AI面临供应链危机,英伟达GPU价格飙升、HBM内存售罄 | 经济学人

🤖 导读

AI需求激增正遭遇硬件供应链危机。芯片、网络设备、冷却系统等制造商投资远不及科技巨头，导致算力短缺。政治阻力也在增加，数据中心项目遭延迟。虽然英伟达GPU价格飙升、HBM内存售罄，但台积电等厂商扩产谨慎。AI热潮可能很快撞上供给天花板。

AI is confronting a supply-chain crunch

人工智能正面临供应链危机

Hardware-makers are failing to invest enough to keep up with demand

硬件制造商投资不足，难以跟上需求

—

Para 1:

Anew craze has taken hold of Silicon Valley in recent months. Techies looking to prove they are among the vanguard of artificial-intelligence adoption have taken to “tokenmaxxing”, competing against one another to burn through the most tokens (as the chunks of text processed by AI models are known). Between January and March, the weekly tokens processed by OpenRouter, a marketplace for accessing models, quadrupled.

近几个月来，硅谷掀起了一股新热潮。那些希望证明自己处于人工智能应用前沿的技术极客们开始“token最大化”，相互竞争消耗最多的token（即AI模型处理的文本块）。今年1月至3月间，模型接入市场OpenRouter每周处理的token数量翻了两番。

📌 生词/短语:

· craze /kreɪz/ n. 狂热，热潮

· take hold of /teɪk həʊld əv/ 占据，蔓延

· vanguard /ˈvænɡɑːd/ n. 先锋，前沿

· token /ˈtəʊkən/ n. （AI模型处理的）文本单元

· burn through /bɜːn θruː/ 大量消耗

· quadruple /kwɒˈdruːpl/ v. 翻两番，变为四倍

📌 翻译重难点:

· “tokenmaxxing” 为自造词，保留引号并意译为“token最大化”，加括号说明。

· “chunks of text” 译为“文本块”通俗易懂。

📌 文化背景:

Token 是大语言模型计费和处理的基本单位，约等于0.75个英文单词。OpenRouter 是一个聚合多个AI模型的API市场。

—

Para 2:

As demand for AI soars, the industry behind it is struggling to keep up. In March Anthropic, an AI lab whose models are popular with businesses, began throttling access to its tools at busy times, and has since been altering its subscription plans seemingly in a bid to curb usage. In April its service has experienced outages averaging around 30 minutes per day. And it is not alone. In March OpenAI, a rival, abruptly shut Sora, its video-generation tool, to redirect scarce computing power toward more lucrative uses. On April 20th GitHub, a coding-collaboration site owned by Microsoft, stopped accepting new subscriptions for its programming bot.

随着AI需求飙升，其背后的产业正努力追赶。3月，模型深受企业欢迎的AI实验室Anthropic开始在高峰时段限制对其工具的访问，此后一直在调整订阅方案，似乎是为了遏制使用量。4月，其服务平均每天出现约30分钟的中断。这并非孤例。3月，竞争对手OpenAI突然关闭了其视频生成工具Sora，将稀缺的算力重新分配给更有利可图的用途。4月20日，微软旗下的代码协作网站GitHub停止接受其编程机器人的新订阅。

📌 生词/短语:

· soar /sɔː/ v. 飙升

· throttle /ˈθrɒtl/ v. 限制，节流

· curb /kɜːb/ v. 抑制

· outage /ˈaʊtɪdʒ/ n. 中断，停运

· abruptly /əˈbrʌptli/ adv. 突然地

· lucrative /ˈluːkrətɪv/ adj. 盈利的，有利可图的

· programming bot /ˈprəʊɡræmɪŋ bɒt/ n. 编程机器人

📌 翻译重难点:

· “throttling access” 动词throttle原指“节流阀”，技术语境下译为“限制访问”。

· “redirect…toward” 译为“重新分配给”。

📌 文化背景:

Anthropic由前OpenAI员工创立，主打Claude系列模型。GitHub Copilot是首个大规模商用的AI编程助手，2025年曾因算力不足暂停新用户注册。

—

Para 3:

The industry’s response has been to pour ever larger sums of money into new infrastructure. On April 20th Anthropic announced a $100bn partnership with Amazon to secure up to five gigawatts (GW) of server capacity, with nearly a fifth due to come online by the end of the year. On April 24th it said that Google would also invest $40bn to help the lab meet its computing needs. On April 27th OpenAI announced that it was reworking its partnership with Microsoft to allow it to distribute all its products through any cloud provider, giving it greater flexibility to tap into computing supply.

