AI热潮遇产能瓶颈：硬件供应危机显现

Hardware-makers are failing to invest enough to keep up with demand 硬件厂商投资乏力，追不上需求狂奔

本文揭示了AI热潮背后隐藏的硬件供应危机。随着AI需求激增，Anthropic、OpenAI等公司频频出现服务中断、限流甚至停服，根源在于算力供给跟不上。五大超大规模运营商（谷歌、亚马逊、Meta、微软、甲骨文）正以前所未有的规模投资数据中心，今年计划资本支出超7500亿美元。然而，硬件供应链的投资增速远远落后于需求：芯片制造商、设备商的资本支出仅增长一半，今年投资额不足云巨头的三分之一。从GPU、HBM高带宽内存到CPU，整个硬件链条都在告急。台积电等代工厂产能已拉满，新建晶圆厂需时两至三年，短期内无解。马斯克计划建造的"Terafab"最早也要到2028年才能投产。软件进步只需数月，供应链扩张却需数年——这场"tokenmaxxing"的狂欢，恐怕即将戛然而止。

原文 A newcrazehasgrippedSilicon Valleyof late. Techies looking to prove they are in thevanguardof artificial-intelligence adoptionhave taken to"tokenmaxxing", competing with one another to burn through the most tokens (as the chunks of text processed by AI models are known). Yet as demand for AIsoars, those tokens are in increasingly short supply.

翻译 近来，一股新狂潮攫住了硅谷。科技达人们为了证明自己站在AI应用的最前列，纷纷迷上了"tokenmaxxing"——互相攀比，看谁烧掉的tokens最多（tokens即AI模型处理的文本块）。然而，随着AI需求暴涨，这些tokens的供给正日益紧张。

原文 In March Anthropic, an AI lab whose models are popular with businesses, beganthrottlingaccess to its tools at busy times. It has since been altering its subscription plans, seeminglyin a bid tocurbusage. In April its service has experiencedoutagesof around 30 minutes a day. In March OpenAI, a rival,abruptlyshut Sora, its video-generation tool, to redirectscarcecomputing power. On April 20th GitHub, a coding-collaboration site that is owned by Microsoft, stopped accepting new subscriptions for its programming bot.

翻译 今年3月，AI实验室Anthropic——其模型深受企业青睐——开始在繁忙时段限制用户访问。此后它不断调整订阅方案，似乎意在控制用量。4月，其服务日均中断约30分钟。3月，竞争对手OpenAI突然关闭视频生成工具Sora，以便将稀缺算力重新分配。4月20日，微软旗下的编程协作平台GitHub停止接受其编程机器人的新订阅。

原文 They have had little choice. Demand is rising faster than they can add capacity: between January and March the weekly tokens processed by OpenRouter, a model marketplace,quadrupled. Meanwhile, the industry is racing to build new infrastructure. On April 20th Anthropic announced a $100bn partnership with Amazon tosecure up tofivegigawattsof server capacity, with nearly a fifth 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 said it was reworking its partnership with Microsoft to allow it to distribute all its products through any cloud provider, giving it greater flexibility totap intocomputing supply.

翻译 它们别无选择。需求增速已超扩容能力：1月至3月间，模型交易平台OpenRouter每周处理的tokens翻了两番。与此同时，整个行业正争相扩建基础设施。4月20日，Anthropic宣布与亚马逊达成1000亿美元合作，以锁定多达5千兆瓦的服务器容量，其中近五分之一将于年底前上线。4月24日，该公司表示谷歌也将投资400亿美元，帮助实验室满足算力需求。4月27日，OpenAI表示正重新协商与微软的合作关系，允许其通过任意云服务商分发所有产品，从而更灵活地获取算力供给。

原文 Fivehyperscalers—Alphabet, Amazon, Meta, Microsoft and Oracle —are investing ever larger sums of money in data centres. Alphabet, Amazon and Oracle have already raised more than $100bnin debtbetween them this year. To free up cash, Meta recently said that it wouldlay off10% of its workforce, while Microsoft said that it would offer voluntaryredundanciesto about 7% of its workers.

翻译 五大超大规模运营商——谷歌母公司Alphabet、亚马逊、Meta、微软和甲骨文——正不断加码数据中心投资。今年以来，Alphabet、亚马逊和甲骨文已合计举债逾千亿美元。为腾出资金，Meta近期表示将裁员10%，微软则计划向约7%的员工提供自愿离职方案。

原文 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. Thesqueezeon capacity, then,looks set toworsen.

