Amazon’s ‘Switzerland’ Days Are Over; Are LLMs Cheating on Benchmark Tests?; OpenAI Quietly Develops Customizable GPTs

Looks like Amazon’s finally made its AI bet.

On Monday, model developer Anthropic announced that it had raised as much as $4 billion from Amazon. As part of the deal, Amazon Web Services will become Anthropic’s primary cloud provider, and Anthropic will train and deploy its future models on AWS’ training- and inference-specialized chips, Trainium and Inferentia. 

What that means isn’t totally clear, however, since Google in February invested $400 million in Anthropic and announced it would be Anthropic’s “preferred” cloud provider. I’d imagine that label is trumped by AWS’ “primary” designation. That isn’t the only unknown in the deal either—we have yet to learn Anthropic’s latest valuation, the terms of the deal or whether it comes with a revenue or profit-sharing agreement between the two companies.

The Amazon investment is the latest in a series of AI bets from the major cloud providers, but this one is a bit more unusual than most. AWS has consistently framed itself as the “Switzerland” of the hyperscalers by providing customers with a range of large-language models. That puts it in stark contrast with Microsoft, the only cloud provider that resells OpenAI software.