.

乔治，很高兴再次见到你。

George, great to see you again.

很高兴能来参加。

Great to be here.

这一定会很有趣，乔治，因为今天我们请来的人，在过去几年里对AI的看法跟你有很大分歧，我觉得他持的是完全不同的观点。

And this should be fun, George, because today we are talking to someone who, least in the past several years, has really disagreed a fair amount, I think, or taken a different view than you on AI.

我们的嘉宾是吉姆·科韦洛，高盛全球股票研究主管。

Our guest is Jim Covello, head of global equity research here at Goldman Sachs.

你和吉姆就这个话题已经争论过很多次了。

And, again, you've had many debates with Jim about this topic.

首先，很高兴吉姆能来。

Well, first of it's great to have Jim here.

他不仅是位好朋友，也是位非常有思想的人。

He is both a great friend and a great thinker.

虽然我们在一些AI相关的问题上看法不同，但其实我们达成共识的地方更多，这些年能持续这样的对话，真的非常愉快。

And while we differ on some matters related to AI, we actually there's much that we agree on, and it's been very fun to have this dialogue over multiple years.

当然。

So For sure.

欢迎你，吉姆。

Welcome, Jim.

对。

Yeah.

不是。

No.

很高兴能来这里。

It's great to be here.

谢谢。

Thank you.

我同意。

And I agree.

乔治身上集中了所有让我觉得高盛如此出色的东西，能和你一路走来，如今再次相聚，真的非常棒。

George is everything that makes Goldman Sachs great to me, and it's been incredible going on this journey with you, and here we are again.

我们又见面了。

Here we are again.

没错。

Exactly.

吉姆，就像我们几年前提到的，你当时对生成式人工智能持一种相当怀疑、甚至与主流观点相左的看法，特别是对这项技术的经济性提出了疑问。

So, Jim, as we mentioned a couple of years ago, you came out with what I would characterize as a pretty skeptical, somewhat out of consensus view of generative AI, and you particularly questioned the economics of the technology.

你当时对这项技术能否带来足够的回报，来证明过去几年里投入的巨额资本支出感到非常怀疑。

You had a lot of doubts about whether returns that technology would generate would ever really justify all of this CapEx you have seen pouring into the technology over the last couple of years.

那么，你觉得你哪些地方是对的？

So tears on where do you think you've been right Yeah.

哪些地方是错的？

And where have you been wrong?

是的。

Yeah.

我想先从我们错的地方说起。

So so I like to start off with where we've been wrong.

我们最近刚发布了一份新报告，开头就谈到了我们之前哪里判断错了。

And so we just published another report most recently where we started off with where we've been wrong.

我们把它叫做‘两年后的公允价值重估’，和你我之前合写的那份报告做对比。

We called it the mark to market two years later versus the report that you and I worked on together.

首先，消费者对人工智能的接受程度远超我的预期，简直太出色了。

So firstly, consumer adoption of AI has been magnificent, much greater than I expected.

乔治当初的预测完全准确。

George accurately predicted that spot on.

这一点上，我们确实判断错了。

So we've been wrong about that.

我们在原始报告里讨论过，这次报告里也再次提到，大多数消费者仍在使用免费版的AI工具。

One of the things that we talked a lot about in the original report and we talked about in this report is most consumers are still using a free version of AI.

所以，要真正触及经济层面的核心问题，我们还得把重点放在企业市场。

So really to get to the heart of the economic issue, we still really need to focus on the enterprise.

但我真的觉得，必须承认消费者端的采纳势头有多强劲，乔治在这方面有多准。

But I do really think it's important to acknowledge how great consumer adoption has been and just how accurate George has been about that.

我们在最新报告中提到的第二个错误预测是，两年前我们以为，如果大型云服务商的股价长期表现不佳，它们会削减资本支出。

The second thing we talk about in the most recent report where we've been wrong was we predicted two years ago that if the hyperscalers stocks underperformed for a significant period of time, we would expect that they would scale back on the CapEx.

它们的股价确实表现不佳，原因正是巨额的资本支出拖累了自由现金流。

And they have underperformed because of the significant investment in CapEx and the negative impact on their free cash flow.

但结果它们非但没削减资本支出，反而进一步增加了投入。

But instead of cutting the CapEx, they've actually raised the CapEx.

这让我更怀疑未来的经济模型是否成立，但现实是，尽管股价低迷，它们还是大幅提升了资本支出，这完全超出了我们的预期。

So I think that that calls into question the economics even more going forward, but the reality of it is that they've massively increased the CapEx despite the stocks underperforming, which is not what we expected.

另外，我觉得技术本身的发展速度惊人，完全印证了乔治的预测，任何讨论都必须强调并承认这一点。

And then I would add that I think the technology itself has made incredible progress, very consistent with George's predictions, and I think any conversation has to really emphasize that and acknowledge that.

话虽如此，我认为经济层面的问题依然悬而未决。

All of that said, I think the economics are still very much in question.

说真的，哪怕技术进步如此显著，我现在的经济前景判断可能比以前更悲观。

And if anything, I'm probably as or more skeptical on the economics today than I was before despite how incredible the technology is.

在乔治发表意见之前，我想先问你一个问题。

Before George weighs in, let me just ask you.

