Amazon Discussing New Multibillion-Dollar Investment in Anthropic
The Takeaway
• The funding discussion shows how cloud providers and AI developers they partner with have competing priorities
• An Amazon official has said Anthropic’s CEO expressed interest in a large supercomputing cluster of servers
• Anthropic’s ability to use servers from two different cloud providers might be an advantage
Amazon is discussing making a second multibillion-dollar investment in OpenAI rival Anthropic, according to a person involved in the discussions. The new deal is similar to Amazon’s initial $4 billion dollar investment in the startup, which was struck last year. But this time, Amazon wants Anthropic to make a concession.
The cloud giant is asking Anthropic, which uses Amazon’s cloud services to train its AI, to use a large number of servers powered by chips developed by Amazon, this person said. The problem is that Anthropic prefers to use Amazon servers powered by Nvidia-designed AI chips.
The size of Amazon’s total investment in Anthropic could depend on the outcome of this discussion, specifically on the number of Amazon chips Anthropic agrees to use, this person said. The status of the talks couldn’t be learned.
The discussions are an example of the competing priorities of large cloud providers and developers of conversational AI that have formed alliances due to the high cost and complexity of producing the technology. The first such marriage in the industry, between Microsoft and OpenAI, has been remarkably beneficial to both companies but has lately become fraught over OpenAI’s concerns that it isn’t getting enough servers from Microsoft to stay ahead of smaller AI rivals. And while Microsoft is developing its own AI server chip, which it hopes OpenAI will want to use, OpenAI hasn't been interested in it, said a person with direct knowledge of the situation. (OpenAI is also developing a chip to run its AI models.)
Shifting to the Amazon server chip could be technically challenging for Anthropic because the Amazon software that developers must use with the Traininum chips isn’t as mature as Nvidia’s Cuda software, which AI developers have become accustomed to. Such a move could also lock Anthropic into using Amazon Trainium servers, making it more difficult for the AI startup to use other cloud providers or to lease its own data centers in the future, as Amazon doesn’t make its hardware available to facilities run by other companies.
Amazon, though, has good reason to get Anthropic to use its own chips, known as Trainium: The cloud giant could reduce the number of Nvidia chips it has to buy. If it can get its cloud customers to agree to use Trainium-powered servers, it won’t need as many Nvidia chips. As part of Amazon's initial deal with Anthropic, the startup agreed to use some Trainium servers but mainly relied on Nvidia servers in Amazon data centers, said the person who has been involved in the discussions involving the companies.
Spokespeople from Amazon and Anthropic declined to comment.
Amazon has increasingly been selling artificial intelligence services powered by Anthropic, a major OpenAI rival, to Amazon’s cloud customers.
Anthropic recently sought funding from investment firms at a valuation of $30 billion to $40 billion, and any investment deal with Amazon could come in the form of convertible notes that become equity after Anthropic raises capital from other investors. In addition to the investment, the companies are negotiating a cloud deal in which the companies share revenue from the sale of Anthropic’s model to Amazon cloud customers such as Doordash and Goldman Sachs, and Anthropic agrees to rent out specialized servers from Amazon to develop its technology.
Amazon a year ago agreed to invest up to $4 billion into Anthropic in a similar deal that was completed earlier this year. Since then, Anthropic has likely spent hundreds of millions of dollars to rent Amazon servers and shared hundreds of millions of additional dollars with Amazon for reselling its models to cloud customers. Amazon also uses Anthropic to power its Q coding assistant for software developers, which competes with ChatGPT, Microsoft’s GitHub Copilot and coding assistant startup Cursor.
The partnership has helped Amazon’s cloud unit maintain its revenue growth rate of 19% in the third quarter though companies using generative AI have been making spending cuts elsewhere in their IT and cloud budgets, an Amazon Web Services executive told The Information.
An Anthropic Supercomputer?
A new deal could bring the companies even closer. A senior Amazon official privately said Anthropic CEO Dario Amodei earlier this year discussed his interest in using a large-scale AI data center server cluster to develop technology, similar to the ambitious data center plans of rivals such as Elon Musk’s xAI and OpenAI, according to a person who spoke to the official. It isn’t clear whether Amazon has committed to building a supercomputing cluster for Anthropic.
Anthropic has a similar but smaller cloud partnership with Google, which has also invested billions of dollars into Anthropic. Google has also been selling its own AI, Gemini, to Google Cloud customers, meaning it hasn’t been as reliant on Anthropic as Amazon has.
Amazon also has been developing its own AI that could eventually rival Anthropic’s.
OpenAI is in a much stronger financial position than Anthropic, in terms of losses as a percentage of revenue. Anthropic recently projected it would generate $83 million of revenue a month by the end of this year, with 25% to 50% of that figure being paid out to Anthropic’s cloud partners in the form of revenue sharing. OpenAI generates four to five times more revenue than Anthropic and pays a smaller percentage to Microsoft, its exclusive cloud provider.
Anthropic’s ability to use servers from two different cloud providers might be an advantage compared to OpenAI’s situation. OpenAI has been frustrated with Microsoft’s ability to provide it with servers, prompting OpenAI to seek an alternative provider—Oracle and Crusoe.