人工智能的三万亿美元之问：如何支付账单？

欢迎回到《市场观点》，欢迎收看我们在科技、媒体与电信会议现场对话的第二部分。

Welcome back to thoughts on the market and welcome to part two of our conversation live from the technology media and telecom conference.

我是米歇尔·韦弗，摩根士丹利的美国主题与股票策略师。

I'm Michelle Weaver, US Thematic and Equity Strategist at Morgan Stanley.

今天我们将继续与斯蒂芬·伯德、乔什·贝尔和林赛·泰勒的对话，这次将聚焦于人工智能的融资以及该领域面临的一些风险。

Today we're continuing our conversation with Stephen Byrd, Josh Baer and Lindsay Tyler this time looking at financing AI and some of the risks to the story.

今天是3月6日星期五，上午11点。

It's Friday March 6 at eleven a.

M.

M.

在旧金山。

In San Francisco.

昨天我们讨论了人工智能的采用情况，虽然这个话题令人兴奋，但也出现了一些担忧。

So yesterday we spoke about AI adoption and while there's a lot of excitement on this theme there've also been some concerns bubbling up.

Lindsay，我想先从你开始谈谈融资问题，这是人工智能建设的另一个关键组成部分。

Lindsay I want to start with you around financing that's another critical component of the AI build out.

您最新的关于数据中心融资缺口规模的分析是什么？

What's your latest on the magnitude of the data center financing gap?

信贷市场在这里扮演了什么角色？

And what role is credit markets playing here?

是的。

Yeah.

去年夏天，我们与主题研究团队、斯蒂芬及其同事，以及固定收益研究部门合作，发布了一份关于数据中心融资缺口的报告。

In partnership with Thematic Research, Stephen and team, and colleagues across fixed income research, last summer, we did put out a note thinking about the data center financing gap.

对吧？

Right?

因此，斯蒂芬及其团队估算出未来四年全球数据中心资本支出需求为3万亿美元。

So Stephen and team modeled a $3,000,000,000,000 global data center CapEx need over a four year time frame.

因此，我们与固定收益部门 across 各类资产合作，开始思考：

So in partnership with fixed income across asset classes, we thought, okay.

这笔资金究竟该如何筹集？

How will that really be funded?

我们得出的结论是，大型云服务商，即高质量的大型云服务商，能够产生可观的现金流。

And we came to the conclusion that the hyperscalers, the high quality hyperscalers generate a good amount of cash flow.

对吧？

Right?

因此，经营活动产生的现金可以覆盖大约一半的资金需求。

So there's cash from ops that can fund approximately half of that.

但我们认为，固定收益市场对于填补剩余的资金缺口至关重要。

But then we think that fixed income markets are critical to fund the rest of the funding gap.

实际上，私人信贷是主力，同时得到公司信贷和证券化信贷的支持。

And really private credit is the leader in that and then aided by corporate credit and also securitized credit.

自那以来，我们看到私人信贷确实发挥了作用。

What we've seen since is that, yes, private credit has served a role.

私人信贷1.0和私人信贷2.0之间存在差异，前者更多是面向中型市场的直接贷款，后者则更多是资产融资（ABF）或资产支持融资。

There is this difference between private credit one point zero, which is more of that middle market direct lending, and then private credit two point zero, which is more ABF, Asset Based Finance or Asset Backed Finance.

我们在此看到的是对大型云服务商租户租赁业务的兴趣。

And what we see there is an interest in leases of hyperscaler tenants.

对吧？

Right?

在过去的九个月左右，我们还看到超大规模服务商发行了投资级债券，这显然是它们利用现金流的一种方式。

We've also seen in the market over the past nine months or so investment grade bond issuance by hyperscalers, obviously a use of cash flow by hyperscalers.

根据报告，我们还看到了与银行提供的建设贷款以及私人信贷相关的融资。

We've seen the construction loans with banks and and also private credit per reports.

我们还看到了高收益债券的发行，这在建设融资中是一种新兴趋势。

We've also seen high yield bond issuance, which is kind of a new trend for construction financing.

我们也看到了资产支持证券（ABS）和商业抵押贷款支持证券（CMBS）。

We've seen ABS and CMBS as well.

然后，一个新兴且日益受到投资者关注的新方向是更多以芯片或计算合同为支撑的融资，也就是更创新的解决方案。

And then something new that's that's emerging in in focus for investors is more of a chip backed or compute contract backed financings, like more creative solutions.