你认为为什么我们一直看到这么多资本支出？

Why do you think we have continued to see all of this CapEx?

是的。

Yep.

我觉得整个供应链的每个环节都存在严重的错失恐惧症，但这并不意味着这些投入没有道理。

I think there's a tremendous amount of FOMO at every level of the supply chain, and it doesn't mean that it's not justified.

我只是觉得，我们现在花的钱远远超出了当前经济状况所能支撑的水平，原因在于大家都担心：万一这项技术真的爆发，找到了大量有利可图的应用场景，而你的对手已经抢先布局，你却一无所获，那该怎么办？

I just think that we're spending well in advance of where the economics are right now, and I think it's because everybody is afraid of what happens if the technology really takes off and finds significant positive economic use cases, and your competitors have that figured out and you don't.

这种心态从企业层面，到模型层，再到超大规模云服务商，无处不在。

And I think that's everything from the enterprise level to the model layer to the hyperscalers layer.

我们在报告中经常提到的一点是，所有的经济价值，全都持续地流向了半导体公司。

And one of the things that we talk a lot about in our report is all of the value, all the economic value has continued to accrue to the semiconductor companies.

半导体公司获得的经济回报简直惊人。

It's been incredible economic value that's accrued to the semiconductor companies.

报告里我们也反复强调，这种局面我们以前从没见过。

And we do talk a lot in the report that we've really never seen anything like that.

对吧？

Right?

我直接研究半导体股票长达十六年，每个周期里，只要半导体的客户业绩好，半导体股票就表现强劲。

I covered semiconductor stocks directly for sixteen years, and in every cycle, the semiconductor stocks thrive when their customers thrive.

但在这个周期里，半导体公司赚得盆满钵满，却让产业链上层的公司承受了经济压力。

Here in this cycle, the semiconductor companies are thriving at the economic expense of everybody above them in the chain.

所以，这种情况迟早得调整回来。

And so at some point, that has to rectify itself.

要么产业链上层的公司也开始赚钱，要么他们就得不得不减少对半导体的投入。

Either everybody above them in the chain needs to start to generate a profit as well, or they're gonna have to eventually scale back on the semiconductor spending.

而这正是我们这份报告的核心关注点。

And that's where we make the focus of this report.

所以，乔治，吉姆对这些经济问题的担忧有道理吗？

So, George, is Jim right to be concerned about these economics?

是的。

Yeah.

我觉得这正是我们看法一致的地方。我们最近刚发布了一份高盛全球研究院的报告，讨论了这项投资的规模，以及要实现足够回报所面临的巨大挑战。

I think this is actually one of the areas where we agree, and we published a paper recently out of the Goldman Sachs Global Institute that talked about the scale of this investment and just how high the bar will be, how high the hill is that we have to climb to generate sufficient payback.

我们的分析核心，是建立在吉姆之前的研究基础上，认为必须超越传统上对现有利润池被颠覆的思维模式。

And, you know, the nub of our analysis is pivoting off some of the work that Jim's done is that you have to move beyond the traditional notion of disruption of existing profit pools.

吉姆和他的团队对广告行业做了很棒的研究，分析了AI能多大程度上介入并改变这个领域。

Jim and his team did a great piece about the advertising business and how much of that could be intervened by AI players, etcetera.

如果你一个利润池一个利润池地算，把所有机会加起来，还是远远不够覆盖我们预计的七到八万亿美元投入所带来的回报。

And I think if you go profit pool by profit pool and sum up the opportunity, you'd still fall short of significant enough payoff from what we think will be 7 to $8,000,000,000,000 spent here.

但在我看来，这并不是最终目标。

Now that to me is not the end game.

我认为，这项技术真正的机遇，甚至是责任，在于创造全新的经济活动，培育全新的市场空间，开发出我们目前根本想象不到的新可能。

I think the opportunity and in fact the imperative for this technology is to help create net new economic activity, breed new TAMs, create new affordances that we can't imagine.

历史上每一次重大技术浪潮都是如此——无论是农业革命、工业革命，还是计算机革命。

And this has been the history of major technological waves, whether it's agricultural revolution, industrial revolution, computer revolution, etcetera.

不过我确实认同吉姆的观点，要实现这样的回报，前面的路确实非常艰难。

But I certainly would stipulate to Jim's point that there's a big hill to climb for this payoff.

第二点，我认为我们看法一致，只是在时间尺度上有些不同，那就是企业采用这项技术至关重要。

The second thing I think we agree on, though I think we differ in terms of timescale on this, is enterprise adoption is really important to this.

它的推广速度比我们预期的要慢，但我一直提醒自己，这项技术才发展了三年半，它本质上是全新的，和过去的技术完全不同——它是概率性的，而不是确定性的。

It has been slower than we might have hoped or expected, and yet I try to anchor back to the fact that we are three and a half and years into this is a technology that is both novel, it's paradigmatically different than old technologies because it's probabilistic versus deterministic.

在企业中部署它，需要一套全新的技术栈，还要有一整套全新的管控机制，才能确保它被负责任地、有效地、合规地使用。

There's a brand new stack of technologies required to deploy it in the enterprise and there's an entirely new set of control planes necessary to use it responsibly, effectively and compliantly.

