Amazon Develops New AI Chip at Half NVIDIA's Cost

Amazon is advancing its server technology with a new AI-focused design currently under testing at its Austin, Texas, chip lab. This latest development, from Amazon Web Services' (AWS) Annapurna Labs, features proprietary AI chips developed to rival those from industry leader NVIDIA.

The initiative aims to reduce Amazon's dependency on NVIDIA’s high-cost chips, which are used to power certain aspects of AWS's AI cloud services. By developing its own processors, Amazon seeks to enable customers to perform complex computations and manage large datasets more cost-effectively.

This move aligns with similar efforts by major competitors like Microsoft and Alphabet, who are also working on their own technological advancements.

Since acquiring Annapurna Labs in 2015, Amazon has noted a growing demand from its customers for more affordable alternatives to NVIDIA’s chips. According to Rami Sinno, Engineering Director at Annapurna Labs, AWS’s Vice President of Compute and Networking, David Brown, expects performance gains of up to 40% or even 50%, with potential cost reductions of up to 50% compared to NVIDIA’s offerings.

Although Amazon's AI chip development is in the early stages, its flagship processor for non-AI tasks, the Graviton, has been under development for nearly a decade and is now in its fourth generation, known as AWS Graviton4. Newer AI-focused chips, Trainium 2 and Inferentia, represent the latest innovations in Amazon's chip portfolio.

AWS, which contributes nearly 20% to Amazon’s total revenue, experienced a 17% sales increase year-over-year, reaching $25 billion in the first quarter of this year. AWS holds approximately one-third of the cloud computing market, while Microsoft’s Azure occupies about 25%.