翻译 然而，扩大产能正变得愈发困难。在美国及美国以外，针对数据中心建设的政治反对声浪日益高涨。更重要的是，那些制造填充数据中心的硬件——从芯片、网络设备到冷却设备——的公司，投资力度严重不足，根本跟不上需求步伐。那么，产能紧缺的局面，看来注定会进一步恶化。

原文 Start with the politics. In Aprillegislatorsin Maine voted in favour of a bill to ban the construction of data centres above 20megawattsuntil November next year. Although it was subsequentlyvetoedby the governor, lawmakers in more than ten other American states areweighingsimilar measures. According to one count, $156bn-worth of data-centre projects were blocked or delayed last year in America by local opposition andlitigation. Other countries, from Ireland to Brazil, are experiencing a growingbacklash. Concern over the impact ofpower-hungrydata centres on electricity billsin particularhas becomewidespread—and mayintensifyfurther as the war in the Gulf raises energy prices.

翻译 先看政治层面。今年4月，缅因州议员投票赞成一项法案，禁止建设容量超过20兆瓦的数据中心，禁令持续至明年11月。尽管该法案随后被州长否决，但美国另有十余个州的议员正在斟酌类似措施。据一项统计，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叫停或推迟。从爱尔兰到巴西，其他国家也日益出现强烈反弹。公众对耗电惊人的数据中心推高电费的影响担忧尤为普遍——而随着海湾地区战乱推高能源价格，这种担忧可能进一步加剧。

原文 Even when data centres are approved for construction and canget hooked up toa power source—whether the grid or, increasingly, their own means of generation—thoseerectingthem are finding it harder toget their hands onthe computing equipment needed to operate them.

翻译 即便数据中心已获准建设并能接通电源——无论是接入公共电网，还是采用如今愈发常见的自备发电方案——建设方仍发现，越来越难以采购到运营所需的计算设备。

原文 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 haveresortedto 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.

翻译 研究公司SemiAnalysis的Ivan Chiam指出，现有芯片不足以填满正在建设的数据中心。以英伟达设计的图形处理器（GPU）为例，它提供了全球逾三分之二的AI算力。自去年11月以来，2022年发布的H100 GPU租金价格飙升约30%，皆因客户拿不到新款，不得不转而使用旧款。竞争对手的AI处理器也愈发难以获得。今年4月，亚马逊掌门人Andy Jassy表示，公司Trainium2 AI芯片的访问权限已接近售罄。定于明年发布的Trainium4，其相当一部分产能早已被预订一空。

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

翻译 紧缺态势亦已蔓延至存储芯片，尤其是AI模型所依赖的高带宽内存（HBM）。三大生产商——SK海力士、三星和美光——均表示，其2026年的大部分供货已被预订一空。3月曾现一线缓解希望：谷歌发布了TurboQuant算法，旨在削减AI所需的内存量，致使存储芯片制造商的股价短暂下挫。即便如此，未来至少三年内，HBM的需求仍将超过供给。

原文 The shortages are now spreading to central-processing units (CPUs). "Agentic" AI tools, which plan, reason and carry out tasks, rely more heavily on these types of chips to coordinate 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 sorobustthat it hasbreathed new life intoIntel, which not long ago seemed to be heading for collapse. The marketcapitalisationof the American chipmaker, one of the leading producers of CPUs, has more than doubled over the past six months.

翻译 短缺局面现已蔓延至中央处理器（CPU）。能够规划、推理并执行任务的"智能体式"AI工具，在协调工作时会更重度地依赖这类芯片。投资银行摩根士丹利估计，智能体系统每配置1个GPU就需要搭配1个CPU，而聊天机器人式系统的配比为1比12。事实上，CPU需求之旺盛，竟让不久前还看似走向崩溃的英特尔重获生机。这家美国芯片制造商是全球领先的CPU生产商之一，其市值在过去六个月内翻了一倍多。

原文 Thecruxof the problem is that companies along the AI supply chain are investing far less than the hyperscalers. 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. The five hyperscalers havetripledtheir combined capital spending, to more than $750bn, but the hardware suppliers have increased theirs by only half, and will invest less than a third as much as the cloud giants this year.

翻译 问题的症结在于，AI供应链上的企业投资规模远逊于超大规模运营商。我们调研了约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，及其自2024年以来的变化。五大超大规模运营商的合计资本支出增长了两倍，达到7500亿美元以上，但硬件供应商的支出仅增长一半，今年投资额将不足云巨头的三分之一。

原文 TakeTSMC, the world's biggest contract chipmaker and thedominantsupplier ofcutting-edgeGPUs and CPUs. Its most advancedfabs—those making chips that are fivenanometresor smaller—are already runningflat out. C.C. Wei, its boss, admits 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 about65bn in 2027. But as a share of sales, its capital expenditure has fallen from around half in 2022 to a third this year.