业界的应对之策是向新基础设施投入更多资金。4月20日，Anthropic宣布与亚马逊达成1000亿美元的合作，以获得高达5吉瓦的服务器容量，其中近五分之一将在年底前上线。4月24日，该公司表示谷歌也将投资400亿美元，以帮助该实验室满足其计算需求。4月27日，OpenAI宣布正在重新调整与微软的合作关系，允许其通过任何云提供商分发所有产品，从而获得更大的灵活性来获取计算资源。

📌 生词/短语:

· infrastructure /ˈɪnfrəstrʌktʃə/ n. 基础设施

· gigawatt /ˈɡɪɡəwɒt/ n. 吉瓦（十亿瓦特）

· come online /kʌm ˈɒnlaɪn/ 上线，投入使用

· rework /riːˈwɜːk/ v. 重新调整

· tap into /tæp ˈɪntə/ 利用，获取

📌 翻译重难点:

· “pour…money into” 形象表达，译为“投入……资金”。

· “five gigawatts of server capacity” 将电能单位用于衡量数据中心算力，保留直译加注。

📌 文化背景:

5GW约相当于5座核电站的发电量，可见AI数据中心的耗电规模。OpenAI与微软的独家云协议此前曾引发反垄断担忧。

—

Para 4:

The five so-called hyperscalers—Alphabet, Amazon, Meta, Microsoft and Oracle—are investing hundreds of billions of dollars apiece in data centres. Alphabet, Amazon and Oracle have already raised more than $100bn in debt between them this year. To free up cash, Meta recently announced that it would lay off 10% of its workforce, while Microsoft said that it would offer voluntary redundancies to about 7% of its workers.

所谓的五大超大规模企业——Alphabet、亚马逊、Meta、微软和甲骨文——各自在数据中心上投资了数千亿美元。Alphabet、亚马逊和甲骨文今年已合计发行超过1000亿美元的债务。为释放现金，Meta最近宣布将裁员10%，而微软则表示将向约7%的员工提供自愿离职方案。

📌 生词/短语:

· hyperscaler /ˈhaɪpəskeɪlə/ n. 超大规模企业（云巨头）

· apiece /əˈpiːs/ adv. 每个，各

· raise debt /reɪz det/ 发行债务/融资

· lay off /leɪ ɒf/ 裁员

· voluntary redundancy /ˈvɒləntəri rɪˈdʌndənsi/ 自愿裁员/自愿离职

📌 翻译重难点:

· “raised more than $100bn in debt” 译为“发行超过1000亿美元的债务”，更符合金融语境。

· “free up cash” 译为“释放现金”或“腾出现金”。

📌 文化背景:

超大规模企业指运营大规模数据中心的云计算巨头。甲骨文近年积极扩张云业务，试图挑战AWS、Azure和GCP。

—

Para 5:

Adding more capacity, however, is only getting more challenging. In America and beyond, political opposition to the construction of data centres is growing. What is more, the companies making the hardware that fills them—from chips and networking gear to cooling equipment—have been investing far too little to keep pace with demand. The squeeze on capacity, then, looks set to worsen.

然而，增加更多容量正变得愈发困难。在美国及其他地区，反对建设数据中心的政治声浪日益高涨。更重要的是，从芯片、网络设备到冷却设备，制造数据中心所需硬件的公司投资远远不足，无法跟上需求。因此，算力紧缺看来还将恶化。

📌 生词/短语:

· capacity /kəˈpæsəti/ n. 容量，能力

· networking gear /ˈnetwɜːkɪŋ ɡɪə/ n. 网络设备

· cooling equipment /ˈkuːlɪŋ ɪˈkwɪpmənt/ n. 冷却设备

· squeeze /skwiːz/ n. 紧缺，挤压

📌 翻译重难点:

· “The squeeze on capacity” 中squeeze有“挤压/紧缺”双重含义，译为“算力紧缺”。

· “looks set to worsen” 译为“看来还将恶化”，表达趋势判断。

—

Para 6:

Start with the politics. In April legislators in Maine voted in favour of a bill to ban the construction of data centres above 20 megawatts until November 2027. Although it was subsequently vetoed by the governor, lawmakers in more than ten other American states are weighing similar measures. According to one count, $156bn-worth of data-centre projects were blocked or delayed last year in America by local opposition and litigation. Other countries, from Ireland to Brazil, are experiencing a growing backlash. Concern over the impact of power-hungry data centres on electricity bills in particular has become widespread—and may intensify further as the war in the Gulf raises energy prices.

先从政治因素说起。4月，缅因州立法者投票通过一项法案，禁止在2027年11月前建设20兆瓦以上的数据中心。尽管该法案随后被州长否决，但另有十多个州的立法者正在考虑类似措施。据统计，去年美国有价值1560亿美元的数据中心项目因当地反对和诉讼而被搁置或延迟。从爱尔兰到巴西，其他国家也正遭遇日益强烈的反弹。人们对耗电巨大的数据中心影响电费账单的担忧尤其普遍——而随着海湾战争推高能源价格，这种担忧可能进一步加剧。

📌 生词/短语:

· legislator /ˈledʒɪsleɪtə/ n. 立法者

· veto /ˈviːtəʊ/ v./n. 否决

· weigh /weɪ/ v. 权衡，考虑

· litigation /ˌlɪtɪˈɡeɪʃən/ n. 诉讼

· backlash /ˈbæklæʃ/ n. 强烈反对，反弹

· power-hungry /ˈpaʊə ˈhʌŋɡri/ adj. 耗电巨大的

📌 翻译重难点:

· “war in the Gulf” 需结合文章发布时间（2026年4月）理解，指当时海湾地区的冲突。

· “20 megawatts” 兆瓦，1MW=1000千瓦，20MW约可供电1.6万户家庭。

📌 文化背景:

爱尔兰因数据中心耗电量已占全国用电量近20%而暂停新建项目。缅因州法案虽被否决，但反映了地方对能源消耗的焦虑。

—

Para 7:

Even when data centres are approved for construction and can get hooked up to a power source—be that the grid or, increasingly, their own means of generation—those erecting them are finding it harder to get their hands on the computing equipment needed to operate them.

即使数据中心获批建设并能接入电源——无论是电网还是越来越常见的自备发电设施——建设者也发现越来越难获得运营所需的计算设备。

📌 生词/短语:

· get hooked up /ɡet hʊkt ʌp/ 接入，连接

· grid /ɡrɪd/ n. 电网

· means of generation /miːnz əv ˌdʒenəˈreɪʃən/ 发电方式/设施

· erect /ɪˈrekt/ v. 建造，建设

📌 翻译重难点:

· “be that…or…” 让步结构，译为“无论是……还是……”。

· “get their hands on” 口语化表达，译为“获得”。

—

Para 8:

Ivan Chiam of SemiAnalysis, a research firm, points out that there are not enough chips to fill the data centres now being built. Consider the graphics-processing units (GPUs) designed by Nvidia, which provide more than two-thirds of the world’s AI computing power. The price to rent one of its H100 GPUs, launched in 2022, has soared by around 30% since November, as customers unable to get their hands on newer models have looked to older generations. Competing AI processors are also getting more difficult to obtain. In April Andy Jassy, Amazon’s boss, said that his company had nearly sold out access to its Trainium2 AI chips. A significant chunk of the capacity of Trainium4, due next year, “has already been reserved”.

