推荐微信扫码登录以获得更好的体验!另:🎉建议更新到最新版,以体验新功能!🎉 版本更新记录→

《经济学人》英伟达:AI时代的『中央银行』 [B2]

字幕摘录

时间英文中文
0:19It took 30 years for Nvidia, an American firm whose chips power much of the world's artificialNvidia公司花了30年时间 该公司是一家美国公司, 它的芯片能提供世界大部分的人工
0:25intelligence, to reach a valuation of $1 trillion.情报,达到价值1万亿美元。
0:30Getting to $2 trillion took only nine more months.达到两万亿只用了九个月
0:33It passed the $5 trillion mark less过了5万亿马克
0:36than two years after that.两年之后
0:38It is now the world's most valuable company, worth around $5.4 trillion.它现在是世界上最有价值的公司,价值约5.4万亿美元。
0:45Next year its sales are expected almost to double.预计明年其销售额将几乎翻一番。
0:48Some analysts predict it will take有些分析家预测这需要
0:50in $1 trillion in annual revenue by 2029.到2029年年收入1万亿美元。
0:55It is not just chip making that has spurred不只是芯片制造 刺激了
展开字幕全文(293 条)
序号英文中文
1It took 30 years for Nvidia, an American firm whose chips power much of the world's artificialNvidia公司花了30年时间 该公司是一家美国公司, 它的芯片能提供世界大部分的人工
2intelligence, to reach a valuation of $1 trillion.情报,达到价值1万亿美元。
3Getting to $2 trillion took only nine more months.达到两万亿只用了九个月
4It passed the $5 trillion mark less过了5万亿马克
5than two years after that.两年之后
6It is now the world's most valuable company, worth around $5.4 trillion.它现在是世界上最有价值的公司,价值约5.4万亿美元。
7Next year its sales are expected almost to double.预计明年其销售额将几乎翻一番。
8Some analysts predict it will take有些分析家预测这需要
9in $1 trillion in annual revenue by 2029.到2029年年收入1万亿美元。
10It is not just chip making that has spurred不只是芯片制造 刺激了
11Nvidia's stunning growth, however.然而,恩维迪亚的惊人增长。
12Its boss, Jensen Huang, has also resorted to financial其上司黄延生也曾利用金融手段
13engineering to boost demand for its wares.工程可以刺激对它产品的需求
14In mid-August, for example, Nvidia agreed例如,8月中旬,恩维迪亚同意
15to provide a backstop worth up to $105 billion for a vast data centre in Ohio that will use为俄亥俄州一个庞大的数据中心提供价值达1 050亿美元的支持,该中心将使用
16lots of its chips.许多芯片。
17A week earlier it had revealed plans to mobilise more than $500 billion of一周前,它揭露了动员5 000多亿美元
18investment in AI infrastructure with the help of six big Wall Street firms by guaranteeing在六家华尔街大公司的帮助下投资AI基础设施,保证
19the value of the equipment it sells to such projects.向这类项目出售的设备的价值。
20It may underwrite as much as a quarter它可能承担多达四分之一
21of the cost of some investments.一些投资的成本。
22In July Nvidia helped some customers finance big data centres7月,Nvidia帮助一些客户资助大数据中心
23in another way by promising to top up their income if it misses set targets.以另一种方式,保证如果收入未能达到既定目标,则将增加收入。
24Companies公司
25often lend to their customers.经常借给顾客
26Think of the financing arms of car makers, for example.例如,想一想为汽车制造商的武器提供资金。
27Mr Huang argues that Nvidia is simply helping to unlock investment for perfectly viable黄先生说,Nvidia只是帮助解锁投资 完全可行
28projects that might otherwise struggle to borrow enough at the right price.(b) 可能难以以适当价格借贷的项目。
29But to critics,但对于批评者来说,
30these deals carry more than a whiff of the dot-com boom of the late 1990s when telecom20世纪90年代后期 电信业的网络热潮
31equipment makers such as Cisco and Lucent lent billions of dollars to telecoms providersCisco和Lucent等设备制造商向电信供应商贷款数十亿美元
32that bought their gear.买了他们的装备。
33When demand for the telecoms firm's services fell short, some当对电信公司服务的需求不足时,有的
34of them collapsed, landing Cisco and Lucent with big losses.他们崩溃了, 登陆思科和卢森特 损失巨大。
35As Jay Goldberg of Seaport作为海港的杰伊·戈德伯格
36Research Partners, a firm of analysts, puts it, Nvidia is walking a fine line between研究伙伴,一个分析师事务所, 说,恩维迪亚正在走 一条精细的线之间
37enabling demand and creating it.允许需求并创造需求。
38He does not yet think Nvidia has crossed that line, although他还没有认为Nvidia已经越过了这条线,虽然
39it is getting pretty close.越来越近了
40Others are more sceptical still, including Michael Burry,其他人更怀疑 包括迈克尔·布利
41an investor who made a fortune betting against the mortgage-backed securities that precipitated以抵押支持的证券为赌注
42the financial crisis of 2007-9.2007-9年金融危机。
43He and others are asking what might happen if demand for他和其他人问,如果要求
44AI chips grows more slowly than expected.AI芯片的生长比预期的慢.
