Aiming to provide more cost-effective solutions for its Amazon Web Services customers, Amazon is developing AI chips in its Texas lab as an alternative to Nvidia’s expensive ones.
E-commerce giant Amazon is developing and testing its own AI chips to reduce dependence on expensive Nvidia processors, which are crucial for its AI cloud services at Amazon Web Services (AWS), as per a Reuters report last week.
According to the report, engineers are rigorously testing a new server design featuring these chips in their laboratory in Austin, Texas, aiming to compete with Nvidia. The move is part of Amazon’s strategy to minimise reliance on the costly ‘Nvidia tax’ associated with its current AI infrastructure.
With its homegrown chips, Amazon aims to offer customers more cost-effective solutions for complex calculations and large data processing. This effort comes as competitors like Microsoft and Alphabet are pursuing similar strategies.
According to Rami Sinno, director of engineering at Amazon’s Annapurna Labs within AWS, there is a growing demand among customers for cheaper alternatives to Nvidia’s chips.
Amazon acquired Annapurna Labs in 2015. While its AI chip initiatives are still emerging, Amazon has been developing its Graviton chip for non-AI tasks for nearly a decade, now in its fourth generation. In contrast, its newer AI chips, Trainium and Inferentia, are more recent innovations.
According to David Brown, AWS Vice President of Compute and Networking, these new chips offer up to 50% better price performance, potentially making them half as costly as Nvidia’s solutions.
As per reports, AWS, contributing nearly 20% to Amazon’s total revenue, saw a 17% increase in sales to $25 billion for the January-March quarter YoY. AWS holds approximately 33% of the cloud computing market, while Microsoft’s Azure has about 25%.
During their recent sale event, Amazon used 250,000 Graviton chips and 80,000 custom AI chips to manage high activity across its platforms. According to Adobe Analytics, the event resulted in a record $14.2 billion in sales.
