中美科技竞赛

全球两大经济体正在展开一场科技竞赛，有人将其比作美苏冷战时期的核武与太空竞赛。

The world's two largest economies are competing in a tech race that some have compared to the Cold War nuclear and space race between The US and The Soviet Union.

联盟。

Union.

那么目前是美国还是中国处于领先地位？

So is The US or China currently ahead?

而什么因素将决定谁最终在这场科技竞争中占据主导地位？

And what will determine who ultimately takes the lead in this technological rivalry?

我是艾莉森·内森，这里是高盛交易所。

I'm Allison Nathan, and this is Goldman Sachs exchanges.

每个月，我都会与投资者、政策制定者和学者们探讨最紧迫的市场动向问题，为高盛研究的《首要关注报告》提供见解。

Each month, I speak with investors, policymakers, and academics about the most pressing market moving issues for our top of mind report from Goldman Sachs Research.

最近我采访了纽约大学发展研究院WABA战略竞争倡议的创始主任马克·肯尼迪，以及奥尔布赖特石桥集团合伙人保罗·特里奥罗。

I recently spoke with Mark Kennedy, founding director of the WABA Initiative for Strategic Competition at NYU's Development Research Institute, and with Paul Triolo, partner at DGA Albright Stonebridge Group.

我首先询问马克，科技在这场战略竞争中扮演着多么重要的角色，以及当前谁在科技竞赛中领先。

I started by asking Mark just how important a role tech plays in the strategic rivalry and who's winning the tech race today.

技术确实似乎处于美中战略竞争的核心位置。

Technology certainly seems to be at the center of The US China strategic rivalry.

这种对技术在这场竞争中角色的描述准确吗？

Is that an accurate characterization of the role of technology within this rivalry?

还是说您认为这个角色被过分强调了？

Or do you think that there's just too much emphasis on that role and that it's being overplayed?

我认为并没有过分强调。

I don't think it's being overplayed.

技术不仅仅是核心。

Tech isn't just at the center.

它实际上是总控台。

It's really the switchboard.

因为未来谁能主导数据流动、标准制定和算力分配，将真正影响从军事实力到经济影响力再到信息传播的每个领域。

Because whoever is going to route the data, the standards, the compute in the future is really going to impact every domain, from military might to economic influence to how you reach out with your message.

我认为技术的重要性没有被夸大，但被低估的是其地缘政治影响。

And I think it's not overplayed, but I think what's underplayed is the geopolitical implications of that.

当我们思考科技竞赛时，需要认识到这场竞赛实际上包含四个主要领域。

When we think about the tech race, we need to understand that there's really four arenas in this tech race.

谁在技术研发上领先？

Who leads in the technology?

谁在技术应用上领先？谁在构建支撑整个体系的数字基础设施上领先？以及谁在自主可控能力上领先？

Who leads in applying that technology, who leads in installing the digital plumbing that really underpins everything, and who leads in terms of self sufficiency.

在大多数前沿技术领域，美国都处于领先地位——无论是半导体、AI框架、云计算基础设施、量子计算还是全球人才储备。

On most of the leading technologies, US is winning on semiconductors, on AI frameworks, on cloud infrastructure, on quantum computing, on global talent.

但在某些技术领域，中国正在取得优势。

But there's parts of the technology where China is winning.

中国在量子通信、高超音速技术和电池技术方面处于领先地位。

They're ahead on quantum communications and hypersonics, on batteries.

在技术应用层面，中国正全力追赶，已在多个领域实现反超。

In terms of applications, China's really pressing hard and catching up in many areas and ahead in some.

以机器人技术为例，有数据显示在调整员工收入水平差异后，中国制造业应用的机器人数量是其他国家的12倍。

When you look at like robotics, one measure shows China using 12 times as many robotics in their manufacturing when you adjust for the income scale differences of employees.

因此他们正全力推进机器人应用以保持制造业优势。

So they're leaning hard into trying to keep their manufacturing edge by staying ahead on applying robotics.

而在我们最亲密盟友之外的全球设备安装领域，中国占据主导地位。

And in global installations outside of our closest allies, China's dominating.

以华为为例，其规模相当于紧随其后的两大竞争对手之和。

If you look at Huawei, they're as big as their next two largest competitors.

他们在全球南方国家占据统治地位。

They're dominant in the global South.

在自主性方面，中国正通过双循环等战略有意识地追求独立自主。

And when it comes to independence, China is deliberately, through their dual circulation and other strategies focusing on achieving independence.

近期政府禁止部分企业采购西方芯片的政令正体现了这一点。

And you're seeing that with the recent edicts of the government not allowing some of their companies to buy Western chips.

若训练模型时不使用西方芯片，他们还会提供能源折扣。

They're giving energy discount if you're training your model without using Western chips.

