Anthropic 申请上市或引发科技板块重估：AI 龙头股若表现不佳恐引连锁反应
英文标题：🔶 Anthropic IPO might kick off tech reset if the AI stock underperforms
AI 领军企业 Anthropic 正式递交 IPO 申请，有望成为史上规模最大的首次公开募股之一，潜在估值直奔万亿美元区间。尽管今年上半年 Anthropic 亏损超 800 亿美元，但其第二季度营收同比暴增 14 倍至 115.6 亿美元并实现单季运营盈利，展现出强劲的商业化势头。然而，当前资本市场对 AI 基础设施的巨额投入及回报周期日趋焦虑，加之开源权重模型的快速崛起，若 Anthropic 上市表现不及预期，恐将戳破市场泡沫，迫使整个 AI 投资生态与估值体系进行全面重置。
1.9 KB
1.2 KB
1.3 KB
530 B
29.1 KB
49.7 KB
1.8 KB
1.1 KB
78.6 KB
中文
2025年1月23日，Anthropic 联合创始人兼首席执行官 Dario Amodei 出席在瑞士达沃斯举行的世界经济论坛专题讨论会。
Stefan Wermuth | Bloomberg | Getty Images
科技行业需要这场上市取得成功，华尔街同样如此。
Anthropic 周一提交的 IPO 申请文件，让这家 Claude 的缔造者有望迎来历史上规模最大的一次首次公开募股。投资者已经在以数万亿美元的体量来衡量这家成立仅 5 年的公司的潜在市值，将其推向了与全球最具价值公司并驾齐驱的行列。
对于整个科技行业而言，这是一个极其关键的时刻，因为这轮人工智能热潮在很大程度上是由两家公司推动的：Anthropic 和 OpenAI。后者已经秘密提交了上市申请，但表示其挂牌亮相可能要等到 2027 年。
与此同时，股票市场对 AI 相关交易开始流露出焦虑迹象。新型云厂商（Neoclouds）和数据中心公司遭受重创，导致澳大利亚的 Firmus 撤回了其 IPO 计划，SK海力士旗下的美国 Solidigm 业务也同样撤回了申请。
如今，所有的目光都投向了 Anthropic。该公司押注投资者会忽略其上半年超过 800 亿美元的亏损，转而将 14 倍的营收增长视为对其顶尖模型与服务近乎无穷无尽需求的反照。
“如果它的表现不及预期，局面将变得有些像多米诺骨牌般的‘纸牌屋’，”Sapphire Ventures 联合创始人 Jai Das 表示，他的科技投资生涯可以追溯到 1990 年代。
Anthropic 的业务与科技巨头深度交织。其近一半的营收来自通过 微软、亚马逊 和 Alphabet 的云业务部门销售 Claude，而这三家巨头均持有该公司的股份。在本次 IPO 中，公司可向 亚马逊、博通 和 谷歌 发行约 770 亿美元的票据，同时 微软 和 英伟达 同意在 Anthropic 构建计算能力之际投资 150 亿美元。来自英伟达、超威半导体 (AMD)、亚马逊和谷歌的芯片正为 Anthropic 的研发提供算力支持。
云服务提供商正在大幅增加资本开支以满足需求，内存制造商也在克服短缺优先保障 AI 芯片。此外还有债务因素。拥有投资级信用评级的 Alphabet 承诺为 Anthropic 在得克萨斯州的数据中心提供债务兜底担保。协助谷歌为 AI 实验室开发芯片的 博通 正在建立融资池，以帮助 Anthropic 和 OpenAI 部署高性能模型，博通首席执行官陈福阳（Hock Tan）今年早些时候向分析师透露了这一消息。
对于二级市场投资者而言，标普 500 指数中超过三分之一的市值来自科技行业的万亿美元级巨头，而这些公司都在不同程度上深度押注 AI。除此之外，还有像 Adobe 和 Salesforce 这样已经集成了 Anthropic 旗下 Claude 助手的软件巨头。
市场面临重置？
而在私募一级市场，投资单笔金额不断飙升，但 IPO 却屈指可数。Bessemer Venture Partners 和 Menlo Ventures 均向 CNBC 表示，Anthropic 是他们迄今为止最大的一笔投资。
“如果他们表现不佳，他们将成为 AI 行业的风向标，”成长型股权投资公司 FTV Capital 的执行合伙人 Brad Bernstein 表示，“他们的估值将体现出这种业绩表现的分歧，随后将对大量涌入 AI 风险投资领域的过剩资本产生下渗连锁反应。所有东西都将不得不进行某种程度的重新洗牌。”
继沙特阿美（Saudi Aramco）和 埃隆·马斯克 旗下于 6 月挂牌的 SpaceX 之后，Anthropic 将成为第三家估值超过 1 万亿美元上市的公司。OpenAI 可能会成为第四家。
随着企业广泛采用 Anthropic 的代码编写和生产力工具，该公司的营收正以惊人的速度成倍增长。第二季度营收从去年同期的 7.87 亿美元飙升至 115.6 亿美元。
截至 6 月底，该公司持有近 1320 亿美元的现金、现金等价物和短期投资。在经历了连续五个季度的运营亏损后，Anthropic 报告第二季度实现运营利润 9.74 亿美元。
“我认为，这比人们担心的要好，”Burnham Park Capital Markets 合伙人兼首席 AI 官 Rohit Kulkarni 在谈及财务数据时表示，“原本可能会有更多负面细节流出。”
中国济南 - 7月17日：在这张创意图片中，2026年7月17日在中国山东省济南市，智能手机屏幕上显示着 Kimi K3 的标识。（图片来源：VCG/VCG via Getty Images）
Vcg | Visual China Group | Getty Images
投资圈某些角落的一个担忧来源是所谓开源权重（open-weight）模型的兴起，此类模型对企业而言使用成本更低，并且在用户接入模型的部署位置方面提供了定制化和灵活性。
在招股说明书中，Anthropic 点名列出的仅有两家竞争对手是 OpenAI 和谷歌，不过该公司表示其面临来自“更广泛的模型提供商群体”的竞争，包括开源权重或开源模型。
