字号·· | 护眼
彭博社

Anthropic的IPO可能让其客户付出代价Anthropic’s Customers May Pay the Price for Its IPO

点「原文对照」整页切到原文,或双击某段只看那段的原文。

还能用什么方式抓住一个30万亿美元的市场?How else do you capture a $30 trillion market?

Anthropic PBC 本周即将创造历史,成为首家在公开市场交易的前沿人工智能公司,潜在投资者无疑都在想:首席执行官 Dario Amodei 打算如何兑现他所说的、等待着公司的 30 万亿美元潜在营收。

Anthropic PBC prepares to make history this week as the first frontier artificial intelligence company to trade on the public markets, prospective investors are no doubt wondering how Chief Executive Officer Dario Amodei plans to deliver the $30 trillion in potential revenue that he’s said awaits.

其中一个答案是更深入地切入医疗、法律、金融和网络安全等行业。公司已承认这一点,并警告称,随着业务扩张,它“可能会越来越多地与我们自己的客户竞争”。这听起来或许离谱 —— 卖 AI 不是已经是个足够大的市场了吗? —— 但 Anthropic 的商业模式存在一个问题:其核心企业客户可以轻松在 Anthropic 的 Claude、OpenAI 的 ChatGPT 和其他 AI 模型之间切换,并不像使用 Microsoft Corp.、Google 或 Amazon.com Inc. 的云服务那样,必然被“锁定”。

One answer is to go deeper into industries like healthcare, law, finance and cyber security. The company has admitted as much, warning that as it expands, it “may increasingly compete with some of our own customers.”That may sound outlandish — isn’t selling AI a big enough market? — but Anthropic’s business model has a problem: Its core enterprise customers can easily switch between Anthropic’s Claude, OpenAI’s ChatGPT and other AI models, and aren’t necessarily “locked in” like they are with cloud services from Microsoft Corp., Google or Amazon.com Inc.

因此,Anthropic 通用 AI 销售业务面临商品化的风险,公司必须寻找其他长期盈利策略,例如销售针对特定需求的 AI 产品。

Anthropic’s business of selling general-purpose AI is thus in danger of becoming commoditized, and the company must find other long-term strategies for making money, such as selling AI products that address a specific need.

最明显的首次尝试是 Anthropic 在 2025 年成功开始与 AI 编程公司竞争。“最初他们通过我们[和另一家 AI 编程公司 Cursor]销售 Claude Code,”Replit 首席执行官 Amjad Masad 说。“然后突然之间,他们直接面向客户了。”到 2025 年 2 月,Anthropic 的编程工具预计年营收达 25 亿美元。

The obvious first effort was when Anthropic successfully started competing with AI coding firms in 2025. “Initially they were selling Claude Code through us [and another AI coding company] Cursor,” says Replit CEO Amjad Masad. “Then suddenly they’re going directly to the customer.”By February 2025, Anthropic’s coding tool had estimated annual revenue of $2.5 billion.

除了向程序员出售工具外,该公司于2026年1月进军律师市场。它推出了一系列功能,允许客户使用Claude Cowork审查合同并处理其他法律工作流程,从而与Harvey AI Corporation和Legora Inc.等法律AI公司展开竞争,这一举动搅动了企业软件公司市场,被称为“SaaS末日”。随后在4月,Anthropic推出了Claude Design,与网站设计软件制造商Figma Inc.竞争,后者的工具由Claude提供支持。

In addition to selling tools to coders, the company in January 2026 made a play for lawyers. It launched a series of features that allowed customers to use Claude Cowork to review contracts and handle other legal workflows, putting it into competition with legal AI firms like Harvey AI Corporation and Legora Inc., a move that roiled the market for enterprise software companies in a moment dubbed the “SaaSpocalypse.”Then in April, Anthropic launched Claude Design, a rival to website-design software maker Figma Inc., whose tools are powered by Claude.

发布前三天,Anthropic首席产品官退出了Figma董事会。Figma首席执行官随后表示,他对此毫无准备,并补充说Anthropic在其计划方面并未对他“始终坦诚”。

Three days before the release, Anthropic’s chief product officer stepped down from Figma’s board. Figma’s CEO went on to say he’d been blindsided, adding that Anthropic hadn’t been “consistently candid” with him about its plans.