这只是他们为实现自主性所采取的众多激励措施之一。

Just one of a thousand different ways that they've been trying to gain independence and encouraging that.

美国在未来一段时间内仍将依赖中国。

The US will be dependent on China for some time to come.

所以我认为美国在第一个方面大幅领先，但中国正在其他技术竞赛领域迎头赶上甚至超越。

So I'd say America's largely ahead on the first aspect, but China is either catching up or ahead on the other aspects of this tech race.

那么如果我们思考各国拥有的筹码及如何运用这些筹码，美国实施的出口管制对中国技术进步造成了多大影响？

So if we think about where each country has leverage and how it's exercising that leverage, how much have the export controls The US has implemented impacted China's technological advancement?

首先，中国原本就计划依赖美国芯片。

Well, first of all, China was planning dependent on American chips to begin with.

技术限制加速了他们向这个方向的推进。

The tech restrictions have accelerated their push in that direction.

他们正在寻找方法，既能使用不那么先进的芯片，又能在某种程度上获取美国芯片技术。

They're finding ways to be able to do with less advanced chips and do have access to US chip technology at some level.

所以这确实拖慢了他们的步伐，但这不会成为让美国长期保持技术领先的制胜法宝。

So it's slowing them up, but this will not be the silver bullet that will keep The US in the tech lead for an extended period of time.

如果我们考虑中国掌握的筹码，尤其是经常被提及的稀土资源，其对稀土的控制会阻碍美国维持领导地位的能力吗？

And if we think about the leverage China has, and we've heard so much about rare earth minerals, will its grip on these rare earths impede The US's ability to maintain its leadership?

这很关键，他们给了我们一年时间——如果这种宽限能持续那么久的话——暂缓实施对稀土更严格的管控。

It's critical, and they've given us a one year, if it lasts that long, reprieve from the more restrictive controls on rare earths.

美国需要认真对待此事。

The US needs to take that seriously.

但我们早就知道这个问题，却一直停留在赞叹问题的层面，而非真正采取审慎措施。

But we've known about this for a while, and we've admired the problem as opposed to really taking deliberate steps.

最近在稀土投资领域取得了一些股权收购进展，但在这方面我们需要盟友的支持。

There have been some more recent steps of taking ownership interest in rare earths investing, but here we need our allies.

澳大利亚、加拿大等国对这些供应来源至关重要。

So Australia, Canada, others are vital to those sources of supply.

因此我认为我们需要比现在更加审慎，认真对待中国在这方面可能实施封锁的风险。

So I think we need to be even more deliberate than we are and take the risk of China's chokehold on this very seriously.

那么美国还需要采取哪些措施来赢得这场科技竞赛？

So what else does The US have to do to win the tech race?

你们需要拥有最先进的技术。

You need to have the best technology.

我们不是在削减就是对研究型大学的资助保持持平。

We're either flat or reducing our funding for research universities.

我们正在让全球顶尖人才更难进入。

We're making it harder for the world's best talent to come in.

我们必须具备与中国相匹敌的规模，以应对中国为掌控全球南方地区的网络基础设施、数字通道所做的系统性投入，否则我们将失去这些阵地。

We need to have the scale to match the deliberate effort that China's putting in to own the wiring, own the rails, own the digital flows in the global South, or we're going to sacrifice them.

我们是否能达到确保技术进步符合我们一贯秉持的价值观与标准的要求？

Are we going to live up to the standards that will make sure that the tech we advance is meeting the values and standards that we've always held?

在前者方面存在疑问，而关于可靠性的质疑也日益增多。

There's some question on the former and there's an increasing question on reliability.

若无法达到许多人期待的标准，这将削弱我们的可信度。

Without having the same standards that many people are looking for, that's going to mitigate our trust.

另一严重损害我们信誉的行为包括：曾切断乌克兰卫星信号长达一天，或是出台可能拒绝向某些国家出售AI技术的规则。

The other thing that's very much mitigating our trust is things like cutting off satellite access for Ukraine, for a day, or the AI rule that says that we may decide not to sell to you.

因此面对中国时，各国可能不信任设备安装后的可靠性，但中国正在兜售'我们比美国更可靠'的供应承诺。

So with China, countries may not trust the installation once it comes in there, but China is selling the, we're a more reliable supplier to you than America.

能源将变得非常重要。

And energy is gonna be very important.

我们能否将能源规模扩大到所需水平，以支持数据集和增长，从而获得赢得最佳技术竞赛所需的计算能力？

Can we scale up our energy at the level that we need to power the data set and growth that's gonna be necessary to have the compute power to win at that level of the best technology?

如果你看看中国在可再生能源方面所做的努力，包括风能、太阳能、核能等。

If you look at what China is doing in the renewable energies in terms of wind and solar and nuclear and others.

他们正大力投入其中。

They're leaning very heavily into that.

