亚马逊收购 Anthropic，意在切断英伟达的芯片垄断

原文作者：Kate Clark,Anissa Gardizy and Stephanie Palazzolo 编辑：阿 a，注： 在这篇文章中我们使用了 Notion AI 进行翻译 

据三位直接知情人士透露，在亚马逊上周承诺向该公司投资 12.5 亿美元后，OpenAI 的主要竞争对手之一 Anthropic 正在与投资者谈判，以筹集至少 20 亿美元的新资金。其中两人表示，Anthropic 告诉投资者，谷歌在 2022 年购买了这家初创公司约 10%的股份，预计将投资这一轮。 

据其中一人称，这家成立两年的初创公司出售与 OpenAI 的 ChatGPT 竞争的聊天机器人 Claude，其估值在 200 亿至 300 亿美元之间，包括新投资。这将使该公司自 3 月以来的估值增加五倍，当时投资者对该公司定价 40 亿美元，并使其股票在收入估值倍数方面远高于 OpenAI 的股价。 

要点 ‍ 

• Anthropic 正在与竞争对手 OpenAI 进行衡量 

• Anthropic 的收入是 OpenAI 的十分之一 

• 投资者一直愿意为人工智能初创股票支付高价 

这一轮尚未敲定，交易条款可能会发生变化。一位了解筹款工作的人表示，Anthropic 也在与主权财富基金和其他机构投资者交谈。Anthropic 的发言人没有立即发表评论。 

此举表明，领先的人工智能开发人员如何竞相获得尽可能多的资本和计算资源，以开发下一代软件，即大型语言模型。人工智能开发特别昂贵，因为它需要专门的服务器芯片，比处理传统计算工作的处理器需要更多的功率。与此同时，私人投资者一直愿意为领先的人工智能开发人员的股票承受更高的价格。 

在 OpenAI 与今年早些时候为这家初创公司投入超过 100 亿美元资金的主要云服务器提供商微软结盟后，其他独立的人工智能模型开发人员，如 Anthropic，也试图找到自己的计算捐助者。这促使 Anthropic 最初达成协议，使用谷歌云和搜索公司的投资，Cohere 与甲骨文的合作，以及 Inflection AI 从微软和 AI 芯片设计师 Nvidia 等公司筹集的资本。 

除了计算资源外，这些初创公司还指望云提供商帮助向客户销售他们的人工智能软件。他们希望追随 OpenAI 的脚步，OpenAI 最近正准备每年创造超过 10 亿美元的收入，部分原因是其软件通过微软的云部门销售。 

两位直接知情人士表示，Anthropic 告诉一些投资者，它一直在以 1 亿美元的年化率创造收入，这意味着它每月创造了超过 800 万美元的收入。这意味着对 Anthropic 的投资估值为 200 亿美元，未来收入的估值倍数将达到 200 倍，远高于 OpenAI 的估值倍数。 

收入希望

其中一位知情人士说，Anthropic 预计，到今年年底，公司的年化营收将达到 2 亿美元，这意味着每月营收将接近 1700 万美元。此人说，到 2024 年底，Anthropic 希望每月收入超过 4,000 万美元，年化收入达到 5 亿美元。Anthropic 最近与亚马逊(Amazon)达成协议，后者将向云客户出售该初创公司的人工智能软件，这是该公司预计增长的一个关键原因。亚马逊表示，包括研究提供商 LexisNexis、资产管理公司 Bridgewater Associates 和旅游出版商 Lonely Planet 在内的客户都在通过亚马逊网络服务使用 Claude。 

Anthropic 的人工智能软件被认为是新兴行业中最好的软件之一，该公司长期以来一直与 OpenAI 相比，OpenAI 在今年早些时候的员工股票销售中价值约为 270 亿美元。据一位知情人士透露，OpenAI 希望以更高的估值再次出售员工股票。据《华尔街日报》报道，OpenAI 的目标是股价，这意味着该公司的估值至少为 800 亿美元。 

以这个价格，OpenAI 投资者将支付大约 80 倍于收入的估值倍，远高于公共科技股的估值。 

投资者也会收购一家具有不同寻常公司结构的公司。Anthropic 成立于 2021 年，拥有一个由五名个人组成的独立机构，在公司中没有财务股份，可以选举和罢免一些 Anthropic 的董事会成员。Anthropic 表示，该集团旨在使 Anthropic 的目标与公众的利益保持一致。 