所以，这一切都需要时间。

And so all of that just takes time.

我对这项技术重塑商业运作方式的潜力，依然充满信心。

I continue to be very optimistic about the potential for this technology to reshape the way businesses work.

只是它的普及速度，会比吉姆提到的消费者端的快速采纳慢一些。

It's just going to take a little bit longer than the avid consumer adoption that Jim referenced.

现在不是已经有这么多工具了吗？

Aren't there all these tools?

过去几个月最大的进展，就是这些智能代理工具的出现，企业开始看清了如何将这些技术融入自己的业务。

That's been the big development in the last several months, has been these agentic tools that are now enterprises are seeing the vision ahead of how to incorporate some of this technology.

这难道不让你感到乐观，觉得有希望吗，吉姆？

So doesn't that give you some optimism, some reason for hope, Jim?

是的。

Yeah.

这项技术本身非常出色。

The technology itself is terrific.

就像乔治几年前预测的那样，这项技术的进步速度真的很快。

And as George predicted a couple years ago, the pace of improvement of the technology is terrific.

但这些技术的经济成本依然非常严峻。

The economics of those same technologies is still really challenging.

这并不意味着它们永远都会这么难，但事实上，很多公司现在用这项技术花的钱，比两年前还多。

Now it doesn't mean they're always gonna be challenging, but in a lot of ways, companies are losing more money today implementing this technology than they were two years ago.

所以，正如乔治准确指出的，眼下要克服的困难比以往更大了——我们投入了更多钱，而且围绕这些工具还有太多其他问题。

So as George accurately described, the hell that has to get climbed is even steeper today than it was before because we spent more money, and there's a lot of things that go along with just the tools.

我们在报告里经常提到一点，就是现在已经有非常出色的智能代理了。

You know, one of the things that we talk a lot about in the report is there are agents today that are terrific.

现在有些模型表现得非常出色。

There are models today that are terrific.

但在很多情况下，数据其实还没准备好让代理来处理。

But in some cases, the data in a lot of cases, actually, the data isn't ready to be agented yet.

所以我们把代理直接搭在了还没准备好的数据上面。

So we're we're putting agents on top of data that isn't ready to be agented.

这又给企业带来了新的经济压力。

And that's creating another economic challenge for companies.

这些问题是能解决的，但我觉得，市场对这些技术热情高涨的时候，我们常常忽略了那些基础却关键的工作。

Again, all of these things can get addressed, but I think sometimes in the euphoria of the market talking about these technologies, we lose sight of some of the blocking and tackling things that needs to happen.

数据管理方面存在问题。

There's data management issues.

模型优化或者编排层也有不少问题。

There's model optimization or orchestration layer issues.

SLM和LLM之间的关系，这个问题我和乔治已经讨论了三年，现在依然很关键。

There's still very much the question of the SLM versus LLM dynamic that George and I have been talking about for three years now.

所以在经济层面，还有很多问题需要解决。

So there's a lot of things that need to be addressed on the economic side of things.

我想说的是，代理这个概念和我之前提到的主题很相似。

And I would just say, I think the agent thing is similar to my last theme.

它甚至比一般的生成式AI还要新颖。

It's even more novel than generative AI generally.

推理范式大概有一年半到两年的历史，而代理真正大规模兴起是在过去一年到一年半之间。

The reasoning paradigm is a year and a half, two years old, agents really became more prolific in the last year to year and a half.

可以说，代理式编程真正实现产品与市场的契合，其实只是去年年底的事。

You could argue that really the takeoff and product market fit for agentic coding really only occurred at the end of last year.

所以，虽然发展速度很快，但这些技术本身非常复杂。

And so again, it's a fast takeoff, but these are complex technologies.

要构建出吉姆提到的那种控制层和协调机制，需要投入大量工作。

It takes a lot to create that control plane, that orchestration that Jim is referencing.

这些代理功能强大，因此必须谨慎、周全地设置安全边界，这本身就是一个巨大的挑战，但同时也蕴含着巨大的潜力。

And these agents are so powerful that the need to guardrail them carefully and thoughtfully, that's its own hill to climb here that, again, I think holds massive potential.

前几天我查了下，如果把各家独立模型公司的公开营收数据加起来，再跟云计算时代达到同样营收水平所花的时间对比，你会发现，模型公司只用了三年就实现了相当可观的收入。

One thing I looked up the other day is if you take the public pronouncements of what the independent model companies are generating in terms of their revenue and sum that up and you compare it to how fast we got to that level of revenue in the cloud era, it's like, hey, we've gotten a significant amount of revenue in three years from the model companies.

而云服务公司整体花了一共十五到十七年才走到这一步。

It took something like fifteen, sixteen, seventeen years for the cloud companies in aggregate to get there.

虽然还处于早期阶段，但增长速度真的很快，我觉得编程领域的市场契合度特别值得重视。

So early days, but very fast takeoff and I think the product market fit in coding is something we should take account of.

我们还应该注意到，编程可能是这项技术最适合的应用场景。

We also should note that coding is probably the best application for this technology.

因为编程是一个可以明确验证的领域。

It is a verifiable domain.