而美国，如你所知，总是走走停停，虽然我们拥有大量传统能源。

America, as you know, is start stop, go pause, and we have a lot of natural energy from traditional sources.

但中国正在非常努力地确保他们拥有这些能源来源。

But China is working very hard to make sure they have the sources.

我们现在也开始倾斜。

We are leaning now.

我们正大力00:00:00,000 --> 00:00:04,000 我们正大力投入核能，这是必要的，但这需要时间才能实现。

We're heavily into nuclear, and we need to, but then it's gonna take a while to achieve that.

美国政府似乎确实在关键领域采取更积极的角色，以推动这些目标的实现。

The US government does seem to be taking a more active role in important sectors to make progress towards these goals.

例如，我们看到有报道称政府正在资助或与稀土领域的公司合作。

We've seen, for example, reports that the government is funding or partnering with companies in the rare earth space.

那么这类政策能起到任何推动作用吗？

So will those kinds of policies move the needle at all?

嗯，这些矿产资源的获取权限确实会有所帮助。

Well, the access to the minerals that they will allow us will help.

但这是我们必须要做的，因为这些是波动性很大、周期性强的商品市场。

But it's something we are gonna need to do because these are very volatile, cyclical commodity markets.

而中国凭借其主导地位，有能力压低价格，挤出竞争对手。

And China, with their dominance, has the ability to basically bring the price down, squeeze out competitors.

因此，如果我们想确保某些最关键矿产的供应是不现实的。

So it's unrealistic for us to have assured supplies of some of the most vital minerals.

除非我们愿意签订某种承购协议，保证一个他们至少能维持生产的价格水平。

If we're not willing to have some kind of an off take agreement, we're guaranteeing a level of price that they can at least sustain their production.

这些措施对于确保矿物层面的供应来源是必要的，这样才能让我们在数字层面施展魔法。

Those are going to be necessary to have the sources of supply at the mineral level that will allow us to do the magic at the digital level.

但尚不清楚例如持有英特尔股权将如何带来长期利益，或者这种'黄金股'（无论是实际存在还是被认为存在）是否会阻碍英特尔进行必要的艰难重组行动，而这些行动正是英特尔重新获得全球竞争力所必需的。

But it's unclear how, for example, taking an ownership interest in intel is going to be long term beneficial, or could that be negative to the extent that either a real or perceived golden share prevents some of the tough actions for restructuring that are going to be necessary for Intel to become globally competitive again.

毫无疑问，拥有我们自己的晶圆厂至关重要。

There's absolutely no question that having our own foundry is vital.

而目前在大规模生产方面，我们唯一的玩家就是英特尔。

And right now, at scale, the only player we have is Intel.

虽然台积电正在美国投资，但他们的主要利益还是在台湾，维持那个'硅盾'。

We have TSMC investing in us, but their primary interest is Taiwan and keeping that silicon shield.

他们在美国的投资规模与在台湾持续建设和投入的规模相比只是杯水车薪。

And what they're doing in America is a shadow of what they're continuing to do and building on in Taiwan.

所以投资意愿是存在的，但很难找到美国投资商业企业——特别是成熟商业企业——产生长期积极影响的例子。

So the interest is there, but it's hard to point to examples of The US investing in commercial enterprises, particularly mature commercial enterprises that have had a long term beneficial impact.

让我们把目光转回中国。

Let's look back to China.

我们刚才讨论了美国的情况。

We talked about The US.

中国还需要采取哪些进一步措施来赢得科技竞赛？

What further steps must China take to win the tech race?

首先，他们有很多做法是正确的。

First of all, they're doing a lot of things right.

最有效的是他们持续、有规划、坚持路线、目标明确。

Most effective is that they are sustained, deliberate, stay on course, target their efforts.

你看他们投入这些技术的研究，与美国相比，美国在总研究量上仍略领先，但中国将所有资源都集中在他们认为将决定未来的几项关键技术。

You know, when you look at the research that they're investing into these technologies and compare it to The US, US is still somewhat ahead on total research, but China is all directed at a few key technologies that they think are going to determine the future.

因此我认为保持持续的投资水平对他们非常有利。

And so I think having a sustained level of investment has been very helpful to them.

他们还注重资源来源和市场开拓。

They're also focused on sources and markets.

他们并非昨天或去年或十年前才开始考虑稀土问题。

They didn't just think about rare earths yesterday or last year or last decade.

他们对此已经深思熟虑了很久。

They've been thinking about it for a long time.

但这不仅仅是稀土资源的问题。

But it's not just rare earths.

如果你观察港口，观察许多其他领域，他们清楚知道需要持续获取哪些资源。

If you look at ports, if you look at many other areas, they know what they need to have continuing to come into them.

他们也深知需要开拓市场。

And they also know that they need to have markets.

