Amazon AI chief: Hope to catch OpenAI, Anthropic in the ‘coming year’

Amazon has been looking to catch up to OpenAI and Anthropic on frontier AI with its Nova models.
AI chief Peter DeSantis said he hopes Amazon is in the conversation about leading models “in the coming year.”
Amazon is looking to build on its AI strategy, which includes the development of custom chips.

Amazon’s top artificial intelligence executive told CNBC on Wednesday that he hopes the company will be able to compete with OpenAI and Anthropic on frontier models in the “coming year” after falling behind the two leading labs.

“I think it’s a fair narrative that our models haven’t been at the very frontier for the very largest, most demanding workloads,” Peter DeSantis, a senior vice president at Amazon who heads up the company’s semiconductor, AI and quantum efforts, told CNBC.

The “frontier” refers to the most advanced AI models.

“We’ve been taking a very deliberate approach to get our foundations right, our data, our architecture, our infrastructure. And you know, we’re on a path that we want to be on,” he said.

The aspiration underscores the effort Amazon is putting into its model development as it looks to reassure investors that it is a key player in the AI boom.

Amazon’s approach to AI models has been two-pronged. On the one hand, it has a product called Bedrock, which is a sort of marketplace for models from different companies that its cloud computing customers can access. On the other hand, Amazon released Nova2, its latest AI model, in December in a bid to compete with the likes of OpenAI and Anthropic.

“We’ve got about 50,000 customers for Nova2, so we’re pretty excited about it,” DeSantis said.

“Our aspiration is to have a model that people think about as one of the very most capable intelligent models out there,” he said. “I’m not sure we’re there yet with Nova2, but that’s our aspiration.”

Amazon’s AI strategy also revolves around its semiconductors under the Trainium and Graviton brands.

The tech giant has been designing its own custom chips for several…

Source