45Supply increases and prices fall, which would供应增加和价格下降,这将
46not only reduce Nvidia's own profits but potentially also generate losses on its lavish不仅会减少Nvidia的利润 而且还会给它的奢侈品带来损失
47support for its customers.支持客户。
48The question is pertinent because Nvidia's financial commitments这个问题很重要,因为Nvidia的财政承诺
49are huge and growing fast.是巨大的,增长很快。
50Over the past three years, it has pledged over $70 billion in在过去三年中,它认捐了700多亿美元。
51investment in start-ups and offered $300 billion in financial support to its customers.向客户提供3 000亿美元的财政支持。
52Some有点
53have taken to calling Nvidia the central bank of AI because it plays so pivotal a将Nvidia称为AI的中央银行 因为它的作用如此关键
54role in financing the industry.(c) 在为工业融资方面发挥作用。
55That will indeed, as Mr Huang argues, help the industry正如黄先生所说,这确实能帮助这个行业
56grow, but it also carries risks.增长,但也带来风险。
57Nvidia's financial engineering is partly a responseNvidia的金融工程是部分反应
58to its biggest customer's transformation into rivals.给最大的客户 变成对手。
59Hyperscalers, tech giants such超规模的,技术巨头
60作为亚马逊,Google,Meta和微软,约占Nvidia收入的一半.
61This这个
62year they are projected to invest around $800 billion, largely on AI infrastructure, but预计每年投资约8 000亿美元,主要用于AI基础设施,但
63most of them have begun designing their own chips, which puts their future purchases from他们中的大多数已经开始设计自己的芯片,这使得他们未来的购买来自
64Nvidia in doubt.恩维迪亚在怀疑中。
65For the hyperscalers, these custom chips are much cheaper, costing between对于超规模机,这些自定义芯片成本要低得多,成本在
66和Nvidia一样多
67They can also perform some tasks better because他们也可以更好地执行一些任务,因为
68the hyperscalers can tailor them to their own software.超标器可以适应自己的软件
69Worse, from Nvidia's point更糟的是,从恩维迪亚的角度
70of view, a few of the tech giants are not simply developing chips for their own use,从观点看,少数技术巨头 不只是开发芯片供自己使用,
71but also marketing them to others.还要推销给别人
72Google, for example, has sold some specialised processors例如,Google出售了一些专门的处理器
73to Anthropic, a big free-standing AI lab.去Anthropic 一个独立的人工智能实验室
74Amazon also expects its custom chip business亚马逊也期望它的自定义芯片业务
75to become a sizeable source of revenue.成为可观的收入来源。
76Bloomberg Intelligence, a data provider,彭博情报公司 数据提供者
77predicts that custom chips will gradually eat into Nvidia's sales, accounting for about 50%预测定制芯片将逐渐进入Nvidia的销售,占50%左右
78of the market for processors used in AI by the end of the decade, up from roughly 40% this year.到本十年末,人工智能使用的加工器的市场从今年的大约40%上升。
79Nvidia, therefore, must look elsewhere for growth.因此,恩维迪亚必须寻找其他地方的增长。
80It is courting governments that want to build正是求爱的政府想要建立
81domestic AI infrastructure, companies constructing their own data centres and国内人工智能基础设施、建造自己数据中心的公司和
82neoclouds, firms that rent out AI computing capacity.新云 租赁AI计算能力的公司
83Such plans require huge upfront这样的计划需要巨大的前期
84investments.投资。
85A one-gigawatt facility, the kind used by large AI labs, contains around half a大型AI实验室使用的单基瓦设施大约含有半个
86million chips.百万筹码.