AWS 的理由

尽管如此，云提供商有很多理由收购 Anthropic。上周，一直在努力为云客户开发自己的人工智能产品的亚马逊表示，Anthropic 将使用 AWS 的服务器开发人工智能。亚马逊表示，除了最初的 12.5 亿美元外，它可能会额外提供 27.5 亿美元的资金，该资金的形式是可转换票据，当 Anthropic 筹集更多资金时，该债券将变成股票。 

亚马逊与 Anthropic 的交易还有另一个原因。亚马逊开发了特殊的服务器芯片 Trainium 和 Inferentia，以推动人工智能开发，并希望更多的云客户使用这些芯片，而不是 Nvidia 的图形处理单元，后者一直供不应求。作为亚马逊-Anthropic 交易的一部分，Anthropic 同意使用亚马逊的芯片。AWS 是英伟达芯片的最大买家之一，但随着英伟达开发自己的云服务，这些公司越来越不一致。 

如果 Anthropic 从谷歌筹集更多资金，这将引发人们对这家初创公司计划如何在两家竞争对手云提供商之间瓜分效忠的问题。谷歌有自己的人工智能模型，并计划将其出售给云客户，而 AWS 没有相同的内部人工智能口径，因此它需要与 Anthropic 等公司合作和销售软件。（AWS 最近向所有客户提供了自己的 Titan LLM。） 

虽然 Anthropic、OpenAI 和 Google 开发的 LLM 远非完美，但它们已经展示了帮助软件开发人员更快地编写代码的能力，并帮助业务经理总结文档或自动生成广告和其他营销内容。一些 LLM 开发人员开始将该技术称为下一个软件操作系统，因为模型能够编写和运行代码、访问互联网以及检索和引用文件。 

不包括最新的亚马逊投资，Anthropic 已经从包括 Spark Capital 和 Salesforce Ventures 在内的投资者那里筹集了超过 11 亿美元的资金。据彭博社报道，2022 年 4 月，这家 LLM 开发商还在现在破产的加密交易所 FTX 的创始人兼首席执行官 Sam Bankman-Fried 的带领下筹集了 5.8 亿美元。据彭博社报道，FTX 此后停止了出售其 Anthropic 股份。 

Kate Clark 是 The Information 负责风险资本覆盖的副局长。她是《Dealmaker》的作者，这是一篇关于 VC 的每周专栏。她住在纽约，可以在推特上找到 @KateClarkTweets。您可以通过信号+1（415）-409-9095 与她联系。 

Anissa Gardizy 是 The Information 报道云计算的记者。她之前是《波士顿环球报》的科技记者。Anissa 位于旧金山，可以通过 anissa@theinformation.com 或推特 @anissagardizy8 联系 

Stephanie Palazzolo 是 The Information 报道人工智能的记者。她之前曾在 Insider 和摩根士丹利工作。总部位于纽约，可以通过 stephanie@theinformation.com 或推特 @steph_palazzolo 与她联系。 

Anthropic, one of the main rivals of OpenAI, is in talks with investors to raise at least $2 billion in new funding, following a commitment from Amazon last week to invest $1.25 billion in the company, according to three people with direct knowledge. Anthropic has told investors that Google, which bought a roughly 10% stake in the startup in 2022, is expected to invest in the round, two of the people said. 

The 2-year-old startup, which sells Claude, a chatbot that competes with OpenAI’s ChatGPT, wants a valuation between $20 billion to $30 billion including the new investment, according to one of those people. That would quintuple the valuation of the company since March, when investors put a $4 billion price tag on the firm, and make its shares far pricier than those of OpenAI in terms of its valuation multiple on revenue. 

THE TAKEAWAY 

• Anthropic is measuring itself against rival OpenAI 

• Anthropic generates one-tenth the revenue of OpenAI 

• Investors have been willing to pay high prices for AI startup shares 

The round is not finalized and terms of the deal could change. One of the people with knowledge of the fundraising effort said Anthropic is also talking to sovereign wealth funds and other institutional investors. Spokespeople for Anthropic didn't immediately comment. 

The move shows how the leading developers of artificial intelligence are racing to secure as much capital and computing resources as possible to develop the next generation of their software, known as large language models. AI development is particularly expensive because it requires specialized server chips that require more power than processors that handle traditional computing work. Private investors, meanwhile, have been willing to stomach ever-higher prices for shares of leading AI developers. 

After OpenAI aligned itself with Microsoft, a major cloud server provider that committed more than $10 billion in funding for the startup earlier this year, other independent AI model developers such as Anthropic have sought to find their own computing benefactors. That prompted Anthropic’s initial pact to use Google Cloud and an investment from the search company, Cohere’s partnership with Oracle, and Inflection AI’s capital raise from Microsoft and AI chip designer Nvidia, among others. 