目前我们还处于支出的早期阶段，因此随着我们逐步推进建设的初期阶段，下一个关注点是：那么芯片怎么办？

We're really in early innings of the spend right now, and so there is this shift as as we start to work through the construction early phases, the next focus is, okay, but what about the chips?

嗯。

Mhmm.

因此，我认为一个重要的关注点是，如果你看一个吉瓦级别的站点，芯片的成本占比超过50%。

And so I think a big focus is that, chips are more than 50% of the spend if you're looking at a gigawatt site.

这取决于芯片的类型和代际，但这也是接下来的另一个重点。

And it depends what type of chips and kind of what generation, but that's the next leg of this too.

所以这在某种程度上

So it's kind

是2026年的关注重点。

of a focus for 2026.

当你考虑超大规模企业的资产负债表杠杆和融资时，你如何看待其债务融资的规模和时间表？

And how do you view balance sheet leverage and financing when you think about hyperscalers, debt raising, magnitude, and timelines?

所以简单来说

So just to bring

如果需要算力，你实际上可能需要两样东西，对吧？

it down to more of a basic level, if you need compute, you really might need two things, right?

一个是带电的机房空间，另一个就是芯片。

A powered shell and then the chips.

因此，如果你在寻找这种计算能力，大致有三种方式。

And so if you're looking for that compute, you could kinda go in three basic ways.

你可以自己建造机房，并且整体采购所有设备。

You could look to build the shell and kind of build and buy the whole thing.

你可以从开发商那里租赁机房，比如那些转做HBC的比特币矿场，然后自己购买芯片并自行安装。

You could lease the shell from, you know, a developer, maybe a Bitcoin miner to that is converted to HBC, and then you kind of buy the chips and you and you put them in yourselves.

或者你可以租赁全部的计算能力。

Or you could lease all the compute.

所谓的租赁。

Quote unquote lease.

这更像是一种合同安排。

It's more of a contract.

从资金角度看，如果你考虑一些大公司的现金流，它们主要会用于芯片支出。

In terms of the funding, if you think about the cash flows of some of the big companies, think of that as primarily being put towards chip spend.

如果你考虑建设成本，这笔开支会在现金资本支出和租赁之间分摊。

If you're thinking about the construction, that's kind of split between cash CapEx but also leases.

因此，我们看到，在四大或五大公司中，未来两到五年内将有超过6000亿美元的未启动租赁义务开始生效。

And so what we've seen is that there is more than $600,000,000,000 of uncommenced lease obligations that will commence over the next two to five years across the big four or five players.

然后，我的股权同行估计，今年这些公司还需要约7000亿美元的现金资本支出。

And then my equity counterparts estimate around 700,000,000,000 of cash CapEx needs this year for some of those players as well.

所以这些数字很大，但基本上，它们就是以这种方式处理一些

So these are big numbers, but that's kind of how, at a basic level, they're approaching some of

融资问题的。

the financing.

这是一种分拆的、分拆的策略。

It's it's a split a split approach.

在最近几天的会议中，关于融资你学到了什么？

And what have you learned around financing the past few days of the conference?

有什么新的信息可以分享吗？

Anything incremental to share there?

当然。

Sure.

是的

Yeah.

我认为我在本次会议上找到了一些关键主题的佐证。

I think I found confirmation of some key themes here at the conference.

首先是存在多种融资渠道。

The first being that numerous funding buckets are available.

去年我们的报告就强调过，你可以从资产层面融资、公共债券、部分股权等方面入手，这些不同的融资渠道都是可用的。

That was a big focus of our note last year is that you can kind of look at asset level financing, you can look at public bonds, you can look at some equity, there are these different funding buckets available.

其次，租户质量对建设融资至关重要。

The second is that tenant quality matters for construction financing.

我觉得在过去两到三周的市场中，我比在本次会议上更频繁地看到这一点，但这一直是交易定价和市场深度的关注焦点。

I think I've seen this more in the markets than maybe at this conference over the past two to three weeks, but that has been a focus of pricing for the deals, but also market depth for the deals.

第三个关键主题的佐证围绕着新兴云服务商以及GPU即服务的商业模式，思考这些创新的融资方式。

A third confirmation of a key theme was around the neo clouds and also the GPU as a service business models, thinking about those creative financings.

对吧？

Right?