把编程领域这种市场契合度扩展到其他领域，可能会更困难一些，需要更多科学和艺术的结合。

Having the kind of product market fit we have in coding expand to other domains may be just a little bit more challenging, require a little bit more science and art.

听好了。

Look.

这些公司是美国企业史上收入增长最快的公司，没有之一。

These are the fastest growing companies in the history of corporate America, right, on the top line.

不过话说回来，目前所有的经济收益和利润都流向了半导体公司，但从营收增速来看，这些公司确实是历史上增长最快的。

Now, again, the economics, the profits are all flowing to the semiconductor companies, but from a top line perspective, these are the fastest growing companies ever.

对。

Yep.

吉姆，我正好有个问题想问你。

So, Jim, I actually have a question on that.

我觉得你的报告特别有意思，尤其是关于基础架构提供商获得的巨额收益这部分。

Your report, I think, was super interesting in terms of the outsized economics going to the foundational infrastructure providers here.

这不一直都是这样吗？

Isn't it always thus?

我记得我们俩都经历过互联网时代，那时候思科、太阳、甲骨文、英特尔都是大赢家，持续了挺长时间。

I think back to the Internet era that we both lived through, Cisco, Sun, Oracle, Intel were big beneficiaries for quite a while.

那次和现在有什么不同吗？

Was that different than this?

也许只是半导体这一层特别突出

And maybe it's just the semiconductor layer in particular

这是产业链上游的部分。

that's side of the share.

尤其是半导体这一层，确实是独特的。

The semiconductor layer in particular that that is unique.

我们几年前发布报告的时候，艾莉森，当时我们就说应该关注卖铲子的公司，因为经济价值最早会集中在这些地方。

I mean, obviously and when we published, Alison, when we published the report a couple years ago, we said at the time we should focus on the picks and shovels because that's where we think the economic value is gonna accrue in the beginning.

事实确实如此。

That did play out.

但迟早这种局面得改变。

At some point, though, that has to shift.

不能只有这些公司获利。

It can't only be those companies that benefit.

特别是半导体公司。

And, again, particularly the semiconductor companies.

当产业链上游的利润已经高得惊人时，总得有个平衡点，情况也该反过来才对。

And when most of the losses upstream in the chain are pretty magnificent, and at some point, that that has to get flipped on its head.

在我看来，未来五年我们每天都可以讨论这个话题，但我认为归根结底就一件事。

In my mind, we can and will talk about this every day for the next five years, but I really think it all boils down to one thing.

企业用人工智能是赚钱了，还是省钱了？

Do the enterprises make or save money implementing AI?

如果答案是肯定的，这项技术才算真正兑现了它的承诺。

If they do, this technology is gonna fulfill its promise.

如果两年后我们还在争论‘还早着呢’，那可能就出问题了。

If we're having the same debate two years from now and we're still saying, well, it's early, then we might have a challenge.

因为什么时候，短期才会变成长期呢？

Because at some point, when does the short term become the long term?

对吧？

Right?

我们当然可以说，企业初期用这项技术可能会亏钱，但总不能一直亏下去。

Because we can say in the short term, companies can lose money implementing the technology, but that can't happen forever.

而且我们之前也聊过这个，乔治，你和我。

And we're also and we talked about this, George, you and I before.

我们现在正经历历史上最强劲的牛市之一，美国企业的利润普遍表现优异，整体经济环境虽然有些挑战，但依然非常不错。

We're in one of the greatest bull markets in history, and we're in a great spot of corporate America profits in general, a great overall economic environment with some challenges, but a great economic environment.

所以在这种环境下，企业长期投资是完全正确的做法——当然，我们都希望这种情况别发生。

And so it's the right thing for companies to do to be making investments for the long term in this kind of environment if we and we all hope this doesn't happen.

但如果市场或经济真的陷入低迷，那些在好时候还能放手花钱的行为，可能会适当收紧一些。

But if we did hit a more of a rough patch in the markets or the economy, some of this spending that the market is okay doing in a good environment might get scaled back a little bit.

不会。

No.

我觉得这也不一定是坏事。

I don't know that would be a bad thing.

对吧？

Right?

在这份报告里，我多次提到，相比半导体股票，我更看好现在的超大规模云服务商股票，这跟两年前的看法正好相反。

Like, I've in this report, I talk a lot about how I would favor the hyperscaler stocks today over the semiconductor stocks, which is a change from two years ago.

原因在于，未来几年可能出现三种情景，我认为在这三种情景里的两种中，超大规模云服务商的股票都会表现得比半导体股票更好。

And the reason for that is if there's three scenarios that you could see playing out over the next couple years, I think the hyperscalers stocks will outperform the semiconductor stocks in two of those three scenarios.

第一种情况是企业开始盈利。

One is corporate start making a profit.

企业开始从这些投资中获得回报。

Enterprises start generating a return on this investment.

那样的话，经济价值就会在整个产业链中传递，而不仅仅是半导体公司受益。

Then I think that there's economic value to flow throughout the whole chain, not just the semiconductor companies.

我认为云服务商的股票会因此受益，因为市场对这一点还存有疑虑。

And I think the hyperscaler stocks would get rewarded because there's doubt on that.

这些股票的表现一直远远落后于大盘。

Those stocks have underperformed the market by so much.

