Amazon Investing $4B in Gen AI Bot Maker Anthropic

Everyone is using arms race analogies to describe the rapidly unfolding competition for hegemony in the emerging technology of generative AI. 

While, at least for the near term, we’re yet to see the AI equivalent of a hydrogen bomb, we are beginning to see the formation of global alliances backed by AI superpowers.

Microsoft ignited the chain reaction at the beginning of the year with its $10B investment in ChatGPT pioneer OpenAI, which produced the generic large language models that are being used to create a series of domain-specific AI co-pilots.

This week comes news of Amazon’s $4B investment in generative AI bot-maker Anthropic, a wide-ranging collaboration that will produce foundation models for Amazon Bedrock using Amazon Web Services (AWS) Trainium and Inferentia chips.

Anthropic said it will offer “enhanced support” of Amazon Bedrock, which produces foundation models, with “secure model customization and fine-tuning” for businesses. Amazon teams will also be able to use Bedrock to build on Anthropic’s models, cointelegraph reported.

The San Francisco-based startup said in a post on social media platform X that the agreement is “part of a broader collaboration to develop reliable and high-performing foundation models.”

Anthropic was formed by a group of OpenAI technicians who broke with the ChatGPT producer over the release of bots without built-in security layers. The company has made safety and reliability the core of the branding for its foundation platform, known as Claude Instant.

In the emerging cyberscape of generative AI, generic foundation models are being marketed like cloud-based APIs—in the words of a proptech player, the basic bots quickly have become “utilities.”

So, it’s not unusual for a successful startup like Anthropic, which inked a cloud deal seven months ago with Google, an early investor, to sign a bigger deal to make AWS its primary cloud provider for “mission critical workloads”—a partnership that is being described by Amazon as acquiring a “minority stake” in Anthropic.

Anthropic, which prior to Amazon’s investment had raised more than $1.1B in venture capital, last month introduced what it called a faster, cheaper foundation model available through an API known as Claude Instant.

Anthropic said version 1.2 is less likely to hallucinate and more resistant to what is known as “jailbreaking” attempts. Hallucinating involves bots basically making stuff up, while jailbreaking is a technique in which users create prompts that bypass built-in safety features of the model.

According to tech industry analysts, the big news in Amazon’s deal with Anthropic involves the agreement to use Trainium and Inferentia chips to “build, train and deploy” new models based on AWS’ AI-programmable semiconductors.

If you think of bots as nukes, the chips are the plutonium at the core of device. Thus far, AI-enabled chips produced by Nvidia have dominated as the primary brain behind the ever-more-powerful gen AI bots, a tilted playing field Amazon is hoping to level off with its own chips.

According to Kunjan Sobhani, a Bloomberg Intelligence analyst, it’s unclear whether the AWS chips can match the firepower of Nvidia’s chips. Sobhani suggested that “at the very least,” the deal with Anthropic positions AWS to produce bots that don’t necessarily require the full capabilities of Nvidia’s technology.

“It’s like having a Ferrari and a BMW. The Ferrari will get you there quicker, but you don’t always need it,” he said.

And, if generative AI leads to a new form of mutually assured destruction, you can skip the cars and get into your Blue Origin rocket, which will get you out of here faster than both.