87Bernstein, a research outfit, estimates that equipping one with Nvidia'sBernstein, 一个研究装备, 估计一个装备 恩维迪亚的
88hardware costs around $47 billion, two-fifths of which goes on the chips alone.硬件花费约470亿美元,其中的五分之二仅用于芯片.
89Nvidia reportedly据报,恩维迪亚
90plans to raise the price of its latest processors by more than 15%, blaming the soaring cost of计划将其最新处理器的价格提高15%以上,指责其成本飞涨。
91memory chips, an important component.内存芯片,一个重要的组件.
92Hyperscalers have little trouble financing all this.超规模公司在为这一切融资方面没有什么困难。
93They have他们有
94typically used their prodigious cash flow to pay for new data centres.通常使用其庞大的现金流动支付新的数据中心的费用。
95More recently, as their最近,作为他们的
96spending has soared, they have also begun to borrow heavily.支出激增,也开始大量借款。
97Four of them have issued more than其中四个已经发行超过
98$200 billion in debt in 2025 and 2026.2025年及2026年债务2000亿元.
99Hyperscalers have investment-grade credit ratings,超规模者有投资级信用评级,
100which keep their borrowing costs low.使他们的借贷成本低。
101Upstart neoclouds have similar spending needs, but little新星云也有类似的支出需求,但很少
102revenue.收入。
103Their loans are naturally much more expensive.他们的贷款自然要昂贵得多。
104Alphabet, Google's parent company,字母,谷歌的母公司,
105sold $2.75 billion of 50-year bonds in November at an annual interest rate of 5.7%.11月以年利率5.7%。
106The rate at费率为
107which CoreWeave, the biggest neocloud, borrowed $2.6 billion in July was almost double.7月借到26亿美元 几乎翻了一番
108It is没错
109this gap between hyperscalers' borrowing costs and everyone else's that the bank of Nvidia wouldNvidia银行的借款成本与其他人的差额
110like to narrow.喜欢缩小。
111One way it does that is by taking equity stakes in start-ups that will be customers一种方式是把股权股权 给新开办企业 成为客户
112themselves or that will help fuel demand for Nvidia's chips indirectly.或者能间接帮助Nvidia芯片的燃料需求。
113Last year Nvidia made去年恩维迪亚做了
114about 90 such investments, nearly twice as many as two years earlier.大约90项此类投资,几乎是两年前的两倍。
115This year it has already今年已经
116agreed to another 60-odd.同意另外60多码。
117Some of these checks aim to propagate open-weight AI models, which users其中一些检查旨在推广开放量级的AI模型,用户使用这些模型。
118can download free of charge and adapt, unlike proprietary offerings from firms like Anthropic,与Anthropic等公司的专有服务不同,
119OpenAI and Google, which users tend to access via subscriptions and whose inner workings are hidden.OpenAI和Google,用户倾向于通过订阅访问,其内部工作被隐藏.
120In August Nvidia agreed to pay Poolside, a start-up building AI coding models, $6 billion8月,Nvidia同意支付Poolside, 一个启动建筑AI编码模型,60亿美元
121to license its software and a further $1 billion for a stake.(c) 为其软件发放许可证,并另外提供10亿美元作为股份。
122It has also reportedly agreed to buy据报道,它还同意购买
123Huggingface, a platform hosting open-weight models for $12.9 billion.胡景面,是一家托管开放量级模型的平台,耗资129亿美元.
