[新聞] Anthropic Files to Go Public, Setting

[zzahoward (Cheshire Cat)]

原文標題：

Anthropic Files to Go Public, Setting Stage for Huge I.P.O.


原文連結：

https://www.nytimes.com/2026/06/01/technology/anthropic-ipo.html

發布時間：

June 1, 2026

記者署名：

Mike Isaac

原文內容：

以下只翻譯重點

Anthropic, the artificial intelligence company behind the chatbot Claude,
confidentially filed on Monday for an initial public offering, joining what
could be a once-in-a-generation, moneymaking moment on Wall Street.

人工智慧公司Anthropic （旗下擁有聊天機器人 Claude）於週一秘密提交了首次上市申
請

With its I.P.O. filing, Anthropic is expected to be among three high-profile
companies preparing to go public this year, along with the rocket company
SpaceX and OpenAI, which started the A.I. boom in 2022 with its ChatGPT
chatbot.

Anthropic 預計將成為今年準備上市的三家備受矚目的公司之一，另外兩家分別是火箭公
司 SpaceX 和 OpenAI

Their I.P.O.s, which would be among the biggest ever, could create a tsunami
of investment and employee wealth, and mint the world’s first trillionaire
in Elon Musk, who owns about 50 percent of SpaceX. The public offerings could
also flood the nonprofit world with new money, since Anthropic and OpenAI
have both pledged a large part of their shares to charity. SpaceX’s I.P.O.
is expected this month, and OpenAI has been preparing to file in the coming
weeks.

它們的IPO將是史上規模最大的IPO之一，可能引發投資和員工財富的巨變，並造就伊隆馬
斯克——這位持有SpaceX約50%股份的富豪——成為世界首位萬億富翁。由於Anthropic和
OpenAI都承諾將大部分股份捐贈給慈善機構，這些IPO也可能為非營利組織帶來大量新資
金。 SpaceX的IPO預計將於本月進行，而OpenAI也正在準備在未來幾週內提交上市申請。

An I.P.O. filing also ratchets up Anthropic’s competition with OpenAI. Last
week, Anthropic officially passed OpenAI as the world’s highest-flying A.I.
start-up with $65 billion in new financing that valued it at $900 billion
before the inclusion of the new capital. OpenAI’s last valuation was $730
billion.

Anthropic提交IPO申請也加劇了其與OpenAI之間的競爭。上週，Anthropic憑藉650億美元
的新融資，估值達到9,000億美元（未計入新融資），正式超越OpenAI，成為全球估值最
高的AI新創公司。 OpenAI先前的估值為7,300億美元。

In a statement, Anthropic said the filing “gives us the option to go public
” after a review of its paperwork by the Securities and Exchange Commission.
The company did not provide details about the timing or size of an I.P.O. and
said the deal would “depend on market conditions and other factors.” With
the filing on Monday, a public offering could happen as soon as this fall.

Anthropic在聲明中表示，提交上市申請後，美國證券交易委員會將審核其文件，這意味
著公司「可以選擇上市」。該公司並未透露IPO的具體時間和規模，並表示最終交易將「
取決於市場狀況和其他因素」。週一提交的上市申請意味著最早可能在今年秋季進行公開
募資。

Among the three companies racing to go public, Anthropic is the youngest and
fastest growing, thanks in large part to its A.I. tools for automatically
writing computer code. Last week, Anthropic said its revenue run rate, which
is its expected revenue for the year based on its current performance,
crossed $47 billion in May. It is unclear if the company is profitable.

“The coding focus is what makes this filing interesting,” said Shashi
Bellamkonda, a director at Info-Tech Research Group, a technology research
and advisory firm. “Anthropic didn’t try to be everything. No browser, no
image generation, no commerce layer. That discipline is now a $47 billion run
rate.”

