AI 崩溃之后
原文：After the AI Crash | 作者：Doug Dawson, CCG Consulting | 日期：2026-07-29
📝 摘要
Doug Dawson 从电信行业分析师的视角出发，系统梳理了 AI 行业正面临的七大结构性风险：每年 2 万亿美元的资本支出难以为继、少数巨头间的循环收入、依赖债务而非股权的融资模式、地方政府与公众对数据中心的抵制、企业因成本过高而缩减 AI 使用、新一代模型资源消耗递增造成的”规模不经济”，以及穆迪等机构的信用警告。他借鉴 2000 年科技泡沫崩溃的历史，推测 AI 崩溃后的连锁反应——从 20 万亿美元财富蒸发到电力水务公司的沉没成本转嫁，最终论证崩溃可能带来的长期正面效应：迫使行业实现真正的效率和规模经济。
📋 术语表
| 英文 | 中文 | 说明 |
|---|---|---|
| capital expenses (CapEx) | 资本支出 | 用于购置或升级长期资产的投入 |
| circular revenues | 循环收入 | 少数公司之间相互投资和采购形成的封闭收入链条 |
| diseconomies of scale | 规模不经济 | 规模扩大反而导致单位成本上升的反常现象 |
| economies of scale | 规模经济 | 规模扩大带来单位成本下降的正向效应 |
| CLEC | 竞争性本地交换运营商 | 2000 年电信泡沫中大量涌现又迅速消失的电信公司 |
| middle-mile fiber | 中程光纤 | 连接数据中心与主干网络的中间传输层光纤 |
| stranded investments | 沉没投资 | 已投入但因项目终止无法收回的资产 |
| ratepayers | 缴费用户 | 公共事业（电力、水务）的终端付费用户 |
正文（双语对照）
Everything I read about the AI industry leads me to think there will be an AI crash. Consider the following:
我读到的一切关于 AI 行业的信息都让我相信，一场 AI 崩溃即将来临。不妨看看以下几点。
Unsustainable Capital Expenses. It’s hard to imagine there can ever be enough revenue to pay for the huge capital investments in data centers and electronics. Several analysts have estimated that it will take $2 trillion a year in revenue to pay for the infrastructure that has already been built, and there are no believable forecasts for generating even half that much revenue. The capital needs of the industry are relentless since expensive AI data center electronics have to be replaced within five years, or less.
不可持续的资本支出。很难想象能产生足够的收入来支撑数据中心和电子设备方面巨大的资本投入。多位分析师估计，仅维持已建成的基础设施，每年就需要 2 万亿美元的收入，而即使是达到这个数字一半的可信预测都不存在。AI 行业的资本需求是无休止的——昂贵的数据中心电子设备必须在五年甚至更短时间内更换。
Circular Revenues. A small handful of tech firms, chip manufacturers, and AI companies are propping each other up by investing and buying from each other. If one stumbles, they might all fall.
循环收入。少数几家科技公司、芯片制造商和 AI 企业通过互相投资和采购来支撑彼此。只要一家倒下，可能全部都会崩塌。
Huge Debt. Much of the industry is being funded through debt, which has to eventually be repaid, instead of through equity.
巨额债务。AI 行业的大部分资金来自债务而非股权融资，而债务终究要偿还。
Public Pushback. Local governments and people are increasingly pushing back hard against the creation of new data centers. Most new technologies have been welcomed by the public with open arms.
公众抵制。地方政府和民众正日益强烈地抵制新建数据中心。历史上，大多数新技术都曾被公众张开双臂欢迎。
Increasing Corporate Skepticism. The news is full of stories of corporations that are throttling the employee use of AI since the costs to use the software are a lot higher than expected. There are many companies having second thoughts about replacing people with AI. The AI industry needs complete corporate buy-in to have any chance of succeeding, and large companies are generally still on the sidelines.
日益增长的企业怀疑态度。新闻中充斥着企业因使用成本远超预期而限制员工使用 AI 的报道。许多公司正在重新考虑用 AI 替代人工的计划。AI 行业要想成功，必须获得企业的全力支持，而大型企业目前普遍仍在观望。
Diseconomies of Scale. Every new technology I can think of thrived, in part, due to economies of scale, where the larger the industry grew, the more efficient it got. AI is going in the opposite direction, where every new AI model consumes more resources than its predecessors. This may turn out to be the fatal flaw — the bigger the industry gets, the more its operating costs increase.
规模不经济。我能想到的每一项新技术，其繁荣都部分得益于规模经济——行业规模越大，效率越高。AI 却反其道而行之：每一代新模型的资源消耗都超过前一代。这可能会成为致命缺陷——行业越大，运营成本反而越高。
Institutional Warnings. Moody’s recently warned that high AI infrastructure spending threatens the credit of AI companies and their large tech partners. I read recently that the number one question being fielded by investment advisors is people asking how to divest from AI.