研究公司SemiAnalysis的Ivan Chiam指出，没有足够的芯片来填充目前正在建设的数据中心。以英伟达设计的图形处理器为例，它们提供了全球超过三分之二的AI算力。其2022年推出的H100 GPU的租赁价格自去年11月以来已飙升约30%，因为无法获得新模型的客户转而寻求上一代产品。竞争对手的AI处理器也变得越来越难获取。4月，亚马逊老板安迪·贾西表示，公司旗下Trainium2 AI芯片的算力已几乎售罄。预计明年推出的Trainium4的相当一部分容量“已经被预订”。

📌 生词/短语:

· graphics-processing unit (GPU) /ˈɡræfɪks ˈprəʊsesɪŋ ˈjuːnɪt/ n. 图形处理器

· launch /lɔːntʃ/ v. 推出

· soar /sɔː/ v. 飙升

· processor /ˈprəʊsesə/ n. 处理器

· sell out /sel aʊt/ 售罄

· reserve /rɪˈzɜːv/ v. 预订

📌 翻译重难点:

· “access to its Trainium2 AI chips” 芯片本身不卖，卖的是算力访问权，译为“Trainium2 AI芯片的算力”。

· “a significant chunk” 译为“相当大一部分”。

📌 文化背景:

英伟达H100是2022-2024年AI训练的主力芯片，后因H200和B200推出而供需紧张。AWS的Trainium系列是自研AI芯片，旨在挑战英伟达。

—

Chart: The Economist

图表：《经济学人》

—

Para 10:

The squeeze also extends to memory chips, in particular the kind of high-bandwidth memory (HBM) that AI models rely on. All three big producers—SK Hynix, Samsung and Micron—say that most of their supply for 2026 is already sold out. Some hope of relief came in March when Google unveiled TurboQuant, an algorithm meant to reduce the amount of memory AI needs, causing the share prices of the memory-makers to briefly swoon. Even so, demand for HBM is expected to outstrip supply for at least the next three years.

紧缺也延伸到了内存芯片，尤其是AI模型所依赖的高带宽内存（HBM）。三大厂商——SK海力士、三星和美光——均表示，其2026年的大部分供应已售罄。3月，谷歌发布了TurboQuant算法，旨在减少AI所需的内存量，导致内存制造商的股价短暂下跌，给市场带来了一丝缓解的希望。即便如此，预计至少未来三年内HBM的需求仍将超过供给。

📌 生词/短语:

· high-bandwidth memory (HBM) /haɪ ˈbændwɪdθ ˈmeməri/ n. 高带宽内存

· unveil /ʌnˈveɪl/ v. 发布，推出

· algorithm /ˈælɡərɪðəm/ n. 算法

· swoon /swuːn/ v. （股价）暴跌，（人）昏厥

· outstrip /aʊtˈstrɪp/ v. 超过，超越

📌 翻译重难点:

· “briefly swoon” 金融用语，译为“短暂下跌”，避免直译“昏厥”。

· “outstrip supply” 译为“超过供给”。

📌 文化背景:

HBM是高性能AI芯片的关键组件，通过堆叠DRAM实现超高带宽。SK海力士是HBM市场领导者，为英伟达主要供应商。谷歌TurboQuant通过算法优化减少内存占用。

—

Para 11:

The shortages are now spreading to central-processing units (CPUs). “Agentic” AI tools that plan, reason and carry out tasks rely more heavily on these types of chips to co-ordinate their work. Morgan Stanley, an investment bank, estimates that agentic systems require one CPU for every GPU, compared with a ratio of one to 12 for chatbot-style systems. Indeed, demand for CPUs has been so robust that it has breathed new life into Intel, which not long ago looked like it was headed for collapse. The market capitalisation of the American chipmaker, which is one of the leading producers of CPUs, has more than doubled over the past six months (see chart 1).

短缺正蔓延至中央处理器（CPU）。能够规划、推理和执行任务的“智能体式”AI工具更加依赖这类芯片来协调工作。投行摩根士丹利估计，智能体系统每块GPU需要搭配一块CPU，而聊天机器人系统的比例仅为1:12。事实上，CPU需求如此强劲，以至于为不久前看似走向崩溃的英特尔注入了新生。这家美国芯片制造商是CPU的主要生产商之一，其市值在过去六个月内翻了一倍多（见图表1）。

📌 生词/短语:

