Amazon is developing new AI chips to outperform and be cheaper than NVIDIA’s current offerings. This move is part of Amazon’s strategy to reduce its dependence on NVIDIA’s expensive chips, which currently power many artificial intelligence (AI) services in Amazon Web Services (AWS), its primary growth driver.
A team of engineers has been working on a new server design that integrates these new AI chips in Amazon’s chip lab in Austin, Texas. During a recent visit to the lab, Amazon executive Rami Sinno showcased the server, highlighting how it competes directly with NVIDIA’s market-leading chips.
Amazon’s drive to develop its own processors aims to provide more cost-effective solutions for customers who need to perform complex calculations and process large volumes of data. By producing its own chips, Amazon aims to mitigate the so-called “NVIDIA tax,” which refers to the high cost associated with NVIDIA’s AI chips.
Amazon’s efforts are not unique; other tech giants like Microsoft and Alphabet are also developing AI chips to reduce costs and improve performance. Amazon’s Annapurna Labs, acquired in 2015, oversees this area. According to Sinno, the director of engineering at Annapurna Labs, there is a growing demand among Amazon’s customers for cheaper alternatives to NVIDIA’s chips.
Though Amazon’s AI chip initiative is still in its early stages, the company has a solid foundation in chip development with its Graviton processors, which handle non-AI computing tasks. The Graviton chip, now in its fourth generation, has been in development for nearly a decade.
On the AI front, Amazon’s new chip designs, Trainium and Inferentia, are poised to offer significant improvements in price and performance.
According to David Brown, Vice President of Compute and Networking at AWS, these new chips can offer up to 50 percent better price performance in some cases, potentially making them half as expensive as running similar models on NVIDIA hardware.
AWS, which accounts for nearly 20 percent of Amazon’s total revenue, saw its sales rise by 17 percent to $25 billion in the first quarter compared to the previous year. AWS controls about a third of the cloud computing market, with Microsoft’s Azure around 25 percent.
During Amazon’s recent Prime Day event, the company deployed many custom chips to handle the increased demand. Specifically, Amazon used a quarter of a million Graviton chips and 80,000 of its custom AI chips to manage the surge in activity across its platforms. According to Adobe Analytics, this event generated record sales of $14.2 billion.
Amazon’s foray into AI chip development is a strategic move to enhance its cloud services and offer more competitive pricing to its customers. By reducing its reliance on NVIDIA and developing its high-performance, cost-effective AI chips, Amazon aims to maintain its edge in the cloud computing market and drive growth in its AWS division.
This development positions Amazon as a formidable player in the AI chip market. It highlights the broader industry trend of tech giants investing heavily in custom chip development to optimize performance and reduce costs.