在开拓全球南方市场方面，他们比其他地区做得更出色，并愿意为此做出牺牲。

They're doing a much better job of establishing markets in the Global South than other places, and they're willing to take sacrifices to achieve that.

他们需要确保牢牢把握全球南方作为客户基础，因为他们看到美国、欧洲及其他西方盟友正逐渐疏远他们。

They need to make sure they have the Global South locked up as a customer base because they see US and Europe and other Western allies moving away from them.

但他们也意识到必须快速发展自己的芯片产业。

But they also see that they need to very quickly develop their chip industry.

所以你看到的部分现象是，他们正在扶持本土芯片制造和产业，让商业资金流向这些领域。

So part of what you're seeing is them favoring the local chip manufacturing and industry so that commercial dollars are flowing their ways.

这些可能仍是中国的商业资金，但为了资助并加速这一发展，他们愿意牺牲一些短期表现来实现目标。

It might still be Chinese commercial dollars, but to fund that development and to accelerate that development and are willing to sacrifice some near term performance in order to achieve that.

他们需要做的另一件事是建立信任，因为中国模式是受控的。

The other thing they need to do is they need to be trusted because the Chinese model is controlled.

他们不仅带着控制进入市场，系统还不具备互操作性、不易更改，你会被锁定其中，而且中国网络安全法还要求任何中国企业必须向政府提供其索要的任何数据。

And not only do they come in with control, it's not interoperable, it's not easily changeable, you're kind of locked in, but they also have the China cyber laws that would require any Chinese company to give the government any data that it asks for.

这类行为会削弱信任。

So those types of things reduce trust.

这在很大程度上是一场关于信任的较量。

This is in many ways a battle for trust.

如果我们集中精力，美国能赢得这场较量。

America can win that battle if we focus on it.

中国在这方面较为落后，他们需要确保人们不仅认为其技术价廉物美，还要值得信赖。

China's further behind in that, and that's an area where they need to make sure that people not only think their tech is cheap and reasonably good, but is trusted.

那么这场科技竞赛谁会赢，美国还是中国？

So who's gonna win this tech race, The US or China?

我认为现在还无法下定论，不过我之前已经谈过这个问题的四个方面。

I don't think we can say, but again, I talked about the four pieces of it.

我可以预见一个场景：美国拥有最尖端的技术，但中国拥有最多的实际应用。

I can see a world where America wins the best technology, but China has the most installations.

当你比较两国每年培养的电子工程专业毕业生和博士数量时，他们远超我们。

When you look at the number of electronic engineering graduates and PhDs they graduate every year compared to us, they're dramatically more than us.

这种数量优势会自然转化为质量优势。

And that quantity takes on a quality of its own.

因此我认为他们很可能在应用领域胜出。

So I can see them winning that application track.

特别是在美国及其最亲密盟友之外的地区，他们很可能在设备安装领域占据优势。

I can see them winning the installation track, particularly outside America and its closest allies.

而且我认为他们会在'我们独立于你们，但你们仍依赖我们'的竞争中获胜。

And I can see them winning the we're independent of you, but you're still dependent on us race.

所以可能会出现我们掌握设计图纸，但他们拥有实体建筑的状况。

So you could have a situation where we're owning the blueprints, but they're owning the buildings.

我们在代码领域领先，而他们在电路领域领先。

We're in the lead in the code, but they're in the lead in the circuits.

这对美国的地缘政治和经济将产生巨大影响。

And that has huge geopolitical and economic consequences for America.

DGA Albright Stonebridge集团的合伙人保罗·特里奥罗也认为，美国能否保持其技术领先地位，特别是在构建全球最强大AI模型方面的领先优势，目前尚无定论。

Paul Triolo, partner at DGA Albright Stonebridge Group, agrees that the jury is still out on whether The US can maintain its tech lead and specifically its lead in building the world's most powerful AI models.

他表示这在很大程度上取决于中国能否突破半导体供应链中的关键瓶颈。

He says much will depend on whether China can overcome critical choke points in the semiconductor supply chain.

当我们审视全球半导体供应链时——这确实是AI技术的核心所在——瓶颈在哪里？谁控制着这些环节？

When we think about the global semiconductor supply chain, which really is at the heart of this AI technology, where are the choke points and who controls them?

好问题。

Great question.

首先我认为并不存在单一的供应链，对吧？

So I think, first of all, there isn't one supply chain, right?

实际上存在多个相互交织的全球供应链网络，产品在各大洲之间往返流动。

There are multiple interlocking global supply chains and products go back and forth across continents.

有些图表显示晶圆从这里开始加工，运到这里进行某些增值步骤，再运往他处，最终完成封装后进入产品。

There are maps that show the wafer starts here, it goes here, some value added here, it goes here, then it's finally packaged, then it goes into a product.

因此'供应链'这个词并不能充分代表这个行业的实际情况。

So the word supply chain is not all that representative of what the reality of the industry is.