124The intention behind such其意图是:
125investments is to fuel demand for Nvidia's chips by creating a proliferation of AI products and投资是为了刺激对Nvidia芯片的需求,办法是制造AI产品的扩散,
126companies that are independent of the hyperscalers.公司独立于超规模公司。
127Increasingly, however, Nvidia is not simply但是,Nvidia越来越不简单
128investing in promising AI firms but also helping them finance big projects more directly.投资于有前途的AI公司,但也帮助他们更直接地资助大型项目。
129In early早点
130July it announced a new stratagem in which it promises to top up NeoCloud's income from new7月,它宣布了一个新的策略,其中它承诺将填补NeoCloud的新收入。
131data centres to an agreed floor.数据中心到商定的楼层。
132These undertakings, the exact terms of which vary from deal to deal,这些承诺的具体条款因交易而异,
133often last for six years.经常持续6年。
134Throughout that period Nvidia promises to pay a set price for compute,在此期间恩维迪亚承诺支付计算费的固定价格,
135as the jargon has it.说得没错
136If the NeoCloud manages to sell the capacity in question at a higher price,如果NeoCloud能以更高的价格出售有关能力,
137Nvidia receives a share of the difference.Nvidia收到一部分差额。
138This safety net makes NeoCloud's future revenues much这个安全网让NeoCloud未来的收入大增
139more predictable and so lowers the cost of the debt they take on to build new data centres.更可预测,这样就能降低他们为建造新的数据中心而承担的债务成本。
140That,这个
141in turn, spurs demand for Nvidia's processors.进而刺激对Nvidia处理器的需求。
142Charon AI,一个澳大利亚NeoCloud, 期望制作
143use of a backstop of this sort worth about $4.9 billion to deploy around 40,000 Nvidia chips.利用这种价值49亿美元的支援 部署大约4万块Nvidia芯片
144Firmus,另一个NeoCloud,计划购买多达17万.
145Dan Nischball,Semi的分析师
146Analysis, a consultancy, argues such arrangements need not be a permanent feature of the industry作为顾问的分析认为,这种安排不一定是该行业的永久特征。
147but instead buy time for lenders to become more comfortable about lending to such projects.却为放款人争取时间,让他们更愿意向这类项目贷款。
148In the meantime, the combination of these different forms of support is creating a complex同时,这些不同形式的支助的结合正在形成一个复杂的结构。
149tangle of commitments.承诺之交。
150Take CoreWeave, a NeoCloud in which Nvidia has accumulated an 11% stake forNvidia在其中积累了11%的股份。
151more than $2 billion.超过20亿美元。
152CoreWeave hopes to build more than 5 gigawatts of data centre capacityCoreWeave希望建设5千兆瓦以上的数据中心能力
153using Nvidia's chips by 2030.到2030年使用Nvidia的芯片.
154Nvidia同意花费63亿美元 用于CoreWeave的未使用
155capacity by 2032 if need be.如有必要,可在2032年前完成。
156Nvidia同时是CoreWeave的投资者,供应商和客户.
157Nvidia is also taking a creative approach with big AI labs which need huge amounts of computeNvidia也正在采取创造性的方法 与大型AI实验室 需要大量的计算
158but are bleeding cash.但血流成河
159The giant data centre it is backing in Ohio is owned by SB Energy,它在俄亥俄州支持的巨型数据中心 由SB能源公司拥有
160a unit of SoftBank, a Japanese conglomerate.软邦克(日语:软邦克),是日本的一个联合体.
161OpenAI将成为租户.