「這份文件最引人注目之處在於它專注於程式碼開發，」技術研究顧問公司Info-Tech
Research Group的董事Shashi Bellamkonda表示。 “Anthropico並沒有什麼都要。
它沒有瀏覽器、圖像生成或電商平台。如今，這種專注的領域已經創造了470億美元的年
收入。”

Founded in 2021, Anthropic was started in San Francisco by Dario Amodei, its
chief executive, and a handful of other researchers who worked with him at
OpenAI. Since he started the company, Mr. Amodei has repeatedly described
what he believes is the transformative nature of so-called artificial general
intelligence, or A.G.I., an as-yet-unattained artificial intelligence that
can match the abilities of the human brain.

Anthropic成立於2021年，由CEO Dario Amodei和幾位曾在OpenAI與他共事的研究人員在舊金
山創立。自公司成立以來，Amodei一再闡述了他認為所謂的通用人工智慧(AGI)的變革性本
質。 AGI是一種尚未實現的人工智慧，其能力可以媲美人腦。


Anthropic has also stressed that this technology could be dangerous to
humanity. That focus on safety has helped Anthropic maintain a positive
public image, but it has also led to tension. This year, the Pentagon banned
Anthropic’s technology from its systems because the company wanted to place
some limits on what military and intelligence officials could do with it.

More recently, Anthropic’s relationship with the federal government has been
in flux after it released a powerful model called Mythos, which can identify
and fix security vulnerabilities in software systems. Mr. Amodei has visited
the White House as intelligence agencies grapple with how Mythos works.

Other A.I. labs, like OpenAI and Google’s DeepMind, have built a wider array
of products, including A.I.-focused web browsers, commerce tools and image
generation apps. But Anthropic has focused mostly on building A.I. that
excels in coding software and tools for business customers.

Anthropic也強調，這項技術可能對人類構成危險。這種對安全的重視幫助Anthropic維持了
良好的公眾形象，但也引發了一些緊張局勢。今年，五角大廈禁止在其係統中使用其技術，
因為該公司希望限制軍方和情報官員使用該技術的可能性。

最近，Anthropic與聯邦政府的關係變得不穩定，該公司之前發布了一款名為Mythos的強大
模型，該模型能夠識別並修復軟體系統中的安全漏洞。在情報機構努力研究Mythos的
工作原理之際，Amodei曾經訪問白宮。

其他人工智慧實驗室，例如 OpenAI 和Google旗下的 DeepMind，已經開發出更廣泛的產
品，包括專注於人工智慧的網路瀏覽器、電商工具和圖像生成應用。但 Anthropic 主要
專注於建立能夠為企業客戶編寫卓越軟體和工具的人工智慧系統。

Last fall, that focus appeared to pay off. Anthropic produced a breakthrough
model with Claude Opus 4.5, which the company described as “a meaningful
step forward in what A.I. systems can do,” with an emphasis on software
coding. Its use soared, and Anthropic revenues skyrocketed as more business
customers began paying for the service.

“Claude’s latest advancements have driven large-scale adoption among the
world’s most demanding organizations,” said Brad Gerstner, founder of
Altimeter Capital, which is an Anthropic investor.

There has long been interest to own a piece of Anthropic, with investment
firms such as Menlo Ventures, Greenoaks Capital, Coatue and XN pouring money
in, as well as technology companies including Amazon, Google, Samsung, Micron
and SK Hynix.

A significant amount of the money Anthropic has raised is likely to pay for
the rising cost of computing power, known as “compute,” a scarce resource
given the increasing demand for A.I. software. Mr. Amodei has said Anthropic
is seeing “unprecedented demand” for its products, which requires it to pay
for even more compute from companies like Microsoft, Amazon and Google to
keep up.

Anthropic also faces stiff competition. OpenAI has reoriented itself to focus
on coding products with Codex, its software coding tool. Mr. Musk, who runs
SpaceX, agreed to team up with and potentially acquire Cursor, a competing
software coding-focused start-up. Larger companies, including Google and
Microsoft, also offer their own coding tools to developers.

But SpaceX has also become a partner to Anthropic. Anthropic recently said it
had reached an agreement to use all the computing power at the rocket company
’s Colossus 1 data center in Memphis. The move gives Anthropic access to
more than 220,000 A.I. chips and opens the door to working with SpaceX to
create A.I. data centers in space.