这是全球范围内最复杂的技术集合，经过跨国协作才能最终形成成品。

It's the most complicated set of technologies that crisscross the globe to get put into a final product.

其中包含若干关键技术环节。

There are these key technologies that are part of that.

第一个被广泛讨论的是先进光刻技术。

So the first one that is fairly commonly talked about is advanced lithography.

这种设备最终将半导体电路图案印刷到硅晶圆上。

This is the tool that eventually prints the picture of the semiconductor layout onto a silicon wafer.

现在这项工艺的精细程度已经远超人类大脑的理解范畴。

And that's being done now at levels that's really hard for the human brain to comprehend.

2纳米工艺，对吧？这比单根头发丝还要细小得多。

Two nanometers, right, which is way smaller than a single strand of hair.

所以这些半导体元件的尺寸，我们所说的特征尺寸，这些特征尺寸真的非常非常微小，而且在过去三四十年间一直在迅速缩小。

So the dimensions of these semiconductors, the dimensions which we're talking about for the features, these are the feature sizes, are really, really teeny and have been shrinking rapidly over the last thirty or forty years.

这正是推动摩尔定律发展的动力——每两年翻倍的容量和能力。

This is what's been driving Moore's Law, the doubling of capacity and capability every two years.

因此光刻技术成为了这个关键的门控功能，，因为荷兰，这家荷兰公司，在过去二十年里开发开发出了最先进的光刻，极紫外光刻工具，经历了各种类型的光刻工具的迭代。

And so lithography has ended up being this key gating function, if you will, because ASML, this Dutch company, ended up developing the most advanced lithography, extreme ultraviolet lithography tools over the course of the last twenty years, moving through successive types of lithography tools.

所以现在这是一个很大的瓶颈，因为要获得最先进的架构和特征尺寸，对于半导体来说，你需要极紫外光刻技术。

So right now, that's a big choke point because to get to the most advanced architectures and feature sizes, for semiconductors, you need EUV.

另外两个瓶颈是，随着特征尺寸越来越小，工具制造行业的其他部分也变得重要起来。

And the other two choke points, as you get smaller and smaller feature sizes, other parts of the tool making sector become important.

比如蚀刻。

So etch.

这是其中的一个阶段。

This is the stage.

在光刻之后，你会使用各种化学物质和先进工具进行蚀刻，以暴露出底层的电路。

After the lithography, you etch away using various chemicals and advanced tools to expose the underlying circuit.

作为光刻工艺的一部分，你需要先放置掩模版，下一步就是蚀刻。

And so as part of the lithography process, you're putting a mask down, and then the next step is etching.

在极紫外光刻中，你确实需要非常精良的蚀刻设备来完成这一步骤。

And in EUV lithography, you really need really good etching equipment to do this.

美国设备制造商如泛林集团、应用材料公司，以及日本的东京电子，都是该领域的重要参与者。

And there are US toolmakers like Lam and Applied Materials and Tel from Japan are also really important players there.

最后还存在其他瓶颈，即如何整合所有这些设备来制造具有先进制程的芯片。

And then finally, there's the other choke points in terms of the ability to put all these tools together and manufacture something at that advanced feature length.

台积电（台湾积体电路制造公司）的商业模式本身成为了瓶颈，因为他们的定位是专注于制造环节，不与客户竞争。

TSMC, Taiwan Semiconductor Manufacturing Corporation, it turned out that their business model became a choke point because their model was, we'll focus just on the manufacturing process, and we won't compete with our clients.

因此像英特尔和三星这样的其他代工厂也在生产自己的设备。

And so other foundries like Intel and Samsung are also making devices of their own.

但台积电能成为值得信赖的首选厂商，部分原因是他们不与客户竞争，部分原因是他们的工程技术确实出色。

But TSMC became the trusted go to player in part because they weren't competing with their clients and in part because they're just really good at engineering.

他们与设备制造商和客户都建立了可靠的合作关系。

And they've developed trusted relationships with both the toolmakers and their clients.

因此他们能够独特地根据客户需求推动制造创新。

And so they're able to uniquely push the manufacturing innovation to the needs of their customers.

所以这个模式被证明是非常了不起的。

And so that model proved to be really amazing.

现在他们生产了超过90%的最先进制程芯片。

So now they produce over 90% of the most advanced nodes.

那么任何一个国家有可能在半导体领域实现自给自足吗？

So is it possible for any country to become self sufficient in semiconductors?

如果有谁能做到这一点，那很可能就是中国和中国企业，因为这是个大国。

If anybody can do this, it's probably China and Chinese companies because it's a big country.

他们有很多STEM教育项目，每年培养出大量工程师。

They have a lot of STEM education programs, and they're cranking out a lot of engineers per year.