162Nvidia, for its part,就恩维迪亚而言,
163will provide guarantees for the lease on the land and buildings for the power purchase contract将为购买电力合同的土地和建筑物的租赁提供担保
164on which it will rely.它将依赖它。
165In return, the project will use 1.5 million Nvidia processors with multiple作为回报,该项目将使用150万个有多个Nvidia处理器
166upgrade cycles expected over the life of the project.预计在项目实施期间会有升级周期。
167即使是Anthropic,它从购买芯片
168Amazon and Google, has not escaped this web.亚马逊与谷歌,
169Through a convoluted series of agreements,通过一系列复杂的协议,
170it has reportedly signed a $35 billion deal to rent cloud computing capacity from Lambda,据报道,它签署了350亿美元的交易 租借云计算能力 从Lambda,
171a Nvidia有利害关系的NeoCloud。
172Lambda则使用HUT8开发的设施.
173another NeoCloud whose entire capacity Nvidia had leased and may now have largely sublet.Nvidia已经租了另一辆NeoCloud,现在可能基本上有转租。
174Nvidia is exploring creative ways to draw more outside capital into such projects.Nvidia正在探索创新方法,以吸引更多外部资本参与这些项目。
175Its $500 billion partnership with big Wall Street firms will court institutional investors它与华尔街大公司 5000亿美元的伙伴关系 将法院机构投资者
176such as sovereign wealth funds, insurers and pension funds.例如主权财富基金、保险人和养恤基金。
177These investors will finance这些投资者将出资
178independent vehicles that will buy Nvidia hardware, build infrastructure and sell compute.独立车辆,将购买Nvidia硬件,建设基础设施和销售计算。
179Nvidia will neither contribute any cash up front for these deals nor take on any debt.Nvidia既不为这些交易提供任何现金,也不承担任何债务。
180In some在某些
181cases, though, it will provide residual value support of up to a quarter of the deal's total不过,它将提供最多占交易总额的四分之一的剩余价值支持。
182price to cover shortfalls in the value of the hardware involved after a certain period.价格以弥补所涉硬件价值在一段时间后出现的短缺。
183It这个
184will also provide technical support, giving lenders more confidence that the equipment they finance还将提供技术支持,使放款人更加相信他们资助的设备
185will be used well.将善用。
186Mr. Huang frames this as a way to make compute an investable asset class.黄先生将这个设定为计算可投资资产类别的方法。
187Underpinning this web of obligations are two fundamental assumptions, that Nvidia's chips这个义务网的基础有两个基本假设, 恩维迪亚的筹码
188will retain their value and that demand for compute will continue to grow at a rapid pace.计算需求将继续迅速增长。
189Neither is assured.也没有保证。
190Start with the chips.从筹码开始
191Mr. Huang says they are fungible and durable,黄先生说它们是可替换和耐用的
192with software upgrades enhancing their usefulness for years after they are sold.软件的升级在售出之后的数年内提高了它们的效用。
193They should be他们应该是
194seen as bankable assets, he argues, that can serve as collateral for loans and guarantees.他认为,可被视为银行可担保的资产,可作为贷款和担保的抵押。
195But how long an AI chip retains its value is a matter of debate.但AI芯片保留其价值多久是一个争论的问题.
196Mr.劳伦斯先生
197Bury has argued that big cloud providers have inflated profits by depreciating themBury认为,大云供应商通过贬值来夸大利润
198over five or six years rather than two or three.超过5年或6年,而不是2年或3年。
199So far Mr. Huang has the better of the argument.目前黄先生的辩词比较好.
200Older chips are still in demand.老筹码仍在需求之中。
2018月,CoreWeave宣布了合同
202involving Nvidia's A100 chips that runs into 2029.涉及Nvidia的A100芯片 运行到2029年.
203A100于2020年和Nvidia推出
204has since rolled out multiple newer models.之后推出了多个较新的模型。
205There is evidence, too, that old chips retain their也有证据表明 旧的芯片保留了他们的
206value.数值。
207Semi-Analysis estimates that renting Nvidia's H100, a workhorse of the AI industry,半解析估计,租借Nvidia的H100, AI行业的一匹工作马,
208costs around $2.80 an hour on a one-year contract.一个一年合同每小时花费2.80美元。
209That is only about a tenth less than when the chip这只比芯片少十分之一
210launched in early 2023.2023年初推出.
