OpenAI, Anthropic and SpaceX are together worth roughly $5.2 trillion — more than every US tech company that went public from 1980 to 2025 combined ($4.1 trillion across 3,365 deals). The gap shows how much value now gets created in private markets, out of public reach.
If you're wondering how huge the AI segment's march toward the public markets is,look no further thanthe valuations of the biggest companies in the space.
如果你想知道 AI 板块迈向公开市场的步子到底有多大，不用往别处看，就看这个领域里最大几家公司的估值。
PARA 02
Take OpenAI,which is raising cash at a $1.2 trillion valuation, according to The Wall Street Journal,throw inAnthropic's potential $2 trillion value when it goes public, and SpaceX's current $2 trillion market cap,and you're sitting at just north of $5 trillion,The Financial Times reported.
Thatis more thanthe value of all of the initial public offerings from 1980 through 2025,The FT says, citing data from University of Florida Warrington College of Business emeritus professor Jay Ritter.
Morning Joe economic analyst and New York Times Op-Ed contributing writer Steve Rattnerillustratedthe valuationsin a chart he posted on X.Healso notesthat the historical numbers aren't adjusted for inflation.
《早安乔》经济分析师、《纽约时报》评论版特约撰稿人史蒂夫·拉特纳（Steve Rattner）用了一张图表把这几个估值摆了出来（图发在 X 上）。他还指出，历史数据并没有做通胀调整。
PARA 05
Anthropic and OpenAIaretwo of the largest and most advanced AI labs on the planet,thanks to their Claude and GPT model families.
Anthropic 和 OpenAI，是全球规模最大、技术最先进的两家 AI 实验室，这要归功于它们的 Claude 和 GPT 模型家族。
PARA 06
Dario Amodei-led Anthropicis expected to go publicsometime next month,though that could also slip to later in the fall.OpenAI,which was also expected to go public this year,is now targetingan IPO in 2027.
The companies' massive valuationsindicatejust how much early backers and Wall Street are banking on AI living up to the hype.Various tech executiveshave comparedthe technology's importanceto that of fire or the internet itself.
The duo and SpaceXhave also turnedthe collection of hardware companiesthat produce the GPUs and memory chips that power their AI modelsinto an indispensable part of the AI trade.
这两家加上 SpaceX，还让那一批硬件公司 ——也就是生产 GPU 和存储芯片、给它们的 AI 模型供电的公司 ——变成了 AI 交易里不可或缺的一环。
PARA 09
Take a look atNvidia,which was valued at $5.5 trillion as of Tuesday,or Micron,whose stock price has soared more than 1,100% over the last two years.
ButAI companiesare also facinga series of challenges,including fears that the technology could harm humans.Thatis in addition toexisting concerns that cyber criminals and authoritarian regimes could abuse AI models for their own purposes.
但AI 公司也正面对一连串挑战，其中包括人们担心这项技术会伤害人类。除此之外，还有早就存在的顾虑：网络罪犯和威权政权可能滥用 AI 模型来达到自己的目的。
PARA 11
Last week,Amodeicalled forAI labs to slow the pace of development of high-powered AI models,sending shares of AI adjacent companies lower.Buta view of the market so far this weekshowsthat investors aren't ready to abandon the AI boom just yet.
上周，阿莫代伊呼吁各家 AI 实验室放缓高性能模型的研发节奏，这令 AI 相关公司的股价走低。但本周到现在的市场表现说明，投资人还不打算现在就放弃这轮 AI 热潮。
配图与图表 · Visuals
三张实景配图 + 两张按原文数据自绘的图表。图表数据全部取自文中与所引报道。
火箭发射。SpaceX 是这三家里唯一已经真正上市的 —— 6 月挂牌，首日收盘涨 19%，估值突破 2 万亿美元。
A rocket launch. SpaceX is the only one of the three that has actually gone public: it listed in June, closed its first day up 19%, and debuted above a $2 trillion valuation.本篇的核心论据：三条腿的凳子，两条腿还没落地。
The article's central comparison — but two of the three legs have not actually landed yet.AI 数据中心的机柜与线缆。文章第 8、9 段说的就是这条线：AI 模型靠 GPU 和存储芯片跑，把这些硬件公司变成了「AI 交易」里绕不开的一环。
Cabling inside a data centre. Paragraphs 8 and 9 make the point: AI models run on GPUs and memory chips, which turned the hardware makers into an indispensable part of the AI trade.支撑 2 万亿估值的那条曲线：一年不到涨了十倍以上。
The curve behind the $2 trillion ask — more than a tenfold jump in under a year.夜色中的地球。文章最后一段提到的更大背景：十年前还不存在（或完全是另一个样子）的三家公司，如今纸面价值压过美国 45 年的科技上市史。
The broader backdrop in the closing paragraph: three companies that barely existed a decade ago now outweigh 45 years of US tech listings on paper.