由于美国的出口管制，他们现在有很强的动力，而且基本上被切断了获取EUV技术的渠道。

They have the incentive now because of US export controls, and they've been largely cut off from access to EUV.

大约从2018年开始，美国与盟国合作基本上阻止了中国企业购买EUV设备。

Starting in around 2018, The US working with allied partners basically was able to prevent Chinese companies from purchasing EUV equipment.

因此自那时起，中国体系一直在努力探索如何发展所有这些开发EUV所需的并行技术。

And so since then, the Chinese system has been trying to figure out how do we develop all these parallel technologies that you need to develop EUV.

但让光刻技术如此困难的原因在于它本质上是一个系统集成问题。

But what makes photography so difficult is that it's really a systems integration problem.

你需要大规模集成复杂系统：光学元件、光源、大量软件、许多小部件，以及光刻胶等材料。

So you're integrating complex systems at scale, optics, the light source, lots of software, and lots of little pieces of the puzzle, and also materials like photoresists.

因此光刻机必须与光刻胶协同工作，还要与制造流程中的所有其他设备配合。

So the lithography has to work with photoresist, and then it has to work with all these other tools that are part of the manufacturing process.

此外，真正重要且困难的部分是维护这些设备，因为它们非常精密。

And also, the really important and hard part is to service these pieces of equipment because they're very delicate.

正如一位业内知情人士曾告诉我的：'造出原型机很容易，但要维护晶圆厂里24/7运转的数百台设备并保持其持续运行却极其困难。'

As one industry insider once told me, said, It's easy to build a prototype, but it's really hard to service hundreds of these things 20 fourseven operating in foundries to keep them up and running.

因为这些设备不仅是复杂机械，还需要备件供应、软件更新，更需要一支真正专业的技术团队来支持这些复杂工具的全球部署——面对要求严苛的客户，他们期望设备保持最高运行时间且几乎零缺陷。

Because in addition to being complex machinery, they need spare parts, there's software updates, and you need a really capable workforce to be able to support deployments of those complex tools for very demanding customers worldwide that are expecting maximum uptime and very few defects.

所以部分挑战在于：你需要大量经验丰富的工程师来进行设备维护。

And so that's part of the challenge is you need a lot of very seasoned engineers to service that.

那些制造数百万尖端半导体的公司，会不愿意冒险采用未经证实的技术。

And companies that are manufacturing millions of semiconductors at the cutting edge, they're gonna be reluctant to risk an unproven technology.

你提到了对中国的出口限制。

You mentioned export restrictions to China.

如果能听听你的看法，这些政策在延缓中国实现自主可控以及人工智能领域发展目标方面的效果如何，会很有帮助。

It would be useful to get your answer on how effective those policies have been in slowing China's progress towards its own self sufficiency and everything it's trying to pursue in terms of AI.

是的。

Yeah.

美国已禁止向中国出售最先进的GPU芯片。

The US has prevented the sale of the most advanced GPUs to China.

GPU可以说是开发先进AI模型的计算单元。

GPUs are the unit, if you will, of compute for developing advanced AI models.

因此美国的政策在切断获取渠道方面是有效的，但情况比这更复杂，因为渠道的切断迫使中国产业提升自身水平。

So The US policies have been effective in cutting off access, but it's a little more complicated than that because the cutting off of access has forced the Chinese industry to get better.

这在半导体制造和GPU设计研发领域产生了连带效应，迫使中国企业提升能力，发展替代技术。

There has been a collateral effect both on the semiconductor manufacturing side and on the GPU design and development side of forcing Chinese companies to get better, to be able to develop alternative capabilities.

所以我们现在正处于这个过程的中间阶段。

And so we're in the middle of that process right now.

就效果而言，我认为结果还有待观察。

In terms of the effectiveness, I guess the outcome remains to be seen.

如果中国能够发展出一个相当不错的替代制造供应链，使其能够生产虽不及英伟达但相当不错的GPU，并支持开发出具有竞争力的人工智能模型，那么我们就需要重新评估这些政策的有效性。

If China, for example, is capable of developing a pretty good alternative supply chain for manufacturing that allows it to manufacture GPUs that are maybe not as good as NVIDIA's but are pretty good and enable them to develop advanced AI models that are pretty good and competitive, then we have to reassess the effectiveness of them.

因此，有效性取决于目标是什么，而这正是你问题的第二部分——中国开发先进人工智能模型的能力。

So the effectiveness depends on what the goal is, and that really is the second part of your question, which is China's ability to develop advanced AI models.

而在这一领域，所有领先的开源和开放权重模型都来自中国。

And there, all of the leading open source and open weight models are all Chinese.

它们都来自阿里巴巴、腾讯、百度、月之暗面、深度求索等公司。

They're all from companies like Alibaba, Tencent, Baidu, Moonshot, Drupo AI, and DeepSeek.

这真是个奇怪的情况，对吧？

And that's a really weird situation, right?