翻译 且看台积电——全球最大的合约芯片制造商，也是尖端GPU与CPU的主导供应商。其最先进的晶圆厂——生产5纳米及以下制程芯片的产线——早已开足马力。台积电掌门人魏哲家承认供应"非常紧张"，但"没有捷径可走"：新建一座晶圆厂需要两到三年。公司计划2026年投入约550亿美元，较2025年增长34%；分析师预计2027年这一数字将升至650亿美元。然而，资本支出占销售额的比例，已从2022年的约一半降至今年的三分之一。

原文 TSMC's cautionhasfrustratedits customers. Sam Altman, OpenAI's boss, has urged it to "just build more capacity". Elon Musk, boss of Tesla and SpaceX, has said he will build a so-called "Terafab" with the modest ambition ofchurning outmore processing power annually than the entire global semiconductor industry today.

翻译 台积电的谨慎令其客户颇感挫败。OpenAI掌门人Sam Altman已敦促其"只管扩产便是"。特斯拉与SpaceX的老板埃隆·马斯克则表示，他将建造一座所谓"Terafab"，怀揣一个"谦逊的野心"——每年产出的算力超过当今全球半导体产业的总和。

原文 The facility, which Mr. Musk hasenlistedIntel to help set up, is unlikely to start production until 2028 at the earliest, and even then at a fraction of the scaleenvisaged. 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 arewary ofover-building and being stuck withidle capacity. The craze for "tokenmaxxing" may soon becut short.

翻译 这座设施——马斯克已邀英特尔助其建设——最早也要到2028年方能投产，即便届时开工，规模也仅为当初设想的一小部分。更甚者，马斯克恐怕难以筹齐运营这座工厂所需的足量先进设备——而那些设备，同样供不应求。这一切，恰恰映现出当下正笼罩AI未来的供需失衡。软件迭代，不过数月之功；供应链扩张，却需数年之力。硬件制造商对过度扩建心存忌惮，唯恐背上产能过剩的包袱。"tokenmaxxing"的热潮，恐怕即将戛然而止。

原文：A new craze has gripped Silicon Valley of late. Techies looking to prove they are in the vanguard of artificial-intelligence adoption have taken to "tokenmaxxing", competing with one another to burn through the most tokens (as the chunks of text processed by AI models are known).

翻译：近来，一股新的狂热席卷了硅谷。科技人士为了证明自己站在人工智能应用的最前沿，纷纷迷上了"tokenmaxxing"——比拼谁消耗的AI tokens最多。

解析：

原文：In March Anthropic, an AI lab whose models are popular with businesses, began throttling access to its tools at busy times. It has since been altering its subscription plans, seemingly in a bid to curb usage.

翻译：今年3月，AI实验室Anthropic——其模型深受企业用户欢迎——开始在繁忙时段限制工具访问。此后它不断调整订阅方案，似乎意在压制使用量。

解析：

原文：Demand is rising faster than they can add capacity: between January and March the weekly tokens processed by OpenRouter, a model marketplace, quadrupled.

翻译：需求增长速度超过了扩容速度：1月至3月间，模型交易平台OpenRouter每周处理的tokens增长了三倍。

解析：

原文：On April 20th Anthropic announced a $100bn partnership with Amazon to secure up to five gigawatts of server capacity, with nearly a fifth to come online by the end of the year.

翻译：4月20日，Anthropic宣布与亚马逊达成1000亿美元合作，以确保最多5千兆瓦的服务器容量，其中近五分之一将在年底前上线。

解析：

原文：To free up cash, Meta recently said that it would lay off 10% of its workforce, while Microsoft said that it would offer voluntary redundancies to about 7% of its workers.

翻译：为腾出资金，Meta近期表示将裁员10%，微软则称将向约7%的员工提供自愿裁员方案。

解析：

原文：The crux of the problem is that companies along the AI supply chain are investing far less than the hyperscalers.

翻译：问题的症结在于，AI供应链上的企业投资规模远逊于超大规模运算符。我们对约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进行了调研，考察了它们今年的计划资本支出及自2024年以来的变化。五大超大规模运算符的合计资本支出已增至三倍，超过7500亿美元，但硬件供应商的支出仅增长一半，今年投资额将不足云巨头的三分之一。

解析：

原文：Its most advanced fabs—those making chips that are five nanometres or smaller—are already running flat out.

翻译：其最先进的晶圆厂——生产5纳米及以下制程芯片的产线——已满负荷运转。

解析：

原文：C.C. Wei, its boss, admits that supply is "very tight", but that "there are no shortcuts": building a new fab takes two to three years.

翻译：台积电掌门人魏哲家承认供应"非常紧张"，但"没有捷径可走"：新建一座晶圆厂需要两到三年。

解析：

原文：Elon Musk, boss of Tesla and SpaceX, has said he will build a so-called "Terafab" with the modest ambition of churning out more processing power annually than the entire global semiconductor industry today.

翻译：特斯拉和SpaceX的老板埃隆·马斯克则表示，他将建造一座所谓"Terafab"，其"modest ambition"（modest此处为讽刺用法）是每年产出的算力超过当今全球半导体产业的总和。

解析：

原文：Improving software takes months, whereas expanding supply chains takes years.

翻译：改进软件只需数月，而扩张供应链却需数年。

解析：