· central-processing unit (CPU) /ˈsentrəl ˈprəʊsesɪŋ ˈjuːnɪt/ n. 中央处理器

· agentic /eɪˈdʒentɪk/ adj. 智能体的，代理的

· co-ordinate /kəʊˈɔːdɪneɪt/ v. 协调

· robust /rəʊˈbʌst/ adj. 强劲的

· breathe new life into /briːð njuː laɪf ˈɪntə/ 注入新生

· market capitalisation /ˈmɑːkɪt ˌkæpɪtəlaɪˈzeɪʃən/ n. 市值

📌 翻译重难点:

· “Agentic AI” 译为“智能体式AI”，与“聊天机器人式”形成对比。

· “looked like it was headed for collapse” 译为“看似走向崩溃”。

📌 文化背景:

智能体AI（Agentic AI）指能自主完成多步骤任务的AI系统，如AutoGPT。英特尔曾因制程落后陷入危机，后因AI CPU需求回升。

—

Para 12:

Semi-detached

半脱离

📌 注： 此为小标题，原文如此，为双关语。“Semi-detached” 字面意为“半独立式房屋”，此处暗指供应链中硬件厂商与超大规模企业的“半脱钩”状态。译为“半脱离”保留双关意味。

—

Para 13:

The crux of the problem is that companies along the AI supply chain are investing far less than the hyperscalers in expanding their capacity. We examined the planned capital spending this year of the 50 or so largest manufacturers of chips, chipmaking tools, servers, networking gear and cooling equipment, and how it has changed since 2024. Over that period, the five hyperscalers have increased their combined capital spending by 190%, from $234bn to $677bn, whereas the hardware suppliers they rely on have increased theirs by only 45%, from $153bn to $223bn (see chart 2).

问题的核心在于，AI供应链上的公司在扩大产能方面的投资远低于超大规模企业。我们考察了约50家最大的芯片、芯片制造设备、服务器、网络设备和冷却设备制造商今年的计划资本支出，以及自2024年以来的变化情况。在此期间，五大超大规模企业的合计资本支出增长了190%，从2340亿美元增至6770亿美元，而它们所依赖的硬件供应商仅增长了45%，从1530亿美元增至2230亿美元（见图表2）。

📌 生词/短语:

· crux /krʌks/ n. 核心，关键

· capital spending /ˈkæpɪtəl ˈspendɪŋ/ n. 资本支出

· manufacturer /ˌmænjʊˈfæktʃərə/ n. 制造商

· combined /kəmˈbaɪnd/ adj. 合计的，共同的

· whereas /weərˈæz/ conj. 而，然而

📌 翻译重难点:

· “the crux of the problem” 译为“问题的核心”。

· 数字对比是关键信息，需清晰呈现增长率与绝对值的对应关系。

📌 文化背景:

资本支出是衡量企业扩产意愿的关键指标。硬件厂商投资保守源于半导体行业周期性波动带来的历史教训。

—

Para 14:

Take TSMC, the world’s biggest contract chipmaker and the dominant supplier of cutting-edge GPUs and CPUs. Its most advanced fabs—those making chips that are five nanometres or smaller—are already running flat out. C.C. Wei, its boss, has admitted that supply is “very tight”, but that “there are no shortcuts”: building a new fab takes two to three years. The company plans to spend about $55bn in 2026, up by 34% from a year earlier; analysts expect the figure to rise to $65bn in 2027. As a share of sales, however, its capital expenditure has fallen from around half in 2022 to a third this year.

以全球最大的芯片代工厂、尖端GPU和CPU的主要供应商台积电为例。其最先进的工厂——生产5纳米及更小制程芯片的工厂——已在满负荷运转。其老板魏哲家承认供应“非常紧张”，但“没有捷径”：建造一座新工厂需要两到三年时间。该公司计划2026年支出约550亿美元，同比增长34%；分析师预计2027年这一数字将增至650亿美元。然而，占销售额的比重，其资本支出已从2022年的一半左右降至今年的三分之一。

📌 生词/短语:

· contract chipmaker /ˈkɒntrækt ˈtʃɪpmeɪkə/ n. 芯片代工厂

· dominant /ˈdɒmɪnənt/ adj. 主导的

· cutting-edge /ˈkʌtɪŋ edʒ/ adj. 尖端的

· fab /fæb/ n. 晶圆厂（fabrication plant缩写）

· flat out /flæt aʊt/ 满负荷，全力以赴

· shortcut /ˈʃɔːtkʌt/ n. 捷径

📌 翻译重难点:

· “running flat out” 译为“满负荷运转”。

· “As a share of sales” 译为“占销售额的比重”，注意与“资本支出占销售额比例”概念对应。

📌 文化背景:

台积电是英伟达、AMD、苹果等公司的核心供应商。5纳米及以下制程目前仅台积电、三星等少数厂商能够生产。新建晶圆厂需2-3年、耗资超200亿美元。

—

Para 15:

TSMC’s caution has frustrated its customers. Sam Altman, OpenAI’s boss, has urged the firm to “just build more capacity”. In March Elon Musk, boss of Tesla and SpaceX, announced plans to build a so-called “Terafab” with the modest ambition of churning out more processing power annually than the entirety of the global semiconductor industry today.

台积电的谨慎令其客户感到沮丧。OpenAI老板萨姆·奥尔特曼敦促该公司“尽管扩产就好”。3月，特斯拉和SpaceX的老板埃隆·马斯克宣布计划建造所谓的“Terafab”，其“谦虚的”目标是每年产出的算力超过当今全球半导体产业的总和。

📌 生词/短语:

· caution /ˈkɔːʃən/ n. 谨慎

· frustrate /frʌˈstreɪt/ v. 使沮丧，使受挫

· urge /ɜːdʒ/ v. 敦促

· modest /ˈmɒdɪst/ adj. 谦虚的（此处为反讽）

· churn out /tʃɜːn aʊt/ 大量生产

📌 翻译重难点:

· “Terafab” 造词，由Tera-（万亿）+ Fab（晶圆厂）构成，音译为“Terafab”加引号。

· “the modest ambition” 反讽语气，译为“谦虚的目标”加引号。

📌 文化背景:

马斯克2024年曾称AI算力将每年增长10倍。Terafab计划被业界普遍认为是夸大其词。奥尔特曼此前也曾与台积电洽谈自建芯片工厂。

—

Para 16:

The facility, which Mr Musk has enlisted Intel to help set up, is unlikely to start production until 2028 at the earliest, and even then at a fraction of the scale envisioned. What is more, Mr Musk may struggle to get his hands on enough of the advanced machines he will need to operate it, which are also in short supply. That illustrates the mismatch that now clouds the future of AI. Improving software takes months, whereas expanding supply chains takes years. Hardware-makers are wary of over-building and being stuck with idle capacity. The craze for “tokenmaxxing”, then, might soon be cut short.

马斯克已请英特尔协助建设的这一设施，最早也要到2028年才能投产，而且届时也只能达到其设想规模的一小部分。此外，马斯克可能难以获得足够多的先进设备来运营该工厂——这些设备本身也供应短缺。这凸显了目前笼罩AI未来的错配问题。改进软件只需数月，而扩展供应链需要数年。硬件制造商担心过度建设后陷入产能闲置。那么，“token最大化”的热潮可能很快就要戛然而止了。

📌 生词/短语:

· enlist /ɪnˈlɪst/ v. 争取， enlisted sb to help 请…协助

· at a fraction of /æt ə ˈfrækʃən əv/ 仅为…的一小部分

· envision /ɪnˈvɪʒən/ v. 设想，想象

· illustrate /ˈɪləstreɪt/ v. 说明，彰显

· cloud /klaʊd/ v. 笼罩，使蒙上阴影

· be wary of /bi ˈweəri əv/ 对…保持警惕

· idle /ˈaɪdl/ adj. 闲置的

· cut short /kʌt ʃɔːt/ 提前结束，中断

📌 翻译重难点:

· “at a fraction of the scale envisioned” 译为“只能达到其设想规模的一小部分”。

· “clouds the future” 译为“笼罩……的未来”，形象表达。

· “cut short” 译为“戛然而止”，符合狂热突然中断的语境。

📌 文化背景:

芯片制造设备中最关键的是ASML的极紫外光刻机（EUV），年产量仅数十台，订单已排到2028年后。硬件厂商因2000年互联网泡沫和2023年半导体寒冬而拒绝激进扩产。