美国通过控制半导体来延缓中国AI发展，但中国企业仍能获取英伟达旧款芯片、华为新款芯片和GPU，并通过自主研发内部芯片，持续设计出非常强大的人工智能模型。

So The US has controlled semiconductors to slow China's AI development, but Chinese companies have access to older NVIDIA chips and to new Huawei chips and GPUs and also developing their own internal in house chips have been able to continue to design very capable AI models.

一年前情况并非如此。

And a year ago, this wasn't the case.

一年前，开源开放权重模型领域的排行榜上领先的是Meta和美国公司。

A year ago, the leaderboards, for example, in the open source, open weight model arena were Meta and U.

S。

S.

公司以及法国的Mistral。

Companies and Mistral from France.

但现在全是中国的公司，因为像DeepSeek这样的中国企业决定走开源路线，这促进了大量快速创新。

But now they're all Chinese companies because the Chinese companies like DeepSeek decided to go the open source route, and that's allowed a lot of rapid innovation.

所以目前领先的模型是由美国的OpenAI和Anthropic开发的，它们都是封闭的专有模型。

So right now, the leading models are being developed in The US at OpenAI and Anthropic, which are both closed proprietary models.

但从模型采用率来看，由于中国模型是开源开放权重的，初始部署成本更低，这些开源模型正被越来越多的公司使用。

But in terms of uptake of models, because Chinese models are open source, open weight models and are cheaper to initially deploy, all those open source models are being more and more used by companies.

这确实是一个非常有趣的趋势。

So that's a really interesting trend.

因此当谈到减缓中国发展时，这些模型的扩散问题虽不常被讨论，但却至关重要，因为AI的成功既取决于能力和模型的开发，也取决于使用者的广泛程度。

And so when you talk about slowing down China, the issue of diffusion of these models is not often discussed, but it's an important thing because the success of AI is based on the development of capabilities and models and also who's using it.

实际上，当前几乎更重要的是这些模型在商业领域的部署应用。

Really, almost more important right now is the deployment of these models in business.

中国政府一直在推行政策，通过在全国建设数据中心来帮助像深度求索这样的企业。

And the Chinese government has been pursuing policies to try to help companies like DeepSeek by building data centers all over China.

所以中国政府试图构建全国性的支持性基础设施，包括电力保障和数据中心建设，让中国企业能更容易获得算力资源。

So the Chinese government has tried to build out a supporting infrastructure nationally, both in terms of power and in terms of data centers, and make it a little bit easier for Chinese companies to gain access to compute.

像阿里巴巴、腾讯和百度这样的大型AI实验室拥有自己的云端基础设施，因此它们并不真正需要依赖国家数据中心。

The bigger AI labs like Alibaba and Tencent and Baidu have their own cloud based infrastructure, and so for them, they don't really need the state data centers to support them.

但其他规模较小的公司未必有能力自建庞大的数据中心集群来训练模型。

But other smaller companies don't necessarily have the capability to have their own massive cluster of data centers to design their models.

对这些企业而言，获得算力支持将非常关键。

So for those companies, having access to compute will be important.

我们还听到一种观点认为，中国在稀土领域的主导地位和充足的电力供应，使其在AI竞赛中相比美国具有显著优势。

We've also heard the argument that China's dominance in rare earths and its abundant power supplies gives it significant advantages over The US in the AI race.

这是真的吗？

Is that true?

就AI数据中心建设而言，稀土问题的重要性相对较低。

The rare earths issue is less salient, arguably, in terms of the AI data center build out.

目前对汽车产业、消费电子以及国防航天等需要磁铁的行业更为关键。

It's more important for industries right now, the auto industry, even consumer electronics, and of course, defense and aerospace, where things like magnets come into play.

半导体行业确实也需要磁铁，但用量非常少。

Now, magnets are important also in the semiconductor industry, but really small numbers there.

中国正在为这些领域颁发许可证。

And the Chinese are issuing licenses for those.

稀土问题会持续引发摩擦，但不会真正影响大规模AI基础设施的建设。

So rare earths are going to be a constant irritation, but they're not going to really affect, I think, the build out of large scale AI infrastructure.

但能源将成为关键制约因素。

But energy is going to be this big gating function.

在美国，电力供应将成为数据中心建设的重大瓶颈，因为这些数据中心需要持续稳定的大量电力。

Power is going to be, in The US, a huge constraint on the build out of these data centers because these data centers require a lot of power, constant, stable power.

以美国为例，若考察所有替代能源方案——如核能与天然气，以及涡轮机、变压器等设备，美国在供应足够数量的硬件以产生充足能源方面存在巨大的供应链问题。

And if you look at all the alternatives in The US, for example, nuclear and natural gas, things like turbines and transformers, there's a huge supply chain problem in The US in terms of supplying sufficient quantities of hardware, really, to generate enough energy.