核心术语 · Glossary
默认只显示英文解释（练阅读）。想看中文，点卡片标题。选词标准：假设读者只有最基础的财经常识。
valuation/ˌvæljuˈeɪʃn/点击看中文解释 ▾
The price a company is thought to be worth — even if it has never been bought or sold. A private company gets its "valuation" when investors put money in and agree on a number. A public company instead has a market cap, which the stock market recalculates every second.
Why it matters here ▸ Every number in this article — $1.2T, $2T, $5.2T — is a valuation. Only one of them was ever paid in cash.
Extended note: A valuation is a label, not cash. It does not change the company's bank account, factories or headcount — only what a buyer is willing to pay. That is why a valuation can double (or halve) overnight while the business itself stays exactly the same.
IPO (initial public offering)/ˌaɪ piː ˈəʊ/点击看中文解释 ▾
The first time a company sells its shares to the general public, after which its shares trade on a stock exchange. In everyday English people also say "going public" or "listing".
Why it matters here ▸ SpaceX did its IPO in June. Anthropic is expected to do one soon. OpenAI says it won't until at least 2027 — which is exactly why its number needs an asterisk.
IPO = 公司第一次把股票卖给公众，之后股票就在交易所挂牌买卖。中文常说「上市」「挂牌」。口语里也叫 going public / listing。
拓展知识：一次典型的 IPO 大致四步：① 向监管机构提交招股文件（美国是 S-1）→ ② 路演（roadshow），向大型机构投资人推销 → ③ 由承销商（underwriters，如摩根士丹利、高盛）定价 → ④ 挂牌首日交易。上市之后，公司必须按季度公开财报 —— 这是「上市」跟「私募融资」最大的区别：从「跟几个有钱人交代」变成「跟全世界交代」。
Extended note: A typical IPO has four steps: file the registration statement (an S-1 in the US) → roadshow, pitching large institutions → pricing, led by underwriters such as Morgan Stanley and Goldman Sachs → first day of trading. After listing, the company must report earnings every quarter. That is the real difference from private fundraising: you stop answering to a few rich backers and start answering to everyone.
private market/ˈpraɪvət ˈmɑːkɪt/点击看中文解释 ▾
Where shares are bought and sold before a company lists. Only certain investors can join — venture capital, private equity, sovereign funds, big family offices. Prices are agreed in private deals, not quoted on an exchange, so outsiders usually cannot see them.
Why it matters here ▸ Fast-growing AI labs are reaching enormous value while still private. By the time ordinary investors get in, the biggest gains may already be booked.
Extended note: The FT estimates US private unicorns are together worth about $5.3 trillion based on their latest funding rounds. That is a structural change: public investors once bought Alphabet, Tesla or Meta while they were still small and grew with them. Today they may only get access after enormous private-market value has already accumulated. The opportunity is not gone — the starting line has moved.
Share price × number of shares. It is the stock market's current price tag for the whole company. Because the share price moves all day, so does the market cap.
Why it matters here ▸ SpaceX's $2 trillion is a market cap. Nvidia's $5.5 trillion is a market cap. Anthropic's $2 trillion is not — it is a target valuation for an IPO that has not happened.
Extended note: Rank the numbers and the distinction becomes obvious: Nvidia $5.5T (listed, actual trades) > SpaceX $2T (listed, actual trades) > Anthropic $2T (a target) > OpenAI $1.2T (under discussion). When reading financial news, first ask which number is a real transaction and which is an asking price. Half the confusion disappears.
annualised revenue run rate/ˈænjuəlaɪzd ˈrevənjuː rʌn reɪt/点击看中文解释 ▾
Take the most recent month (or quarter) of revenue and scale it up to a full year: "if this pace held for twelve months, we would book X". It is a speedometer, not a report card — it describes how fast revenue is growing right now, not what the company actually earned.
Why it matters here ▸ It is the single biggest number behind Anthropic's $2 trillion ask: roughly $9B at the end of 2025, past $100B by September 2026. Banks love it because it turns three months of data into a headline. Be careful: a run rate flatters a business that is growing fast but is not yet profitable.
拓展知识：为什么投行爱用它：三个月的数据就能包装成一个醒目的年化数字。为什么你要警惕它：一家营收翻十倍但毛利率很薄的公司，年化数字可以极其漂亮，同时每个月仍在烧钱。所以专业投资人看完 run rate，一定会接着问一个问题 —— 毛利率（gross margin）是多少。
Extended note: Why banks love it: three months of data becomes one bold headline. Why you should be wary: a company whose revenue grows tenfold but whose margins are thin can post a spectacular run rate while still burning cash every month. So the professional's next question is always the same — what is the gross margin?
gross margin/ɡrəʊs ˈmɑːdʒɪn/点击看中文解释 ▾
(Revenue minus the direct cost of delivering the product) ÷ revenue. For an AI lab, the direct cost is mostly computing power: chips, data centres, electricity. A company can grow explosively and still lose money if the gross margin is thin.
Why it matters here ▸ Anthropic has never disclosed it. It is the one line in the prospectus every banker and analyst will read first.