他们是否希望从计算合作伙伴那里获得预付款？

Are they thinking about from their compute counterparties, would they like upfront payments?

如果他们有一个非常高质量的投资级交易对手，是否会考虑将融资移出资产负债表？

Might they look to move financing off balance sheet if they have a very high quality investment grade rated counterparty?

因此，这些解决方案正在经历一些演变。

So there is some of this evolution around those solutions.

第四个关键主题是信用支持。

And then a fourth key theme is just around the credit support.

斯蒂芬和我都讨论过，关于一些比特币矿工，可能存在一些高质量的投资级参与者，他们愿意提供信用支持，或作为租赁的后盾，帮助生态系统中的其他参与者获得更优惠的建设融资定价。

And Stephen has and I have talked about this around some of the Bitcoin miners is that there can be these higher quality investment grade players that might look to lend their credit support, maybe a lease backstop to other players in the ecosystem in order to get a better pricing on a construction financing.

我们已经开始看到一些媒体报道，探讨这种模式在未来芯片融资中可能如何展开。

And we are seeing some press pick up around how that might play out in in chip financing down the road too.

嗯。

Mhmm.

AI驱动的风险和潜在颠覆，一直是今年迄今为止这一主题价格走势的重要特征。

And AI driven risk and potential disruption has been a big feature of the price action we've seen year to date in this theme.

斯蒂芬，你认为哪些资产类别或企业能够抵御这种颠覆？

Stephen, what are some asset classes or businesses you see as resistant to some of this disruption?

我们经常思考那些能够抵御通缩和颠覆的资产类别，有趣的是，世界上其实有几位经济学家正在对此概念进行出色的研究，他们称之为‘变革性经济的经济学’——包括两位美国人、两位加拿大人、两位英国人以及其他几位学者，他们的研究非常深入，核心是探讨当强大的人工智能进入众多行业、引发价格下降和通缩时，经济会变成什么样？

We a lot of time thinking about sort of asset classes that are resistant to deflation and disruption and what's interesting is there's actually a handful of economists in the world that are doing remarkable work on this concept that they would call it the economics of transformative Americans, two Canadians, two Brits, a number of others who are doing really, really interesting work and essentially what they're looking at is what do economies look like as we see very powerful AI enter many industries, cause price reductions, deflation, what does that do?

他们得出了许多有趣的结论，其中之一是：那些无法被人工智能通缩的资产，其相对价值会上升。

They have a lot of interesting takeaways, but one is this idea that the relative value of assets that cannot be deflated by AI goes up.

这个想法非常简单，但你可以这样理解。

Very simple idea, but think about this way.

比如，考艾岛上只有一个主要度假胜地。

I mean, there's only one principal resort on Kauai.

金属的储量是有限的。

There's a limited amount of metals.

因此，我们列出了一份受到投资者广泛关注的资产清单，这些资产不仅具有抗性，甚至可能升值。

And so what we go through is this list that's gotten a lot of investor attention of resistant asset classes or more than resistant asset classes that can go up in value.

显而易见的包括土地，但房地产方面你需要谨慎一些，比如办公地产可能并不是理想选择，中低收入住房也不太合适，更值得关注的是工业房地产投资信托、高端房地产；此外，还有很多其他类别让我感到有趣。

So there are obvious ones like land, though you have to be a little careful with real estate in the sense that like office real estate probably wouldn't be where you would go nor would you potentially go sort of towards middle income, lower income housing but more think of industrial REITs, higher end real estate but there are a lot of other categories that are interesting to me.

各种基础设施都应该相当抗冲击，各种关键材料和金属在这种环境下表现会非常好。

All kinds of infrastructure should be quite resistant, all kinds of critical materials, metals should do extremely well in this.

但当你进一步深入时，实际上会发现一些有趣的现象。

But then when you go beyond that, it's actually kind of interesting.

我的意思是，除了土地和金属这些经典例子之外，还有一份更长的清单。

Mean, there's a longer list than those classic sort of land and metals examples.

这里的例子包括计算能力。

Examples here would be compute.

我觉得詹森说得很好，如果可用的基础设施是有限的，你当然希望把最好的计算资源投入其中，最终在某种程度上，智能将成为这个世界的新型货币，对吧？

I thought Jensen put it well, know if there's a limited amount of infrastructure available you want to put the best compute and ultimately in some ways intelligence becomes the new coin of the realm in the world, right?

所以我希望拥有智能的提供者。

So I would want to own the purveyors of intelligence.

