Amazon is developing custom artificial intelligence chips to lessen its reliance on NVIDIA, according to a report by the Financial Times. This initiative follows Amazon's investment in a chip design startup in 2015 and is part of its broader strategy to create in-house processors for data center workloads.
The new chips, developed by Amazon's Annapurna Labs, are already being utilized by Anthropic, Amazon's primary AI partner, which provides access to the Claude foundational AI model. This move aligns with a trend among tech giants to decrease dependence on NVIDIA's GPUs, which are currently the market leaders for AI workloads.
Amazon aims to cut costs and enhance efficiency by producing its own AI chips, including the Trainium series designed for large language models. The Trainium2 was unveiled in November 2023, although supply constraints have limited its adoption.
The chips are designed using technology from Alchip and are manufactured by Taiwan Semiconductor Manufacturing Company (TSMC). More than 50,000 AWS customers are currently using Amazon's Graviton processors, which complement the Trainium chips.
Other major companies, including Alphabet and Meta, are also developing their own AI chips to reduce reliance on NVIDIA. Google recently introduced its latest tensor processing unit (TPU), Trillium, which significantly enhances AI training and inference speeds.