211Yet both the longevity and value of such chips may simply be artefacts of然而,这种芯片的寿命和价值都可能只是
212When compute is constrained, firms have no choice but to keep older chips humming.当计算受限时,企业别无选择,只能保持旧的芯片鸣叫.
213Many are using such chips for inference, meaning when an AI model responds to queries,许多人使用这样的芯片进行推论, 意思是当一个AI模型回答询问时,
214a far less demanding task than training models for which the newest hardware is used.与使用最新硬件的培训模式相比,一项艰巨得多的任务。
215Nvidia itself releases spiffy new chips every year, with the explicit intention of supersedingNvidia本身每年都会释放飞跃的新芯片,明确意图是取代
216their predecessors.他们的前任。
217That leaves it in the odd position of arguing both that old chips这让它处于奇怪的地位 争论这两个老筹码
218should retain their value and that its new chips should supplant them.应保留其价值,其新筹码应取代这些筹码。
219Finding a balance will be找到一个平衡
220tricky.难解.
221The price of compute should fall as supply increases.计算价格应随着供应的增加而下降。
222Hyperscalers and even labs超规模甚至实验室
223like OpenAI are rolling out purpose-built inference chips since spending on inference就像OpenAI在推论上花费时间 推出专有的推论芯片
224has overtaken spending on training.已经超过了培训开支。
225Nvidia enjoys enviable gross margins of around 75%Nvidia享有约75%的令人羡慕的总差额。
226against roughly 55% at AMD, a smaller rival.相对在AMD的大约55%, 一个较小的对手。
227Tim Davis 一个AI硬件公司的创始人
228acquired by Qualcomm, a chip designer, believes that as options proliferate,由芯片设计师Qualcomm收购,认为随着选项激增,
229the economics of chip making will start to compress.芯片制造的经济学将开始压缩。
230The biggest risk to chip prices is demand芯片价格的最大风险是需求
231itself.它本身。
232As long as spending on AI keeps booming, Nvidia sells its chips, its customers fill their只要花在AI上的钱不断繁荣 Nvidia卖掉它的芯片 客户就充斥着
233data centres and its various guarantees are seldom invoked.很少使用数据中心及其各种保障。
234But if demand is not quite as expansive如果需求没有那么大
235as markets expect, NeoClouds and other AI providers could struggle to sell their capacity or turn a正如市场所期望的那样,NeoClouds和其他AI供应商可能难以出售其能力或转成
236profit.利润。
237That in turn could trigger Nvidia's backstops, forcing it to buy unused compute,这反过来又会触发Nvidia的后台 迫使它购买未使用的计算,
238cover price shortfalls or pay for unwanted power.弥补价格短缺或支付不必要的电力。
239The same slowdown would also dent Nvidia's sales同样的减速也会证明Nvidia的销量
240and thus the cash flow available to meet those obligations.以及因此可用于履行这些义务的现金流量。
241This scenario does not hinge on a这一设想并不取决于
242collapse in demand.需求崩溃。
243Merely disappointing growth could conceivably spell trouble.仅仅令人失望的增长可能会造成麻烦。
244How much could Nvidia have to cough up?恩维迪亚要咳嗽多少?