科技巨头颠覆其生态圈内的公司并非首次。据报道,Amazon利用其平台上第三方卖家的数据帮助开发Amazon Basics产品。Apple Inc.推出操作系统新功能(从密码管理器到睡眠追踪器)已让一大批第三方应用开发者倒闭。也别忘了,Claude和ChatGPT部分是通过抓取互联网上海量素材构建而成。这种“见猴做猴”的过程是硅谷运作的重要组成部分。

It wouldn’t be the first time a tech giant disrupted the companies in its orbit. Amazon reportedly used data from third-party sellers on its platform to help develop its Amazon Basics products. There’s a passel of third-party app developers that Apple Inc. has put out of business with new features for its operating systems, from password managers to sleep trackers. And don’t forget that Claude and ChatGPT were built in part by scraping vast amounts of material from the internet. The process of monkey-see, monkey-do is a large part of what makes Silicon Valley work.

但这也引发了一个问题:Anthropic的客户应在多大程度上信任该公司。包括Palantir首席执行官Alex Karp和前白宫AI沙皇David Sacks在内的多位科技行业领袖警告称,企业使用AI时,做出的权衡与社交媒体用户在隐私方面做出的权衡属于同一类。

But that also raises questions about how much Anthropic’s customers should trust the company. Several tech industry leaders, including Palantir CEO Alex Karp and former White House AI Czar David Sacks, have warned that when businesses use AI, they are making the same kind of trade-off that social media users do with their privacy.

“本质上,你为智能支付了两次费用,”微软首席执行官萨提亚·纳德拉在7月的一篇博客文章中写道。“一次是用金钱,另一次是用更有价值的东西:为了让这种智能发挥作用,你必须透露的专有知识。”纳德拉建议大型企业构建自己的专有AI模型,以保留所有洞察和知识产权,以及使用这些模型时可能产生的任何数据“废气”。

“You essentially pay for intelligence twice,” Microsoft CEO Satya Nadella wrote in a July blog post. “Once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful.”Nadella recommended that large firms build theirown proprietary AI models to retain all insights and IP, as well as any data “exhaust” that might be emitted when using them.

Anthropic和OpenAI都坚称,客户数据是机密的,且归其客户所有。“默认情况下,我们不会使用商业产品中的客户内容来训练我们的模型。我们的商业条款禁止这样做,”Anthropic的一位发言人告诉我。“我们也不会阅读个别客户的对话、代码或文件,以制定我们的战略、决定开发什么产品,或选择进军哪些市场。”尽管如此,业内一些人仍对Anthropic如何处理客户信息提出质疑。

Both Anthropic and OpenAI insist that customer data is confidential and owned by their clients. “By default, we do not train our models on customer content from our commercial products. Our commercial terms prohibit it,” a spokesman for Anthropic told me. “We also don't read individual customer conversations, code, or files to shape our strategy, decide what products to build, or choose which markets to pursue." Still, some in the industry are raising questions about how Anthropic processes customer information.

一个争议领域是Anthropic的安全监控,因为这可能涉及保留和审查一些对话,以检查是否存在与化学或生物武器等相关的危险请求。目前不清楚Anthropic内部谁可以访问这些信息,以及这些信息可能会如何被使用。

One area of contention is Anthropic’s safety monitoring, since it can involve keeping and reviewing some conversations to check for dangerous requests related to things such as chemical or biological weapons. It’s not clear who inside Anthropic can access that information and how it might be used.

但Anthropic不需要挖掘客户的秘密就能进入他们的市场。医疗保健看起来像是该公司进军的一个可能路径,尽管Alphabet Inc.和International Business Machines Corp.等大型科技公司过去尝试过但均告失败,后者的Watson超级计算机早在2012年就旨在治愈癌症。

But Anthropic doesn’t need to mine its customers’ secrets to move into their markets. And healthcare looks like one likely path for the company to push into, despite a history of scuppered attempts by Big Tech players including Alphabet Inc. and International Business Machines Corp., whose Watson supercomputer was meant to cure cancer way back in 2012.

前生物物理学研究员阿莫代伊在2024年写道,“智能在生物学和医学领域的回报很高”。Anthropic已在旧金山建立了一家生物实验室,并于去年11月收购了生物技术初创公司Mithril Medicine,作为其药物发现工作的一部分。如今,随着它进入一个充满财务和声誉潜力的领域,问题在于它将走多远——以及这对其医疗保健客户意味着什么。

Amodei, a former biophysics researcher, wrote in 2024 that “the returns to intelligence are high” in biology and medicine. Anthropic has already established a biology lab in San Francisco and in November bought biotech startup Mithril Medicine as part of its drug-discovery efforts. Now that it’s entering a field rich with financial and reputational potential, the question is how far it will go — and what that might mean for its healthcare customers.

阅读媒体原文 ↗