Amazon Provides Free Access to Custom AI Chips for Researchers

Amazon.com’s cloud computing division, Amazon Web Services (AWS), announced on Tuesday that it will provide free computing resources to researchers interested in using its custom AI chips, aiming to compete with Nvidia’s popularity among researchers in this field.

AWS said it will offer $110 million in credits for its cloud data centres to researchers working with its Trainium chip, designed for AI model development and competing with chips from Nvidia, Advanced Micro Devices, and Alphabet’s cloud division.

Researchers from Carnegie Mellon University and the University of California, Berkeley, are already participating in the programme, with AWS planning to allocate 40,000 of its first-generation Trainium chips.

This initiative comes as AWS, still the leading cloud computing provider by sales faces increased competition from Microsoft as software developers explore new chip types for AI applications.

Also Read: Amazon abandons acquisition of iRobot

AWS aims to attract attention to its AI chips by taking a different approach from Nvidia. Instead of using Nvidia’s Cuda software, most commonly used to program Nvidia’s chips, AWS will publish documentation on the core instruction set architecture of Trainium, allowing direct programming of the chip.

According to Gadi Hutt, AWS’s head of AI chip business development, this approach targets large clients who may benefit from minor adjustments that could lead to significant performance and cost benefits when using tens of thousands of chips at scale.

“For those investing hundreds of millions of dollars in rented computing power, any performance increase and cost reduction is highly valued,” Hutt explained.