而且我们也没有统一的电网体系，而中国拥有这一优势，这使得电力跨区域调配更为便捷。

And we don't have a unified grid either, which China also has, that makes it easier to move power from place to place.

鉴于美国面临的电力问题，许多数据中心可能会建在中东地区，美国企业将选择使用这些设施，只因中东能获得大量相对廉价的电力资源。

Because of that power issue that The US has, probably a lot of data centers will get built in The Middle East, and US companies will use those just because The Middle East has access to so much fairly cheap power.

那么在这场人工智能竞赛中，最终会是美国还是中国胜出呢？

So who will ultimately prevail in the AI race, The US or China?

首先，认为美国能保持领先的前提是基于规模扩张的理念——即投入的GPU越多，算力越大，模型表现就会越好。

First of all, the premise that The US will be able to stay ahead is premised on the idea of scaling and the idea that the more GPUs, the more compute you throw at these models, the better they will get.

但这一观点目前尚无定论。

And the jury's still out on that.

当然，很大程度上还取决于中国半导体产业能否突破我们之前提到的那些硬件瓶颈。

A lot of it depends, of course, on the ability of the Chinese semiconductor industry to overcome some of those hardware choke points that we mentioned earlier.

坦率地说，中国的重大劣势在于六七十年代对半导体领域关注不足，导致其在这一行业错失了二十多年的发展机遇，而同期美国正全力发展半导体产业。

China's big disadvantage, frankly, is they didn't pay attention to semiconductors in the '60s and much in the '70s, so they lost two decades or more of focus on an industry, while in The US, there was a lot of attention being paid to developing the semiconductor industry.

中国还面临着资本市场不发达的劣势。

And China has the disadvantage of underdeveloped capital markets.

美国曾拥有大量耐心资本来资助半导体行业发展，而中国目前缺乏这种条件。

The US had the luxury of a lot of patient capital in helping to finance the development of our semiconductor industry, and China just doesn't have that.

因此正如我所说，政府必须发挥更大作用。

So the government has to play a bigger role, as I said.

世界上唯一有可能尝试重建大部分半导体供应链的国家就是中国，因为他们确实拥有大量工程师，并且现在政府也有资源激励政策来推动这件事。

The only country in the world that could even conceive of trying to recreate large parts of the semiconductor supply chain is China, because they do have the numbers of engineers, and they do have the government resource incentive now to do this.

但我们没有其他可参考的成功先例，对吧？

But we don't have any other example of this being done to compare to, right?

所以很难评估他们最终能取得多大成功。

So it's hard to assess how successful they will be.

从我在美国的观察来看，似乎普遍认为美国在科技竞赛中领先，尤其是人工智能领域。

There seems to be a perception from where I sit in The US that The US is leading the tech race, especially when it comes to AI.

但我和马克、保罗交谈后最深刻的体会是：中国拥有许多超越美国的优势，这可能最终会让中国胜出。

But perhaps the most striking takeaway I had from my conversations with Mark and Paul is that China has a lot of advantages over The US, and that could lead to China ultimately coming out ahead.

如果中国能获得更多美国先进芯片，这种情况尤其可能发生，而美国政策制定者本周似乎正朝这个方向推进。

That would be especially the case if China has more access to advanced US chips, and that's the direction US policymakers seem to be moving in at least this week.

我们就先聊到这里。

Let's leave it there.

感谢马克·肯尼迪和保罗·特里奥罗。

My thanks to Mark Kennedy and Paul Triolo.

也感谢您收听本期高盛交易所节目。

And thank you for listening to this episode of Goldman Sachs Exchanges.

我是艾莉森·内森。

I'm Alison Nathan.

本文所表达的观点和意见截至发布日期，可能随时变更且不另行通知，并不一定反映高盛或其附属机构的机构观点。

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.

提供的材料仅供信息参考之用，不构成投资建议、高盛任何实体采取特定行动的建议，也不构成购买或出售任何证券或金融产品的要约或招揽。

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.

过往业绩不代表未来表现。

Past performance is not indicative of future results.

高盛及其任何关联方均不对本文件所含陈述或信息的准确性或完整性作出明示或暗示的保证，且对任何用途依赖该等信息的行为概不负责。

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归其所有，未经授权不得使用。此处使用仅用于信息识别目的，不代表任何所有权关联。

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.

为方便起见提供本文件，可能与原视频或音频内容存在差异。

A transcript is provided for convenience and may differ from the original video or audio content.

文本仅供信息参考，不代表任何投资建议。

Goldman Sachs is not responsible for any errors in the transcript.

未经授权不得复制、分发或发表。

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

未经授权不得复制或分发。

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

未经授权不得复制或分发。

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

版权所有 © 2025，高盛集团。

Copyright 2025, Goldman Sachs.

保留所有权利。

All rights reserved.