心得/評論：

AI IPO今年真的要大爆發了

Space X 包裹著X AI、歐噴醬以及Anthropic都在今年要上市

市場越來越瘋狂，資金越吸越大

前幾天跟GE聊了一下，他們能源部門產能滿載，但也不敢擴廠

上個月跟Ecolab開會也是評估明年data center整體CAPEX會冷卻不少

OpenAI提交的IPO文件表示:

2028年營收99B 虧損74B，2030營收280B 開始營利

歐噴醬真的很狂，這財務預估也只有Altman噴得出來

但建議大家還是先想清楚，投資是一輩子的事情

All in 翻身的確是存在這世界，Fomo絕對是最要不得的事情


--
※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 136.226.240.97 (臺灣)
※ 文章網址: https://www.ptt.cc/bbs/Stock/M.1780360763.A.B22.html
推 pponywong   : 900B 估值應該值得 現在寫程式速度比以前x20~100倍
推 tsubasawolfy: 恩戳闢跟歐噴醬真的互幹拐子中XD
→ tsubasawolfy: IPO互扯 token加倍也是
→ pponywong   : 大型軟體公司靠自動化測試+AI輔助 軟體疊代速度
→ pponywong   : 快很多
現在美國部分風向有點變了
四月多的時候去聽Wharton的一場演講，有些人認為AI受益較多的會是低碼公司
大型軟體應用的公司反而會被Token成本限制，同時潛在錯誤成本太高
(沒有替死鬼可以解雇=上斷頭台)
※ 編輯: zzahoward (136.226.240.97 臺灣), 06/02/2026 08:46:52
→ sunnywing   : 這三家看起來會賺錢的只有Anthropic嗎
推 pponywong   : Open醬第一開始走太多彎路 最近的codex還不錯用
推 ck326       : 三間市場流動性被吸3,4T，下場會？？？嘻嘻
推 pponywong   : 很多早期的程式碼 可以用AI幫忙改寫 像是最早從美國
→ pponywong   : 軍方實驗室出來的arpack/GNU scientific lib
→ pponywong   : 銀行用的cobol 很多都是70~80年代的東西
→ zzahoward   : 我目前經驗是太舊的東西有一堆屎山 反而不好去改
推 turndown4wat: dxyz一根?
→ daniel6412  : 重點是早期程式碼那又臭又長你確定該寫完還能跑？
推 hellophoenix: 週一秘密送件 週二全世界都知道了
推 takanasiyaya: cobol現在是AI卡關的地方，證明AI有邊界，沒有海量
→ takanasiyaya: 資料的領域AI都很難發揮
推 zaqimon     : 很多時候寫程式其實是在繞過bug 不知道AI繞的過去嗎
→ zaqimon     : 有些東西有時候正常有時候不正常 AI要怎麼debug?
→ zaqimon     : 就算AI真的繞過去 那個程式碼人類可能完全看不懂
推 BIGBIGFISH  : Zoom昨天一根
→ zaqimon     : 到時候又出問題就只能請AI自己debug
→ zaqimon     : AI你快一點好不好 我想要下班了
→ zaqimon     : 對於古早時代那些看不太懂得程式碼最好不要亂動
推 snowpoint   : AI寫程式導致的問題需要三個月到數年才會浮現，主要
→ snowpoint   : 問題是難以維護，所以短期市場會繼續噴沒問題
→ snowpoint   : 近期才開始看到幾篇碼農的網誌提到這問題，說初期
→ snowpoint   : 建置很快，但現在因需求擴張要重構，試了半天看不懂
→ snowpoint   : 最後全部打掉自己重新手刻XD
推 h0103661    : 並沒有，ai寫的比人還乾淨
→ wr          : 理論上 AI coding是可以讓系統變成"活的"
→ wr          : 當然 用AI讓系統活起來 隨需求變化程式碼還要時間
推 doubi       : 現在都用 Codex
→ dosiris     : 丸子
→ dosiris     : 資金要回美國了
推 BBKOX       : AI成熟後會加劇軟體行業的競爭
→ BBKOX       : 對軟體巨頭來說當然是喜憂參半
推 snowpoint   : 網誌有提到AI code很有條理沒錯，但有當過碼農就知
