AWS takes aim at Nvidia with Graviton4 and Trainium chips

Amazon’s AWS (Amazon Web Services) is set to challenge Nvidia’s market dominance in the AI chip sector as it pushes into its custom chip strategy through Graviton4 CPUs and Trainium series accelerators. The custom chips are engineered to maximize profit margins in AI workloads by slashing data transfer costs in cloud environments.

AWS announced an update to its Graviton4 chip that includes 600 gigabits per second of network bandwidth, which the company calls the highest offering in the public cloud. Ali Saidi, an engineer at AWS, likened the speed to a machine reading 100 music CDs a second. The Graviton4 CPU is one of many chip products from Amazon’s Annapurna Labs in Austin, Texas, and it is a win for the company’s custom strategy, putting it up against traditional players like Intel and AMD.

With Graviton4′s upgrade on the horizon and Project Rainier’s Trainium chips, Amazon demonstrated its ambition to control the entire AI infrastructure stack, from networking to training to inference. As more major AI models like Claude 4 proved they could train successfully on non-NVIDIA hardware, the question was not whether AWS could compete with the chip giant but how much market share it could take.

Hutt says Amazon will provide cheaper alternatives to Nvidia’s GPUs

AWS Senior Director for Customer and Product Engineering Gadi Hutt said Amazon wanted to reduce AI training costs and provide alternatives to Nvidia’s expensive graphics processing units (GPUs). According to AWS, Anthropic’s Claude Opus 4 AI model launched on Trainium2 GPUs, and Project Rainier is powered by over 500K chips—an order that would have generally gone to Nvidia. 

Hutt said that while Nvidia’s Blackwell was a higher-performing chip than Trainium2, the AWS chip offered better cost performance. He also pointed out that Trainium3 was coming up this year, and it was doubling the performance of Trainium2, and it would save energy by an additional 50%. Rami Sinno, Director of Engineering at AWS’s Annapurna Labs, said demand for these chips was already outpacing supply.

“Our supply is very, very large, but every single service that we build has a customer attached to it.”

–Rami Sinno, Director of Engineering at AWS’s Annapurna Labs

The AWS team stressed that while the company acknowledged specific gaps, it preferred to work with smaller, innovative startups such as Anthropic, Scale AI, and Fiddler rather than relying on large vendors. Amazon often supports these companies through strategic investments, forming mutually beneficial relationships in return for providing capital or infrastructure, as in the case of Anthropic. AWS announced Project Rainier—an AI supercomputer built for startup Anthropic—at the Invent 2024 conference last December and reportedly put $8B into backing Anthropic.

AWS’s Graviton4 and Trainium3 chips promise improved performance and efficiency 

The Amazon team disclosed that the upgraded Graviton4 and Trainium3 chips set for late 2025 promised a 4x performance leap and 40% better energy efficiency, further compressing Nvidia’s margins. It added that this was not just a win for AWS’s top-line growth but a direct assault on Nvidia’s GPU premium.

Rahul Kulkarni, Amazon’s Director of Product Management for Compute and AI, said the upgraded Graviton4 promised to deliver three times its predecessor’s compute power and memory, 75% more memory bandwidth, and 30% better performance. He added that collectively, it was expected to deliver more price performance, meaning that users got a lot more performance for every dollar spent.

Patrick Moorhead, the CEO and Chief Analyst at Moor Insights & Strategy, said all AI companies were spending a lot of money on developing chips, adding that they had giant R&D budgets despite not disclosing the exact figures of how much was being invested. Moorhead, who spent over a decade as a vice president at AMD, added that Nvidia still remained a dominant player in the AI chips market. However, there was enough demand to support multiple competitors, including AWS.

Cryptopolitan Academy: Coming Soon - A New Way to Earn Passive Income with DeFi in 2025. Learn More