第二种情况是云服务商说：我们可能需要适当控制一下支出。

The second is that the hyperscaler companies say, look.

我们并不是说企业的资本支出会从每年两千亿美元直接降到零。

We might need to moderate the spending a little bit.

我们并不是说企业的资本支出会从每年两千亿美元直接降到零。

And we're not talking about companies going from 200,000,000,000 a year in CapEx to zero.

我们说的是资本支出节奏稍微放缓一点，好让他们能逐步回收一些自由现金流。

We're talking about slight moderation of the CapEx pace so that they can start to claw back some free cash flow.

在这种情况下，我觉得超大规模云服务商的股票也会跑赢半导体股票。

And I think the hyperscalers stocks would outperform semi stocks in that scenario as well.

唯一能让半导体股票继续表现更好的情况，就是现状持续下去——整个链条里只有半导体公司赚钱。

The one scenario where the semi stocks continue to outperform is the status quo persists where the only companies making money in the chain are the semiconductor companies.

说实话，这种情况持续的时间比我当初预想的要长得多，但不可能永远这样下去。

Again, this has gone on longer than I thought to begin with, but that can't go on forever.

我们不可能让链条上游这么多优秀的企业一直亏钱。

We can't have all these incredible companies upstream in the chain losing money.

因为，就算拿互联网行业打比方，最终上游最出色的企业也还是赚到了钱。

Because, again, even in your Internet analogy, eventually, the best companies upstream started to make money.

完全没错。

So Totally.

我知道现在市场并不是这么看的。

I know that's not how the market is positioned right now.

市场对半导体板块极度看多，对下游所有板块则非常看空。

The positioning of the market is incredibly bullish on semis and bearish on everything below that.

迟早有一天，这种局面得反转。

At some point, that has to turn around.

两年前你说的是十八个月到两年，现在回头看还真是这样。

I would say that at two years ago, you said eighteen months, two years.

对。

Yeah.

可现在时间又过了两年，你却还在说还要再等两年。

And now we're two years forward, and you're talking about another two years.

对。

Yeah.

看来市场对这个预期还挺有信心的。

So the market seems to have some faith in this.

哦，市场就是这么看的。

Oh, the market look.

这又是我们这一代人见过最牛的牛市了。

We're again, this is the best bull market of our lifetime.

这一点毫无疑问。

I I don't think there's any question about that.

这背后是有原因的。

And this is a reason.

有点儿像是个循环。

It's a little bit of a circular.

这项技术正是推动牛市的原因，而这个牛市又反过来为大量投资提供了动力。

This technology is the reason for it, and then that bull market is powering a lot of the investment that's happening.

我觉得迟早有一天，短期的预期会变成长期的现实，但市场现在也给了足够多的耐心，这很合理，因为它看到的是技术上的真实进展，觉得利润迟早会跟上来。

I think at some point, again, the short term has to become the long term, but the market's giving a long leash as well it should because, again, I think the market's looking at the progress on the technology and saying, well, eventually, the profits will flow through.

这种情况很可能会发生。

That may well happen.

我只是希望这一天能早点到来。

I would just like to see that happen at some point.

吉姆和我经常讨论的一个回报问题，我觉得特别值得思考的是：即使你像我一样，极度相信这项技术的力量，企业在应用它时获得的优势也可能只是暂时的。

One of the returns issues that Jim and I have talked a lot about that I think is fascinating to consider here is to what extent, even if you are an extreme believer in the power of this technology like I am, there's a potential in the enterprise that the advantage you gain from deploying it is somewhat fleeting.

换句话说，你如果率先部署这项技术，比如降低工程成本、提升流程效率，确实能赢得一点时间上的优势，但最终同行都会跟上，那时利润空间会不会被竞争抹平？

In other words, there's a chance that you can create some temporal advantage by being the first person on the block to effectively deploy the technology to lower your engineering costs or increase your straight through processing times, whatever, but ultimately everyone in your sector is gonna catch up and will the margin advantages get competed away?

其次，就像历史上很多技术变革一样，我们创造的超额收益，会不会最终都流入了消费者的口袋？

And then second, like many technology evolutions we've seen in history, will the surplus we generate vanish into consumers' pockets in a way?

所以，关于技术的价值捕获和投资回报率的衡量——这本来就是个极其复杂的话题——我认为这是最引人入胜的维度之一。因为从直觉上看，企业部署这项技术似乎理所当然、逻辑清晰，也确实感觉能带来成本优势。

And so this value capture and then measurement of ROI, which is a super complex topic around technology, I think is one of the most fascinating dimensions of this where, in a way, companies are and will deploy this technology because it seems so intuitive, logical, and it feels like it's creating savings advantage, all of that.

但要精确衡量这些无形的收益，往往非常困难。

And yet the measurement of that intangible hard terms is sometimes very elusive.

这也加剧了吉姆提到的不确定性。

And so this adds to the uncertainty that Jim's talking about.

你说得实在太到位了。

That's so incredibly well said.

我一直特别欣赏和乔治讨论的原因，就是他对这个问题的平衡态度。

And one of the things I've always loved about having these discussions with George is he's so balanced on this.

对吧？

Right?