Anthropic 表示，一切仍处于早期阶段，《财富》世界 500 强企业在最近一个季度分配给其模型和服务的支出占其总支出的比例还不到 0.1%。Founder ETFs 的投资组合经理 Michael Monaghan 认为，这正是投资逻辑的一部分。
“我认为像 Anthropic、OpenAI 和 SpaceX 这样的前沿模型会有一席之地，开源权重模型也同样会有空间，”Monaghan 说，“现在断言一种模式会淘汰另一种模式还为时过早。”
面对超大规模云厂商在 AI 基础设施上的投入将达到数万亿美元，Sapphire 的 Das 指出，出现对市场供过于求的担忧是可以理解的。但他将其比作 1990 年代末的电信与互联网泡沫，当时各家公司将光纤设备大量推向市场。最终，这些基础设施得到了充分利用，尽管在此期间大量公司破产倒闭。
“长远来看，我不认为会缺乏需求，”Das 说。
下行风险
一次表现平平的亮相可能会让 OpenAI 的上市时间进一步推迟，并促使投资者重新审视其高昂的估值。与此同时，利率上升意味着企业筹集债务资金的成本正变得越来越高。
金融研究机构 New Constructs 的首席执行官 David Trainer 在上个月的一份报告中写道，AI 热潮建立在廉价信贷和资本市场准入将“唾手可得”的预期之上。
“如果 AI 的流动性枯竭——而且现在有明显的迹象表明确实在枯竭——我们认为标普 500 指数存在 20% 的巨大下行风险，其中大部分跌幅将集中在 AI 相关股票的估值上，”Trainer 写道。
行业专家则采取了不同的视角，将重点放在该技术的强大能力及其迅猛普及上。在行业基准 Artificial Analysis 智能指数上，Claude Opus 5.5 Max 的表现超越了其他所有模型。
Snowflake 首席执行官 Sridhar Ramaswamy 表示，这项技术只会变得愈发不可或缺。
“Anthropic 和 OpenAI 迄今为止已经证明了他们站在 AI 前沿并持续推动创新的能力，”Ramaswamy 说，“它们是非常不可思议的公司，通过不断拓宽可能性的边界为人类作出了巨大贡献，我们非常支持它们成为上市公司并取得巨大的成功。”
即便如此，Snowflake 仍在测试来自中国月之暗面（Moonshot AI）、深度求索（DeepSeek）和智谱 AI 的低成本开源权重模型。
“我们在内部 AI 方面投入巨大，因为我们希望尽可能获取更高的效率，”Ramaswamy 表示。
视频/嵌入内容：FirstMark 的 Rick Heitzmann 表示，Anthropic 似乎将抢先 OpenAI 一步完成 IPO
English / 原文
Dario Amodei, Anthropic's co-founder and CEO, appears on a panel at the World Economic Forum in Davos, Switzerland, on Jan. 23, 2025.
Stefan Wermuth | Bloomberg | Getty Images
The tech industry needs this to go well. So does Wall Street.
Anthropic's IPO filing on Monday puts the Claude creator in position to pursue what could be the largest initial stock offering in history. Investors are already measuring the 5-year-old company's potential market cap in the trillions, putting it in a category with the world's most valuable companies.
It's a critical moment for the whole tech sector, as the artificial intelligence boom has been largely driven by two companies: Anthropic and OpenAI. The latter has confidentially filed to go public, but has said a debut likely won't take place until 2027.
Meanwhile, the stock market is showing signs of jitters on the AI trade. Neoclouds and data center companies have been getting pummeled, prompting Australia's Firmus to withdraw its IPO plans and for SK Hynix's U.S.-based Solidigm unit to do the same.
Now, all eyes turn to Anthropic, which is betting that investors will look past over $80 billion in losses in the first half of the year, and see the 14-fold increase in revenue as a reflection of seemingly insatiable demand for the company's leading models and services.
"If it underperforms expectations, it'll be a little bit of a house of cards," said Jai Das, co-founder of Sapphire Ventures whose career in tech investing dates to the 1990s.