In addition to computing resources, the startups are counting on the cloud providers to help sell their AI software to customers. They hope to follow in the footsteps of OpenAI, which was recently on pace to generate more than $1 billion in revenue annually, boosted in part by sales of its software through Microsoft’s cloud unit. 

Anthropic has told some investors it has been generating revenue at a $100 million annualized rate, implying it has generated more than $8 million in revenue per month, said two people with direct knowledge. That would mean an investment in Anthropic at a valuation of $20 billion would carry a valuation multiple on future revenue of 200 times, far more expensive than OpenAI’s valuation multiple. 

Revenue Hopes 

By the end of this year, Anthropic projected it would generate revenue at a $200 million annualized pace, implying nearly $17 million in monthly revenue, one of the people said. And by the end of 2024, Anthropic hopes to generate more than $40 million in monthly revenue, a $500 million annualized rate, this person said. Anthropic’s recent deal with Amazon, in which the cloud provider will sell the startup’s AI software to cloud customers, is a key reason for the projected growth. Amazon has said customers including research provider LexisNexis, asset management firm Bridgewater Associates, and travel publisher Lonely Planet were using Claude through Amazon Web Services. 

Anthropic’s AI software is considered among the best in the nascent industry and the company has long measured itself against OpenAI, which was valued at about $27 billion in an employee share sale earlier this year. OpenAI is looking to sell employee shares again at a much higher valuation, according to a person with knowledge of the situation. OpenAI is targeting a share price that implies a valuation of at least $80 billion for the company, the Wall Street Journal reported. 

At that price, OpenAI investors would be paying a valuation multiple of around 80 times revenue, far above the valuations of public tech stocks. 

Investors would also be buying into a company with an unusual corporate structure. Anthropic, founded in 2021, has an independent body of five individuals with no financial stake in the company who can elect and remove a number of Anthropic’s board members. Anthropic said this group aims to align Anthropic’s goals with the interests of the general public. 

AWS’ Rationale 

Nonetheless, cloud providers have had plenty of reasons to buy into Anthropic. Last week, Amazon, which had struggled to develop its own AI products for cloud customers, said Anthropic would develop AI using servers from AWS. Amazon said it may provide an additional $2.75 billion in funding on top of the initial $1.25 billion, which came in the form of a convertible note that would turn into stock when Anthropic raises more money. 

Amazon had another reason for its deal with Anthropic. Amazon has developed special server chips, Trainium and Inferentia, to power AI development and has wanted more cloud customers to use those chips rather than Nvidia’s graphics processing units, which have been in short supply. As part of the Amazon-Anthropic deal, Anthropic agreed to use Amazon’s chips. AWS is one of the biggest buyers of Nvidia chips but the companies have increasingly been at odds as Nvidia develops its own cloud service. 

If Anthropic raises more money from Google, it would raise questions about how the startup plans to split its allegiance between the two rival cloud providers. Google has its own AI models and plans to sell them to cloud customers, whereas AWS doesn’t have the same caliber of in-house AI so it has needed to collaborate and sell software from firms such as Anthropic. (AWS recently made its own Titan LLMs available to all its customers.) 

While LLMs developed by Anthropic, OpenAI and Google are far from perfect, they have shown the ability to help software developers code faster and help business managers summarize documents or automatically generate ads and other marketing content. Some LLM developers have begun referring to the technology as the next software operating system because of the models’ ability to write and run code, access the internet and retrieve and reference files. 

Excluding the latest Amazon investment, Anthropic has raised more than $1.1 billion in funding from investors including Spark Capital and Salesforce Ventures. In April 2022, the LLM developer also raised $580 million in a round led by Sam Bankman-Fried, the founder and CEO of the now-bankrupt crypto exchange FTX. FTX has since halted the sale of its Anthropic stake, Bloomberg reported. 

Kate Clark is deputy bureau chief responsible for venture capital coverage at The Information. She's the author of Dealmaker, a weekly column on VC. She is based in New York and can be found on Twitter at @KateClarkTweets. You can reach her via Signal at [+1 (415)-409-9095](tel:+1 (415)-409-9095). 

Anissa Gardizy is a reporter at The Information covering cloud computing. She was previously a tech reporter at The Boston Globe. Anissa is based in San Francisco and can be reached at anissa@theinformation.com or on Twitter at @anissagardizy8 

Stephanie Palazzolo is a reporter at The Information covering artificial intelligence. She previously worked at Insider and Morgan Stanley. Based in New York, she can be reached at stephanie@theinformation.com or on Twitter at @steph_palazzolo.