毛利率 =（营收 − 交付产品的直接成本）÷ 营收。对 AI 实验室来说，直接成本主要是算力：芯片、数据中心、电费。毛利率太薄，营收涨得再猛也可能亏钱。
Extended note: Separating revenue from gross profit is the key to reading AI business models. Anthropic's revenue is put above $100bn on an annualised basis — while it pays out enormous compute bills: reportedly about $1.25bn a month for GPU capacity from SpaceX, plus large leases from Amazon, Google and Meta. Revenue is above the waterline; the compute bill is below it. That is why an undisclosed gross margin matters more to bankers than the $2 trillion headline.
A pre-approved credit line from banks that a company can draw on, repay, and draw on again — like a corporate credit card. It is debt, not equity: no shares change hands and no owner gives up control.
Why it matters here ▸ Anthropic locked in about $15bn before listing, up from roughly $2.5bn a year earlier. Having cash in hand means it never has to list in a hurry just to raise money.
Extended note: Cash in hand is leverage. A company that has locked in a large credit line does not have to accept a low IPO price just to raise money — and the market reads the facility as a signal of exactly that. Read the credit line and the cash balance together and you can usually tell whether a company is listing because it wants to, or because it has to.
anchor investor/ˈæŋkə ɪnˈvestə/点击看中文解释 ▾
A large, respected investor who commits to buy a big slice of an IPO before it is priced. Their participation is a public signal: someone who did the homework is willing to write a big cheque.
Why it matters here ▸ Nvidia is reportedly in talks to commit up to $10bn to Anthropic as an anchor investor — both a financial commitment and a strategic signal, since Anthropic runs heavily on Nvidia GPUs.
Extended note: An anchor investor does two jobs: it supplies certainty (the underwriter knows a chunk is already sold) and it supplies a reference (the name itself is advertising). When the anchor is also a supplier to the company, read carefully — it can be conviction, or it can be lock-in. Nvidia investing in Anthropic is exactly that dual relationship: the more cash Anthropic has, the more GPUs it can buy.
bubble/ˈbʌbl/点击看中文解释 ▾
When asset prices rise far above what the underlying business could ever justify, usually on optimism rather than earnings. Bubbles are easy to name in hindsight and almost impossible to call while you are inside one — and being early is indistinguishable from being wrong.
Why it matters here ▸ Whether this is durable transformation or "a very expensive bet on the future" is the open question the article ends on. Anthropic's listing is the next real test.
拓展知识：判断泡沫的经典难题是：高估值也可能对应真实的技术变革（互联网、铁路、电力都曾如此）。摩根大通 CEO 杰米·戴蒙被问到 AI 投资回报时的比喻意思是：现在预测赢家还太早 —— 互联网泡沫里，很多当年的明星公司没了，一批当时没人听说过的公司成了巨头。所以对这类问题，正确的姿势不是猜「是不是泡沫」，而是盯住三个可验证的量：收入增速、毛利率、以及资本开支是否换来对应的利润。
Extended note: The classic problem with calling a bubble is that high valuations can also go with genuine technological change — the internet, railways and electricity all looked absurd at the time. Jamie Dimon's point about AI returns is essentially this: it is far too early to name the winners. During the dot-com bubble many celebrated names vanished while companies nobody had heard of became giants. So the useful posture is not to guess "bubble or not", but to track three checkable quantities: revenue growth, gross margin, and whether capital spending converts into profit.
读后测验 · Check Your Understanding
3 道题。点选项立即判分；解析需要单独点开。
Q1What caveat does Steve Rattner attach to the historical comparison?
正确答案：A
原文第 4 段："He also notes that the historical numbers aren't adjusted for inflation."
拉特纳自己就点出了这个口径问题：1980 年的 1 美元和 2026 年的 1 美元不是一回事。如果按购买力折算，历史那 4.1 万亿会比字面上更大，也就是说三家公司与历史总量之间的差距实际上没有表面看起来那么夸张。 B and C are not in the text (the FT data covers 3,365 companies from 1980 to 2025, which includes Apple and Microsoft). D reverses the point.
Q2Why does the article treat OpenAI's $1.2 trillion figure more cautiously than the other two?
正确答案：B
原文第 2 段说 OpenAI 是「raising cash at a $1.2 trillion valuation」（正在以该估值募资），第 6 段又说明它把 IPO 目标推到了 2027 年。也就是说：这是正在谈的私募融资估值，不是任何已发生交易的成交价。
顺带记住这个阅读习惯：看到大数字，先分清它是「已成交价格」「在谈价格」还是「目标价格」。 A is wrong on the face of it; C is the opposite of the text (OpenAI targets 2027); D confuses the two companies.
Q3What does the final paragraph suggest about investor sentiment this week?
正确答案：C
原文末句："But a view of the market so far this week shows that investors aren't ready to abandon the AI boom just yet."
注意 "just yet" 这个词 —— 它既是「暂时还没」，也隐含「以后不好说」。作者的态度是整个句子的关键：不否认担忧存在，但指出当下的资金行为还没有转向。 A is contradicted by the same sentence; B contradicts paragraph 6, where OpenAI pushes its IPO to 2027; D is never stated — the article only says the two stocks have risen a lot, not that they are cheap.