Anthropic's business is deeply entwined with tech megacaps. Close to half its revenue comes from Claude being sold through cloud divisions at Microsoft, Amazon and Alphabet, which all hold investments in the company. In the IPO, some $77 billion in notes can be issued to Amazon, Broadcom and Google, while Microsoft and Nvidia agreed to invest $15 billion as the company builds up its computing capacity. Chips from Nvidia, Advanced Micro Devices, Amazon and Google are powering Anthropic's development.
Cloud providers are ratcheting up capital expenditures to meet demand, and memory manufacturers are fighting shortages to prioritize AI chips. And there's the debt factor. Alphabet, with its investment-grade credit rating, committed to backstopping Anthropic for a data center in Texas. Broadcom, which helps Google develop chips for AI labs, is creating financing pools to help Anthropic and OpenAI deploy high-performance models, CEO Hock Tan told analysts earlier this year.
For public market investors, more than one-third of the value of the S&P 500 comes from tech's trillion-dollar companies, which are all, to varying degrees, going deep in AI. Beyond that, there are software giants like Adobe and Salesforce that have integrated with Anthropic's Claude assistant.
Market reset?
And in the private markets, check sizes have ballooned but IPOs are few and far between. Bessemer Venture Partners and Menlo Ventures both told CNBC that Anthropic is their biggest investment ever.
"If they don't perform well, they will become the bellwether for AI," said Brad Bernstein, a managing partner at growth equity firm FTV Capital. "Their valuation will reflect that performance variation, and then it'll just be a trickle-down effect to all the stuff going on with the excess capital that's getting poured into the venture AI world. Everything is going to have to sort of be reset."
Anthropic will be the third company to go public with a valuation above $1 trillion, after Saudi Aramco and Elon Musk's SpaceX, which debuted in June. OpenAI could be the fourth.
Revenue at Anthropic has been multiplying at a staggering pace as businesses adopt the company's coding and productivity tools. Second-quarter revenue soared to $11.56 billion from $787 million a year earlier.
The company held almost $132 billion in cash, equivalents and short-term investments at the end of June. And after five quarters of operating losses, Anthropic reported $974 million in operating income for the second quarter.
"It is, I think, better than feared," said Rohit Kulkarni, partner and chief AI officer at Burnham Park Capital Markets, referring to the financials. "There could have been more negative tidbits."
JINAN, CHINA - JULY 17: In this photo illustration, the logo of Kimi K3 is displayed on a smartphone screen on July 17, 2026 in Jinan, Shandong Province of China. (Photo by VCG/VCG via Getty Images)
Vcg | Visual China Group | Getty Images
One source of concern in some corners of the investing universe is the emergence of so-called open-weight models, which are cheaper for businesses to access and allow for customization and flexibility in terms of where users access the models.
In the prospectus, the only two competitors Anthropic named were OpenAI and Google, though the company said it faces competition from a "broader set of model providers," including open-weight or open-source models.
Anthropic says it's all very early, with Fortune 500 companies committing less than 0.1% of their total expenses to its models and services in the latest quarter. Michael Monaghan, portfolio manager at Founder ETFs, says that's part of the investment thesis.
"I think there's going to be room for frontier models like Anthropic and OpenAI and SpaceX, and I think there will be room for open-weight models," Monaghan said. "It's way too soon to say that one is going to disintermediate the other."
With hyperscaler spending on AI infrastructure set to reach into the trillions of dollars, Sapphire's Das said there's understandable concern about a market glut. But he compares it to the telecommunications and dot-com bubble of the late 1990s, when companies flooded the market with fiber-optic equipment. Eventually, the infrastructure got used, although plenty of companies went belly up in the interim.
"Long term, I don't think there is a lack of demand," Das said.
Downside risks
A lackluster debut could push out OpenAI's offering even further and cause investors to rethink its lofty valuation. At the same time, rising interest rates mean it's getting costlier for companies to raise debt.
David Trainer, CEO of financial research firm New Constructs, wrote in a report last month that the AI boom has been built on the view that cheap credit and access to the capital markets will be "readily available."
"If AI liquidity dries up, and there are clear signs it is, we see large, 20%, downside risk in the S&P 500, with most of that concentrated in the valuation of AI-related stocks," Trainer wrote.
Industry experts take a different approach, focusing on the power and rapid proliferation of the technology. Claude Opus 5.5 Max performs better than any other model on the Artificial Analysis Intelligence Index, an industry benchmark.
Snowflake CEO Sridhar Ramaswamy said the technology is only becoming more essential.
"Anthropic and OpenAI have by now demonstrated their ability to stay at the frontier of AI and keep pushing innovation," Ramaswamy said. "They're incredible companies that are doing a big service to humanity by expanding the frontier of what is possible, and we are very much rooting for them to be public and very successful companies."
Still, Snowflake is experimenting with low-cost open-weight models from China's Moonshot AI, DeepSeek and Zhipu.
"We spend a lot on internal AI because we want to get as much efficiency as we can," Ramaswamy said.
WATCH: Seems like Anthropic will beat OpenAI to IPO, says FirstMark's Rick Heitzmann