你想想这些问题，这些投入现在是不是已经成了做生意的基本成本？

You look at some of the questions around this is, are these expenditures just the cost of doing business now?

它并不会带来特别大的优势。

And it doesn't really give you a big advantage.

只是你做生意必须付出的代价。

It's just the cost of being in business.

然后问题就变成了规模。

And then it becomes an issue of scale.

现在，规模是不是比以往任何时候都更关键？

Is scale really what matters in business now more than ever?

规模一直是个巨大的优势。

I mean, scale's always been a big advantage.

但当这些新模型出现时，你某天突然发现，作为一家企业，你得花五千万美元去买一个根本不在原计划里的新模型。

But when you have some of these models come out and all of a sudden you wake up one day and you have to spend $50,000,000 as an enterprise on a new model that you didn't even know that wasn't part of the original plans.

这并不能直接带来收入。

And that's not really revenue generating.

这更像是防御性的，而不是进攻性的。

That's almost more defense than offense.

现在能负担得起这种投入的公司越来越少。

You know, there are fewer and fewer companies that can afford to do that.

所以在这种环境下，规模优势变得极其关键。

So scale becomes an absolutely massive advantage in this environment.

完全正确。

Absolutely right.

另一方面，我们投入这么多，利润空间却被竞争蚕食；但如果不投，就会永久落后于对手——这正是推动企业持续花钱的一种博弈心态。

I mean, the other interesting side of the coin of, gosh, will we make these investments and while our margin advantage get competed away, the other side of that coin is if I don't make these investments, I'm at a permanent margin disadvantage to my and this is one of the game theoretical things that I think continues to drive spending.

另一个重要话题，吉姆，是这对就业的影响。

The other big topic, of course, Jim, is the implication on employment.

你对华尔街或管理层如何看待这项技术对员工生产力的影响，发表过一些很有意思的看法。

And you've made some pretty interesting comments about Wall Street or management's perception, C Suite's perception of the impact of this technology on worker productivity

是的。

Yep.

但员工实际的体验和科技公司所说的却不一样。

Versus what workers are actually experiencing and what the tech companies are saying.

是的。

Yep.

你能不能聊聊这种认知偏差或差距？

Talk to us a little bit about that misperception or gap in perception Yeah.

你所服务的正是这种差距。

That you're serving.

几乎所有第三方调查都发现，高管层对技术影响的预期，和一线员工对技术实际影响的看法之间，存在巨大差异。

Pretty much every third party survey that has been done finds this big gap between the c suite expectations around the impact of the technology versus the line workers' view on the actual impact that the technology is having.

这其中涉及很多方面。

Now there's a lot of different parts to that.

对吧？

Right?

这里面涉及的因素太多了。

There's so many different components.

几乎每一份调查都提到同样的问题：一线员工从技术中获得的收益，远不如高管们预期的那么多。

Now every survey pretty much says the same thing, which is the line workers aren't getting as much benefit from it as the c suite expected.

每一份调查和分析都得结合背景来看，显而易见，高管们之所以投资，是因为他们对这些投入持乐观态度。

Now every survey, every analysis has to be taken in context, and obviously the c suite is bullish on their investments or they wouldn't be making the investments.

而一线员工有时会担心，这项技术可能会取代他们的工作。

And the line workers in some ways are scared that this technology could replace what they're doing.

所以我们看待这些调查时，得把这一点考虑进去。

So we have to take these surveys with that as context.

不过，所有调查都指向同一个结论：如今的实际生产效率，远低于当初投资时的预期。

However, all the surveys do say the same thing, which is there isn't as much productivity today as where the expectations were when the investments were being made.

但有些办法是可以改变这种状况的。

There's things that can change that.

再说一遍，我觉得数据层在这方面起着至关重要的作用。

Again, I think the data layer is a huge part of that.

对吧？

Right?

因为现在很多人做查询时，得到的结果要么不对，要么不完整。

Because so many times people are doing these queries today, and they're coming up with the incorrect results or incomplete results.

这很大程度上是因为企业内部的数据系统彼此不通。

And a lot of that has to do with the data across an organization not speaking with one another.

所以我觉得现在这方面确实存在明显的脱节。

So I do think there's a pretty significant disconnect today on that.

是的。

Yeah.

吉姆提到的两个问题特别有意思，比如高管和一线员工之间的分歧，反映出整个生态系统的深层矛盾：大公司有根深蒂固的工作习惯、固化流程和老旧系统，想把这种新技术硬塞进去，阻力自然很大，这可能正是吉姆所说的——一线员工觉得‘我得改掉老办法’、‘我不习惯这玩意儿’的原因。

Two issues that Jim touches on there I think are fascinating is this issue of C suite versus Lineworker illustrates a really interesting divergence in this ecosystem, which is as a big company with incumbent work habits, enshrined workflows, ways of doing things, legacy systems, etcetera, retrofitting this brand new technology has its own challenges and there's a certain drag coefficient to it that may be responsible for what Jim's talking about, like line workers going like, I have to change the way I do things, or I'm not used to this, etcetera.

而真正能看到这些技术爆发式提升效率的，往往是那些初创公司，也就是天生就为AI设计的新型企业。

Where you can see the real takeoff and productivity of these technologies is in very young companies, call them new native, AI native companies, which are built from the jump for this technology.

