What you need to know about Anthropic, the AI startup Amazon will invest up to $4 billion in

In a major development in the AI arms race, Amazon announced this morning that it has committed to paying $1.25 billion to take a minority stake in Anthropic, a developer of generative AI models similar to those that power OpenAI’s ChatGPT. Amazon’s investment in Anthropic, which was started by a group of ex-OpenAI members, could go up to $4 billion, the company says. The deal is a big step for Anthropic, but it also sheds some light on the tactics of the combatants in the AI arms race now raging in the tech industry.

Amazon has been working on its own large language models, and it already distributes Anthropic’s LLMs through its AWS. Amazon will likely be on more exclusive terms with Anthropic after the investment, and the relationship could tighten in the future. Amazon could go further with Anthropic’s models, perhaps integrating them or Anthropic’s Claude chatbot, into its marketplace in various ways. This would make the Anthropic/Amazon tie-up look more like OpenAI’s arrangement with Microsoft, which has integrated OpenAI models into its core productivity apps.

The deal shows that huge, well-monied tech companies like Amazon can’t always duplicate the “secret sauce” that highly talented (and highly paid) AI researchers can bake into models at far smaller shops like OpenAI and Anthropic. It also shows that big tech companies are willing to write big checks to get the secret sauce, underlining a growing belief that conversational AI interfaces will be our preferred way of interacting with technology in the future.

Much of the recipe for Anthropic‘s secret sauce was written by the company‘s CEO Dario Amodei, who remains an underexposed figure from the AI world. Before founding Anthropic he was lead researcher at OpenAI, and was heavily involved with that company’s major advances: The company witnessed astonishing performance improvements in its language models after it scaled them up to huge sizes and put huge amounts of computing power underneath them. It’s how ChatGPT began conversing like a human being (sort of).

You might wonder why companies like OpenAI and Anthropic wouldn’t just go it alone, holding out to become the Googles and Apples of tech’s next epoch. While talented researchers are extremely important, success in the AI arms race depends a lot on a company’s access to large amounts of compute power (in the form of powerful GPUs within server clusters), especially as AI models grow larger and work harder. The need for this compute power is likely the reason that Anthropic has agreed to sell off a big chunk of its company to Amazon. Anthropic earlier took investment money from Google/Alphabet ($300 million), but Amazon may have just secured the inside track to Anthropic’s AI work for years into the future.