这可能包括高端奢侈品。

It could include high end luxury.

这也可能包括独特的人类体验。

It could include unique human experiences.

所以我不知道你们中有多少人有上大学的孩子，但我的孩子正是上大学的年纪，他们非常讨厌他们所谓的AI垃圾内容。

So I don't know how many of y'all have children who are sort of college age, but my children are college age and they absolutely hate what they would call AI slop.

他们想要真正的原创人类内容，并且会主动去寻找，当看到拙劣模仿人类内容的作品时，他们简直无法忍受。

They want legit human content and they seek it out and they absolutely hate it when they see bad copies of human content.

因此，我认为在经济的许多领域中，独特的人类体验和原创人类内容仍有其立足之地，去探索这些领域在经济中可能在哪里，是很有趣的。

And so I think there is a place in many parts of the economy for unique human experiences, unique human content, and it's interesting to kind of seek out where that might be in the economy.

所以这些就是一些具有抗压性的资产的例子。

So those would be some examples of resistant assets.

乔什，软件一直是这场AI颠覆辩论的中心。

Josh, software has been really the center of this AI disruption debate.

你如何看待当前软件估值倍数的回调，与以往高度不确定时期的对比？你认为这些担忧有道理吗？你是怎么看待这个问题的？

How would you compare the current pullback in software multiples to prior periods of peak uncertainty and do you think any of these concerns are valid or how are you thinking about that?

很好的问题。

Great question.

也就是说，从秋季以来，以企业价值与销售额之比衡量的软件估值倍数已经下降了30%到35%。

Mean software multiples on the EV to sales basis are down 3035% just from the fall.

我要说的是，这在整个群体中都是如此。

I will say and that's overall in the group.

许多股票的估值倍数在过去一年中下跌了60%到70%，当前定价反映的正是极度的不确定性，以及大量的恐惧。

A lot of stocks multiple handfuls are down 6070% over the last year and what's being priced in is really peak uncertainty, a lot

的

of

而如今四倍市销率的估值，让我们回溯到了大约十年前向云计算转型的时期，这一次在许多方面让我们想起了那个充满恐惧的高峰期。

fear and these multiples now four times sales takes us all the way back about ten years to the shift to cloud and this time in many ways reminds us of that period of peak fear.

在这种情况下，市场定价所反映的是终值风险。

In this case, what's being priced in is terminal value risk.

我们昨天谈到了这个总可服务市场，但你知道谁会赢得这部分市场份额吗？

We talked about this TAM yesterday but you know who is going to win that share?

从竞争格局来看，这些模型提供商、大型语言模型以及新进入者之间，市场份额将如何分配？

How is it divided from a competitive perspective across these model providers, the LLMs with new entrants?

当然，是现有巨头，以及另一个关于内部化的问题。

Of course the incumbents, and this other idea of in housing.

因此，存在竞争风险、商业模式风险，我们的公司需要将定价模式从按座位收费转变为按使用量收费或混合模式，最后还有利润率风险。

So there's competitive risk, there's business model risk, our company's gonna need to change their pricing models from seat based to consumption or hybrid, and then last margin risk.

只是在思考更高的输入成本和更高的资本密集度。

Just thinking about the higher input costs and higher capital intensity.

所以，目前所有这些担忧都已经被反映在价格中了。

So all of those fears are being priced in right now.

当然，我们在会议上邀请了这些公司现场参与。

And we, of course, had a bunch of these companies live with us at conference.

它们是如何应对这些风险的？

How are they responding to some of these risks?

它们是如何回应投资者的担忧的？

How are they addressing these investor concerns?

我们覆盖的大多数公司都是现有的软件供应商，我认为管理团队很好地出来捍卫了它们的竞争护城河，并清晰地阐述了为什么它们处于把握这一机遇的绝佳位置。不同公司的原因可能各不相同，但共同点通常包括企业级的信任、安全、治理以及IT部门的接受度。

Most of the companies here from our coverage are the incumbent software vendors, and I think that the leadership teams did a really nice job coming out and defending their competitive moats and really articulating the story of why they are in a great position to capitalize on the opportunity And the reasons can vary across different companies, but some of the commonalities are around enterprise grade trust, security, governance, acceptance from IT organizations.

当你真正与公司和首席信息官交谈时，那种‘所有人用一键生成所有应用’的想法就会被彻底打消。

The idea of vibe coding all apps in an organization get squashed when you actually talk to companies and chief information officers.

