AWS chips challenge Nvidia with new academic initiative

AWS is enticing researchers with free access to its Trainium AI chips, offering detailed programming capabilities to help optimise performance and cut costs on a massive scale.

AWS, Amazon, Nvidia, AI chips, Trainium

Amazon Web Services (AWS) is offering $110 million in free computing power to researchers to promote its custom AI chips. The programme provides credits for the use of AWS’s Trainium chips, which are designed to compete with Nvidia’s widely-used hardware, as well as Advanced Micro Devices and Alphabet’s cloud technology. Researchers from Carnegie Mellon University and the University of California, Berkeley, are already participating, with AWS planning to make 40,000 Trainium chips available.

AWS, the world’s leading cloud provider by sales, is facing intensified competition from Microsoft, especially as the demand for cutting-edge AI hardware grows. The company is taking a novel approach to lure AI developers by offering detailed documentation for Trainium’s instruction set architecture. This will allow researchers to program the chip directly, unlike Nvidia‘s chips, which usually require the use of proprietary Cuda software.

Gadi Hutt, head of business development for AI chips at AWS, said this strategy is aimed at customers with large-scale operations. Even minor programming adjustments could yield significant performance and cost advantages when using tens of thousands of chips. Hutt emphasised that companies investing hundreds of millions in computing infrastructure would welcome opportunities to improve efficiency and reduce expenses.