它们的组织架构、技术栈都是从零开始为这种技术量身打造的。我住在硅谷附近，亲眼看到在这个时代创办公司的人，能从这些技术里获得多么惊人的效率提升。

They're organizationally designed, their technology stack is designed, and living close to Silicon Valley, I get to see when you're starting a company in this era, the amount of productivity gain you can get from this technology is extraordinary.

不过，从乐观的角度看，我希望等这些大公司完成改造后，也能享受到同样的效率提升，但这还存在不确定性，时间差到底有多长也很难说。

Now, in my optimistic sense, I hope that that means that same kind of productivity leap will be ultimately available to bigger companies as we complete the retrofit, but that is an uncertainty and certainly the time gap there is uncertain.

另外，我们还没怎么聊到最近突然兴起的对人工智能的民粹式反感。

The other thing we haven't talked much about is this somewhat sudden turn towards populist resentment of AI.

比如有人在毕业典礼上被嘘，或者因为负责在某个地区部署数据中心而在家里遭人袭击，诸如此类的事。

And whether it's being, you know, seeing people boot at commencement addresses or being shot at in their homes because they're responsible for deployment of a data center in a legislative region, etcetera.

现在，人工智能在美国变得非常不受欢迎，这在其他地方可不多见，而这一点很可能正是导致这里进展缓慢的重要原因，我一直在密切关注。

We're getting to a point where this is a deeply unpopular technology, interestingly, almost uniquely in The US versus other parts And of the that may be one of the things that really contributes to slower progress here that I'm trying to keep my eye on closely.

是的。

Yeah.

这确实是个很棘手的问题。

It's a really challenging issue.

我一般会避开这些政治类话题，因为我觉得在这些争论里根本赢不了。

I try to stay away from a lot of these political kind of issues because I don't think there's any winning in these debates.

我确实和一些最近卷入这些事件的人聊过，真让人惋惜。

I have had some real conversations with some of the people that have been recently in the middle of some of those things that you just talked about, and it's a shame.

我觉得很多事情可能会在中期选举前后集中爆发，看看人们会不会按这些立场投票。

I think a lot of it's gonna potentially come to a head around the midterm elections where we're gonna see if people are gonna vote along these lines.

我把所有这些都归到更大的经济范畴里。

And I would put all of it in the broader economics bucket.

对吧？

Right?

我通常更关注企业层面的经济问题。

Like, I've tend to focus on enterprise economics.

我关心的是整体支出和回报之间的关系，回报在供应链中怎么分配，有多少能真正回到个人手里。

I think the overall expenditures versus the returns and where those returns are accruing in the supply chain, how much of it is gonna flow back to the individual.

我们能不能说，个人使用这项技术后，经济上得到了好处，足以抵消电费上涨之类的负担？

Can we make a case that individuals are benefiting economically from using the technology, and that's going to offset some of the increased electricity costs and things of that nature?

我觉得这些问题都得好好回答清楚。

I mean, I think those are all questions that need to be answered.

是的。

Yes.

我觉得总结一下，听你们两位这么说，我觉得你们的观点其实差别不大。

I think just to wrap then, as I'm hearing both of you, I mean, I don't think there is too much daylight between your views.

我觉得你们的共识是，未来几年我们还需要多观察、多学习。

I think the consensus between you is that there's just a lot more that we need to observe and learn over the coming years.

吉姆，你其实是在问，如果我们一直看不到这些进步、生产率提升和企业收入增长，到底什么时候会碰到瓶颈？

And I think, Jim, you're really just questioning at what point will we hit a wall if we aren't seeing some of these advances and productivity gains and enterprise revenues.

对。

Yeah.

听我说。

Look.

迟早得赚钱才行。

At some point, you gotta make money.

你投资企业，就是为了获得回报、赚到钱，但过去几年我们反而离这个目标越来越远了。

I mean, you make investments in a business so that you can generate returns and make money, and we've gotten further away from that over the last couple years instead of closer to it.

这不意味着它永远不会发生。

That doesn't mean it's never gonna happen.

这只不过意味着风险更高了。

It just means the stakes are higher.

就像乔治一开始说的，我们要翻越的那堵墙更高了。

And like George started out by saying, the wall we need to climb is a little higher.

这事儿绝对能成。

It can absolutely happen.

在上一份报告里，我试着从技术或准技术的角度，梳理出我认为大家该关注的重点。

And in the last report, I tried to lay out from a technical perspective or quasi technical perspective what the kinds of things I think people need to be focused on.

我觉得投资者不该盲目假设这事一定会发生。

I don't think investors should just be blindly assuming it's gonna happen.

你得在过程中不断提问，但要保持合理平衡，别动不动就说‘这永远不可能’或者‘这肯定能成’，而是该列出一些关键指标，看看现在的情况到底预示着哪种结果最有可能。

I think you need to question along the way, but question with a reasonable balance of instead of saying it's never going to happen or it's definitely going to happen, kind of laying out the markers of here's what needs to happen, you know, moment in time today to gauge the most likely outcomes.

是的。

Yeah.