对于一些公司来说，存在专有数据壁垒和网络效应。

For some companies, there's proprietary data moats, network effects.

所有这些都建立在现有的客户关系之上，这就是我们这些 incumbent 公司传达的信息。

All of that's on top of existing customer relationships and so that was the message from the companies that we had that we're the incumbents.

我们可以像所有这些不同的竞争领域一样使用同样的创新技术，但我们拥有这种差异化和壁垒，因此我们处于有利位置。

We get to use all of the same innovative in the same way that all these different competitive buckets do but we have that differentiation and that moat and so we're in a good place.

我想以一个积极的基调来收尾。

I want to wrap on a positive note.

斯蒂芬，你在会议上听到的最让你兴奋的是什么？

Stephen, what did you hear at the conference that you're most excited about?

我认为是生命科学领域。

I'd say the life sciences.

一些投资者指出，如今人工智能可能面临公关问题，但我确实认为，通过改善健康成果来为人类带来显著益处非常重要——即使不在这次活动之外，我前一周在印度时就看到，人工智能能够为最需要帮助的人群提供强大的医疗工具，以分布式方式发挥作用。

A few investors pointed out that perhaps AI has a PR problem these days and I do think showing a significant benefit to humanity in terms of improved health outcomes whether that's just better diagnosis away from this event but I was in India the week before and AI can have a powerful benefit to people who suffer the most in terms of providing very powerful medical tools in a distributed manner.

我非常看好这个领域。

I'm a big big fan there.

但从根本上说，治愈那些困扰人类的最具挑战性的疾病，这类问题的复杂性以及开发这些疗法的过程，恰恰非常适合人工智能。因此，从宏观角度看，这无疑是我认为最令人兴奋的领域。

But you know in many ways curing the most challenging diseases plaguing humanity, the kind of problems involved in providing those and developing those cures are perfect for AI so that for me stepping way back that is the by far the most exciting thing.

乔什，你呢？你最期待的是什么？

Josh same to you what are you most excited about?

从我的角度来看，这可能是软件行业的转折点——我们能够展现出自己正处于一个关键的拐点和加速期。实际上，我们的软件公司需要时间来开发新的AI技术，将其融入经过测试和验证的产品中，并完成企业采用周期。我们的调研数据显示，我们正站在更广泛采用的门槛上。看到这一拐点，我认为有助于重新评估整个行业。

From my perspective it's potentially the turning point for software the ability to showcase that we are at this inflection point and acceleration to actually see that it takes time for our software companies to develop new AI technologies put that into products that have been tested and proven and go through the enterprise adoption cycle and that we're at the cusp of more adoption that's what our survey work says and to see that inflection I think can help to rerate this sector.

林赛，你呢？同样的问题给你。

Lindsay, same question for you.

也许我可以从市场角度来谈。

Maybe I'll tie it to markets.

在过去几个月里，我和股权投资者的交流明显增多了。

I've already had a lot more conversations with equity investors over the past how many months.

目前固定收益领域备受关注，这对我来说是一个绝佳且非常有趣的机会。

There's a big fixed income focus right now, which is a great, you know, spot and really interesting opportunity in my seat.

如今信贷领域正涌现出许多有趣的新结构，因此我认为现在是一个令人兴奋的时期。

And there's a lot of interesting structures coming to be right now in the credit space, so I think it's it's an exciting time.

是的

Mhmm.

Lindsay、Stephen、Josh，非常感谢你们参与本次回顾，分享你们在会议中学到的内容。

Lindsay, Stephen, Josh, thank you very much for joining to recap the event and let us know what you learned at the conference.

感谢我们的观众实时收听。

To our audience, thank you for listening here live.

也感谢正在收听的观众们。

And to our audience tuning in, thanks for listening.

如果你喜欢关于市场的见解，请在你收听的平台给我们留下评价，并今天将本播客分享给朋友或同事。

If you enjoy thoughts on the market, please leave us a review wherever you listen and share the podcast with a friend or colleague today.

前述内容仅为信息性参考，基于创建时可获得的信息。

The preceding content is informational only and based on information available when created.

它不构成任何要约或招揽，也不构成税务或法律建议。

It is not an offer or solicitation nor is it tax or legal advice.

它未考虑您的财务状况和目标，可能并不适合您。

It does not consider your financial circumstances and objectives and may not be suitable for you.