- AWS urging customers to go from Nvidia to its cheaper
- He says that his equipment offers the same performance with a cost saving of 25%
- Amazon’s argument occurred while Nvidia presented its new equipment in the GTC 2025
While Nvidia held his annual GTC 2025 conference, showing new products such as DGX Spark And DGX station AI superordinators, Amazon tried to convince its Client Clients that they could save money by moving away from Nvidia dear equipment and adopting Amazon’s own ia chips.
Information AWS complaints have presented at least one of its cloud customers to consider renting servers powered by AmazonChip Trainium, saying that they could take advantage of the same performances as the H100 of Nvidia, but at 25% of the cost.
Trainium is one of the many internal chips that Amazon has developed (alongside Graviton and Inferentia), built for the formation of automatic learning models in the AWS cloud and offering an alternative to lower cost to GPU -based systems. Amazon silicon is not intended for a replacement similar to the more advanced products of Nvidia, but it does not need to be.
Part of the AI conversation
Amazon’s offer seems to be part of a wider change in the cloud market, where suppliers like AWS and Google Develop their own chips and offer them to customers as a means of avoiding the cost – and rarity – highly sought -after GPUs from Nvidia.
“What AWS does is intelligent,” said Matt Kimball, vice-president and main analyst for calculation and storage data at Moor Insights & Strategy Networkworld. “That said to the world that there is a profitable alternative which is also efficient for AI training needs. He fits into AI conversation. “
The land here, of course, is access. AWS gives customers the possibility of experimenting with the training and the deduction of workloads without having to wait for months for an NVIDIA GPU or to pay the best dollar for this.
Although a 25% saving is certainly not to sniff, and something that will undoubtedly appeal to a number of AWS customers, there are obvious drawbacks to consider.
As Networkworld Note: “Companies accustomed to working with the unified system architecture (CUDA) of NVIDIA must reflect on the cost of transition to a whole new platform as a trainium. In addition, trainium is only available on AWS, so that users can lock themselves. ”