这正说明了我们这场对话有多有意思，正如吉姆所说，我们还处在非常早期的阶段，这种讨论还会持续好多年。

This is illustrative of fun that we have with this dialogue, and as Jim says, we're so early that this kind of debate will continue for years to come.

我觉得我们合作的一个亮点在于，吉姆非常擅长精准地评估我们当前在这一技术上的位置。

I think one of the great things about our partnership on this is, Jim, is extremely good at trenchantly and accurately pricing the spot of where we are in this technology.

我的倾向，某种程度上也是我的角色，是往更远的未来看。

My inclination, and I suppose in some ways my role, is to look a little bit farther downfield.

这需要在技术和经济层面有一定的前瞻性，但这种平衡让我们对现状和未来方向有了更清晰的把握。

That implies a leap of technological and in some ways economic faith to do so, but that balance I think gets us to a really nice synthesis of where we are today and where we might go.

再次感谢吉姆参与我们的对话。

Thanks again, Jim, for joining us.

这次交流非常棒。

This has been great.

非常感谢你们。

Thank you so much.

一如既往，这是一次愉快的交谈。

It's been a pleasure as always.

所以，乔治，这次对话真是引人入胜。

So George, again, fascinating conversation.

每次我都这么说，但每次都是真心的。

I say that every single time, but I genuinely mean it every single time.

我觉得你很好地总结了大家的观点和立场。

And I think that you did a good job summing up the views and where you guys stand.

你这边有什么收获吗？

Any takeaways from you?

没有。

No.

真的特别有意思。

It's, like, fascinating.

每次和你一起做这个，都特别愉快。

And by the Alison, it's always such a pleasure to do this with you.

我也觉得这些讨论总是特别有启发。

And I agree these are always fascinating discussions.

这次尤其有趣，因为这延续了我和吉姆过去三年半来一直讨论这个话题的脉络，看到我们两人的想法都发生了变化，真的很棒。

This one's particularly fun because it's part of this lineal legacy of us for Jim and I for the past three, three and a half years talking about this issue, and it's really fun just to see how both of us have evolved our thinking.

我们最一致的一点，大概就是前方还有太多未解的问题，这也让讨论变得特别有意思。

And one thing we probably most agree on is the number of unanswered questions ahead of us, which makes this really exciting and fun to talk about.

这意味着我们可能还得再请吉姆回来，你知道的，再来一次。

And it means I think we're gonna have to bring Jim back on, you know, again and

没错。

again.

当然。

Absolutely.

谢谢你，乔治。

Thanks, George.

下次见。

See you next time.

你也是。

You too.

本集《高盛对话》录制于2026年5月26日，星期二。

This episode of Goldman Sachs Exchanges was recorded on Tuesday, 05/26/2026.

我是艾莉森·内森。

I'm Alison Nathan.

感谢收听。

Thanks for listening.

本文所表达的观点和看法截至发布日期，可能随时变更，且不一定代表高盛集团或其关联机构的立场。

The opinions and views expressed herein are as of the date of publication subject to change without notice and may not necessarily reflect the institutional views of Goldman Sachs or its affiliates.

本材料仅供信息参考，不构成任何投资建议，不构成高盛任何实体对采取特定行动的推荐，也不构成购买或出售任何证券或金融产品的要约或邀请。

The material provided is intended for informational purposes only and does not constitute investment advice, a recommendation from any Goldman Sachs entity to take any particular action, or an offer or solicitation to purchase or sell any securities or financial products.

本材料可能包含前瞻性陈述。

This material may contain forward looking statements.

过往表现不代表未来结果。

Past performance is not indicative of future results.

高盛及其关联机构不对本材料中陈述或信息的准确性或完整性作任何明示或暗示的保证，并对任何基于此类信息所作的决策不承担任何责任。

Neither Goldman Sachs nor any of its affiliates make any representations or warranties expressed or implied as to the accuracy or completeness of the statements or information contained herein and disclaim any liability whatsoever for reliance on such information for any purpose.

文中提及的第三方机构名称均为其所属公司的财产，此处仅用于信息说明和识别目的，不代表任何此类公司与高盛之间存在所有权或授权关系。

Each name of a third party organization mentioned is the property of the company to which it relates, is used here strictly for informational and identification purposes only, and is not used to imply any ownership or license rights between any such company and Goldman Sachs.

本文字稿仅供方便参考，可能与原始视频或音频内容存在差异，高盛不对文字稿中的任何错误承担责任。

A transcript is provided for convenience and may differ from the original video or audio Goldman Sachs is not responsible for any errors in the transcript.

未经高盛明确书面同意，任何接收方不得复制、分发、发布、全文或部分复制，或向他人披露本材料。

This material should not be copied, distributed, published, reproduced in whole or in part or disclosed by any recipient to any other person without the express written consent of Goldman Sachs.

有关本材料中提及的任何证券的研究披露信息，可通过您的高盛代表获取，或访问 www.gs.com/research/hedge.html。

Disclosures applicable to research with respect to issue if any, mentioned herein are available through your Goldman Sachs representative or at www.gs.com/research/hedge.html.

高盛不支持任何候选人或任何政党。

Goldman Sachs does not endorse any candidate or any political party.

版权所有 2026 年。

Copyright 2026.

高盛。

Goldman Sachs.

保留所有权利。

All rights reserved.