→ snowpoint   : 有些小問題的workaround會在一些不明顯的地方完成
→ snowpoint   : 當這些由AI幫你完成時，因為不識你自己寫的，所以
推 Haruna1998  : 最近去新公司 接手的案子隨便點個檔案起來都超過5
→ Haruna1998  : 000行= = 沒有Claude真的要烙跑了
→ snowpoint   : 你要花很多心力去找，而系統大到一個程度後就很難
→ snowpoint   : 其實就跟老code不要去動的道理是一模一樣，因為負責
→ snowpoint   : 的人不在了，沒人想去鑽那麼大的code base
→ wr          : 找出關鍵小地方 這種工作理論上AI比人類更適合
→ wr          : 因為人類只能靠老屁股的秘笈 AI只要token夠多
推 qk3380888   : 股份捐贈慈善機構 真假啊
推 pponywong   : 5000行還好吧 沒有AI前 我大概都寫到14~20萬行
→ pponywong   : vibe coding的問題還是在個人 對軟體架構跟專業知識
推 Haruna1998  : 一個檔案放這麼多行幹嘛？拆檔案不好嗎@@
→ pponywong   : 都缺乏的人 AI要怎麼產生跟品質 也很難顧好
→ zzahoward   : 手刻的和AI寫的開發人員掌握度不同啊
→ pponywong   : 我是說整個軟體專案 自己寫的部分
推 Xration     : 988大爆崩 還在吹988的崩潰了沒
→ Xration     : 發錯==
推 treeson     : Codex負責人TiBO每天上X打嘴炮嗆claude code還可以
→ treeson     : 一直上線新功能你就知道ai的能力
推 tsubasawolfy: 別這樣 Tibo還有按下使用量Reset的用途
→ zxcvvcxz    : 大概什麼時候上市啊
推 fhill12     : 越虧越噴
推 jackshadow  : 貸款也要買
推 crohns      : cobol,as400這種有辦法更新重構成現代語言？？？
推 happyendless: dxyz噴起來~
推 mynumber55  : codex的負責人真的滿嘴的，不過小問題更新也很快
→ mynumber55  : 5.6好像這禮拜要發布了
推 sdbb        : 謝謝
推 treeson     : 有大型銀行成功的案例（JPM) ，主要是求穩所以還在
→ treeson     : 按兵不動，未來重構還是趨勢
推 rebel       : data center的資本支出放緩的論點是？太貴了？
推 NoMomoNoLife: COBOL LLM出來前就可以轉成Java啦
→ NoMomoNoLife: 有專門的工具在轉
推 Lhmstu      : 模型巨頭應該就三家了
→ mit2502     : 資本支出你不增加等於是認輸
推 rebel       : 問題是敢不敢轉 XD 我記得美國早期的核彈發射系統
→ rebel       : 就是cobol寫的 沒人敢動 萬一不小心改出bug把核彈
→ rebel       : 射出去就好笑了
→ okbon       : q3大圈錢
推 sunpedro0202: tibo真的整天嘴笑死
推 jy760517    : AI的累積錯誤風險比一般人認知到的嚴重多了...
推 cloud7515   : 現在是誰先縮誰就輸了
推 okbon       : ipo就是撈錢大會
推 ma721       : 唯一真實大獲利
→ BBKOX       : AI查找錯誤的能力也比一般人強的多
推 LipaCat5566 : 利多出盡
→ bnn         : 是不是得開新銀行用新寫的code來測試 沒事才反併購
推 pponywong   : 現在可以用自動化架設kubernetes+docker CI/CD
→ pponywong   : 用AI開發加除錯 也容易很多
→ cct1121     : 都在籌錢
推 siyaoran    : 部分用AI才會這樣，擴充功能, 找錯誤這明明是AI最擅
→ siyaoran    : 長的，我們專案靠AI才解決長年的已知bug
推 siyaoran    : 實務上，接收沒人維護的程式嗎也是有AI更快，問架構
→ siyaoran    : 和問實作更容易了解別人的程式碼
推 AirLee      : 賺爛了
推 capssan     : 這隻我一定要allin
推 CruiseTom   : Ai王