Amazon’s Partnership With Anthropic Shows Size Matters in the AI Industry

5 minute read

Amazon announced today (Sept. 25) it will invest up to $4 billion into Anthropic, an AI developer known for developing language models such as Claude 2, a competitor to OpenAI’s ChatGPT.

As part of the deal, Amazon, the world’s largest provider of cloud infrastructure services through its AWS unit, will become the primary provider of computational processing power, also called compute, for Anthropic. The process of training and running state-of-the-art AI models requires vast amounts of compute, and many analysts expect future AI models to require increasing amounts of compute.

In return, Amazon will acquire a minority ownership position in Anthropic, and Amazon’s engineers will be able to incorporate Anthropic’s AI models into their products and services such as Amazon’s personal assistant, Alexa. Anthropic has also committed to offering its models via Bedrock, Amazon’s online platform on which it hosts foundation models—broadly capable AI models that can be adapted for different tasks.

Anthropic was founded in 2021, after a group of OpenAI employees left over differences in their approach to AI safety. Since then, it’s become one of the leading foundation model developers. (Investors in Anthropic also include Salesforce, where TIME co-chair and owner Marc Benioff is CEO.)

The deal is similar to, although smaller than, Microsoft’s partnership with OpenAI, which began in 2019. Microsoft has reportedly invested $13 billion for access to OpenAI’s models, such as GPT-4, while offering OpenAI the use of Microsoft’s Azure cloud computing services. DeepMind, another AI model developer, was acquired by Google in 2014 for a price reportedly ranging between $400 million and $650 million, and merged with Google Brain to form Google DeepMind this year.

Amazon fights back

Amazon was the only remaining company among the three large cloud computing providers without a large investment in a foundation model developer. Instead, Amazon had chosen to offer access through Bedrock to a range of foundation models produced by third-party developers, such as Anthropic and Cohere. The Anthropic investment is a break from that strategy, says Jai Vipra, a former research fellow at the AI Now Institute, a U.S.-based nonprofit.

Although industry insiders have made much of Amazon falling behind Google and Microsoft, the gap between the companies is fairly small, says Anton Korinek, an economics professor at the University of Virginia. On the other hand, in AI development, “just being 12 months ahead, it makes a world of a difference, because the sector is so dynamic,” he says.

Size matters

Over time, the increasing amounts of capital extended by cloud computing providers has resulted in their exerting more control over AI developers, says Vipra. The relationship between Microsoft and OpenAI began as a partnership, but Microsoft’s large investment in OpenAI now means it has more say over how OpenAI is run, she says. In June, the Wall Street Journal reported that Microsoft released its Bing chatbot, powered by OpenAI’s GPT-4, despite warnings from OpenAI that there were still issues to address.

“DeepMind was acquired by Google, but it had quite a bit of independence for a long time,” says Vipra. “But then, again, compute as a scarce resource is what drove Google Brain and DeepMind to merge.”

Anthropic’s leadership, like the leadership at many other prominent AI developers, have warned that powerful AI systems could go off the rails or be misused. But, in order to understand how to make the most powerful AI models safely, they say they need to continue to build increasingly large AI models. “There’s this intertwinement—it’s one of the things that makes the problem hard—between the safety problems and the kind of inherent capabilities of the model,” Dario Amodei, CEO of Anthropic, told TIME in a previous interview

Vipra says that Anthropic, which prides itself on being particularly safety-conscious, will struggle to avoid being influenced by Amazon. “The announcement says it's all about safety—it's pitched as expanding access to safe AI products," she says. "But it really shows that productization has taken over.”

In the press release, Anthropic stressed its corporate governance setup would allow it to continue to develop AI responsibly. In the same interview with TIME, Amodei argued that Anthropic was not contributing to a reckless race to build increasingly powerful AI systems that would be counterproductive to its goal of building safe AI, but said it needed to build bigger AI models to fulfill its mission. “One way or another, the scaling up of the models is part of our plan,” he said. “Simply not doing it, I think, isn’t a solution.”

AI systems require three inputs—compute, data, and human talent in the engineers and researchers that design algorithms. Although developers like Anthropic and OpenAI have had to partner with big tech companies to access compute, and may increasingly need to do so in future for access to data, they still have an advantage when it comes to talent, says Korinek.

“The top talent …are drawn to these labs, like a magnet,” says Korinek. “It's not going to be easy for larger, more bureaucratic organizations to attract the same talent.” As long as this advantage persists, foundation model developers will be able to extract concessions from the cloud providers they partner with, he says.

Competition concerns

A report published last week by the U.K.’s Competition and Markets Authority, the nation’s antitrust watchdog, raised concerns about partnerships between cloud providers and foundational model developers, such as Amazon’s investment into Anthropic. A blog post published by the U.S. Federal Trade Commission in June raised similar concerns.

“Whenever you have the vertical integration, it allows players to kind of leverage the monopoly power of where they are the strongest, and extend that to the other layers,” says Korinek.

“Amazon has been in the crosshairs of antitrust regulation,” says Vipra. The addition of Amazon, she argues, “might, in fact, be bad for competition, just given all the other markets that Amazon has its hands in.”

More Must-Reads from TIME

Write to Will Henshall at will.henshall@time.com