245It has promised around $25 billion in future equity已经承诺了约250亿美元的未来股权
246investments.投资。
247It owes around $33 billion in debt.它欠债约330亿美元。
248Its potential liabilities to customers amount to它对客户的潜在负债相当于
249about $300 billion, but only come into play in a downturn and so do not appear on its balance sheet.大约3 000亿美元,但只是在经济下滑时才发挥作用,因此,不出现在资产负债表上。
250These include the $105 billion guarantee behind OpenAI's data centre, as much as $125 billion这些包括OpenAI数据中心背后的1 050亿美元的担保,高达1 250亿美元
251through the Wall Street Partnership and around $67 billion in other backstops.通过华尔街伙伴关系和约670亿美元的其他支助。
252Nvidia will恩维迪亚会的
253almost certainly not end up paying anything close to the full amount of its guarantees.几乎肯定不会最终支付任何接近其全部担保额的款项。
254Its commitments are spread over many years, so could not come due all at once.它的承诺分许多年,因此不可能一劳永逸。
255Its proposed提议
256backstop for OpenAI's data centre, for example, runs for 20 years starting in 2028.例如,OpenAI数据中心的支持站从2028年开始运行20年。
257如果 OpenAI 则
258stopped paying the lease, Nvidia could find another tenant, vastly diminishing its exposure.Nvidia可以找到另一个租户,
259What is more, Nvidia's own finances are strong enough to weather these liabilities.此外,Nvidia自己的财政足以应付这些负债。
260Morgan Stanley, an investment bank, reckons Nvidia's all-in debt will rise from $53 billionMorgan Stanley,一家投资银行, 估计Nvidia的全部债务将从530亿美元上升
261early next year to $200 billion by the beginning of 2029 as guarantees come into effect.明年年初,随着担保生效,到2029年初将达到2,000亿美元。
262But that is offset by a stash of cash and liquid securities currently worth $99 billion但被目前价值990亿美元的现金和流动证券所抵消
263and a business that will generate about $200 billion in cash this year.还有今年能赚到两千亿美元现金的企业
264Only a cataclysmic downturn that caused all Nvidia's guarantees to come due and its profits只有灾难性的衰退 导致Nvidia所有的担保 和它的利润
265to evaporate, almost entirely, would imperil the company as things stand.蒸发, 几乎完全, 将会危及公司 目前的情况。
266The picture may图片可能
267change, however, if Nvidia's commitments keep growing.然而,如果Nvidia的承诺继续增加,情况就会发生变化。
268Andy Lee of Credit Sites, 一个金融公司
269research firm, worries that it will keep pushing the pedal until something breaks.研究公司担心它会继续推踏板 直到有东西断裂
270Semi-Analysis半解析
271estimates that Nvidia takes on roughly $5.9 billion of guarantees for every 100 megawatts估计Nvidia每100兆瓦获得约59亿美元的担保
272of data centre capacity covered by its NeoCloud backstop programme.其NeoCloud支援站方案覆盖的数据中心容量。
273If it adds more such guarantees,如果它增加了更多的这种保证,
274Semi-Analysis reckons its exposure could reach $175 billion by the end of 2028 from this initiative半解析估计,到2028年底,该举措的接触量将达到1 750亿美元
275alone.一个人
276Nvidia is not alone in using ever more complex financial instruments to lubricateNvidia并非独自使用日益复杂的金融工具进行润滑
277customer demand.顾客需求。
278AMD has offered to sell big stakes in itself to OpenAI and Meta in exchangeAMD主动提出将大股权本身出售给OpenAI和Meta作为交换.
279for mammoth contracts to buy its chips.给猛毛腿公司买薯片
280In April, Google assembled a consortium of firms to help4月,Google召集了一批公司来帮助
281finance Anthropic's purchase of $35 billion of its chips.资助Anthropic购买了350亿美元的芯片.
282Broadcom,开发芯片在
283Google's behalf, agreed to cover any shortfall if Anthropic fails to pay.谷歌的代理,同意弥补任何短缺 如果Anthropic没有支付。
284Amid the race to lock in在锁定的比赛中
285sales and loans, there is a danger that speculative projects that might otherwise struggle to find投机性项目可能难以找到
286financing will get built.资金将会到位
287Should returns in the industry fall short or take longer to materialise如果行业回报不足或需要更长的时间才能实现
288than these convoluted deals assume, the result will be a profusion of expensive idle chips.比这些扭曲的交易假设,结果将是大量昂贵的闲置芯片。
289No company is more entangled in this web than Nvidia.没有哪个公司比Nvidia更纠缠在这个网上了.
290Mr Huang believes that the AI infrastructure黄先生认为人工智能基础设施
291build-out is at full steam.积聚在全蒸汽。
292But Nvidia's investors should be aware that the more of this boom they但Nvidia的投资者应该知道 他们越是繁荣
293finance, the more of any bust they will have to bear.资金,他们要承受的越多。
该视频共有字幕 293 条。解锁更多字幕为会员功能,请移动到 价格