机构警告。穆迪最近警告称，高昂的 AI 基础设施支出正在威胁 AI 公司及其大型科技合作伙伴的信用评级。我最近读到，投资顾问接到最多的客户问题就是如何从 AI 领域撤资。
I don’t have a crystal ball to foresee the nature of the crash. It could be a total crash like the 2000 tech crash, where four out of five tech startups disappeared practically overnight. I lived in the DC area at the time, and I will never forget the rows of abandoned CLEC headquarters buildings in Northern Virginia. A crash could be milder, where a few firms disappear, with the outlooks for the survivors greatly diminished, and industry expectations are reset to something more realistic.
我没有水晶球来预知这场崩溃的具体形态。它可能像 2000 年科技泡沫崩溃那样彻底——五分之四的科技初创公司几乎一夜之间消失。我当时住在华盛顿特区地区，永远不会忘记北弗吉尼亚那一排排被废弃的 CLEC（竞争性本地交换运营商）总部大楼。崩溃也可能温和一些：少数公司消失，幸存者的前景大幅缩水，行业预期被重置到更现实的水平。
The reason I wrote the blog is to speculate about what happens after an AI crash. I foresee some of the following consequences of an AI crash.
我写这篇博客的原因是想推演 AI 崩溃之后会发生什么。我预见到崩溃可能带来以下一些后果。
An article in the Economist said a total crash would wipe out $20 trillion in U.S. wealth. That means wiping out the wealth of the investors in the new technology, along with a huge hit on the stock market.
《经济学人》的一篇文章指出，全面崩溃将抹去 20 万亿美元的美国财富。这意味着新技术投资者的财富将化为乌有，股市也将遭受重创。
Data center construction would stop dead, and unfinished projects would collapse. Communities that contributed to the costs of bringing data centers will end up eating those investments.
数据中心建设将戛然而止，未完工的项目将全部烂尾。那些为引进数据中心而承担了成本的地方社区，最终将独自吞下这些投入。
There will be stranded investments by electric utilities and water companies that built new infrastructure to support data centers. They won’t eat these losses, though, which will all be passed on to ratepayers in the form of higher electric and water rates.
电力公司和供水公司为建设支持数据中心的新基础设施而产生的投资将成为沉没成本。不过它们不会自己承担这些损失——全部将以更高电价和水价的形式转嫁给终端用户。
A lot of vendors will be in big trouble. Companies that pivoted to supporting data center electronics, like Micron, might fold. But a lot of other vendors also would take a big hit. For example, Corning announced investments in three new fiber factories just to support data centers.
大量供应商将陷入严重困境。像美光这样转向支持数据中心电子设备的公司可能会倒闭。但许多其他供应商也会受到重创。例如，康宁宣布投资建设三家新光纤工厂，就只是为了支持数据中心。
There have been some huge investments by carriers in middle-mile fiber to support data centers. The companies that made these investments won’t see the expected revenues.
运营商为支持数据中心在中程光纤方面进行了巨额投资。做出这些投资的公司将无法获得预期收入。
The most interesting thing about a major crash is that it can do as much long-term good as it does short-term harm. I want to again use the analogy from the tech crash. I know of at least a half dozen CLECs that had business plans to capture 30% of the voice and data market in Atlanta. The crash cleaned them all out of the market, but without the crash they would have all failed more slowly. The tech crash brought a sense of reality to the telecom market, which still experienced phenomenal long-term growth after the original tech companies had died.
一场重大崩溃最有趣的地方在于，它带来的长期好处可能与短期伤害一样多。我再次借用科技泡沫崩溃的类比。我知道至少有六家 CLEC，它们的商业计划都是要拿下亚特兰大 30% 的语音和数据市场。泡沫崩溃把它们全部清出了市场，但如果没有那次崩溃，它们只会更缓慢地走向失败。科技泡沫崩溃给电信市场带回了一剂现实的清醒剂，而电信市场在最初那批科技公司消亡后，仍然经历了惊人的长期增长。
I don’t think there is any chance of AI failing as a technology. But that doesn’t mean the early developers are the ones who will see the ultimate success. Most, and maybe all of today’s players might be gone. A crash will bring financial constraints, which would mean that AI companies will have to figure out efficiency and economies of scale. If AI is ever going to be a viable technology, it has to control costs and be able to pay for itself. It’s hard to foresee today’s companies somehow reaching that point without some kind of market reset.
我不认为 AI 作为一项技术会失败。但这并不意味着早期的开发者就是最终的成功者。今天的大多数玩家——甚至可能是全部——可能都会消失。崩溃将带来财务约束，意味着 AI 公司将不得不解决效率和规模经济的问题。如果 AI 要成为一种可行的技术，它必须控制成本并能自负盈亏。很难想象今天的这些公司能在不经历某种市场重置的情况下达到那个临界点。
译者注：Doug Dawson 是 CCG Consulting 的总裁，长期从事电信和宽带行业的咨询工作。他从行业分析师的独特视角出发，将当前 AI 狂热与 2000 年电信泡沫进行了深刻类比。文中 CLEC 是 1996 年美国《电信法》催生的竞争性本地运营商，在 2000 年泡沫中大量涌现又迅速消亡，成为那次泡沫的标志性符号。这篇文章的核心洞见在于：崩溃不是技术的终结，而是淘汰投机者、让真正有效率的模式胜出的残酷筛选机制。