On November 12, 2024, Amazon announced a $110 million investment to support university-led research in generative AI. Through the “Build on Trainium” program, Amazon will provide compute hours to enable researchers to develop new AI architectures, machine learning libraries, and performance optimizations using AWS Trainium UltraClusters—large-scale collections of AI accelerators built for complex computational tasks.
AWS Trainium, Amazon’s custom-built chip for deep learning training and inference, will serve as the foundation of this initiative, with all research outcomes to be open-sourced, allowing for continued innovation across the AI community.
The program supports a broad range of research, from improving AI accelerator performance to large distributed systems studies. AWS created a dedicated Trainium UltraCluster with up to 40,000 Trainium chips, designed to handle the complex needs of AI research at scale. Amazon also plans to provide funding for research projects and education programs, as well as AWS Trainium credits and access to the UltraClusters for selected proposals.
Also Read: Oracle to Invest Over $6.5B in AI and Cloud Computing in Malaysia
With computing resources in high demand and often limited by budget constraints, Build on Trainium offers universities the opportunity to carry out high-performance research without compromise. Carnegie Mellon University (CMU) is one of the institutions participating in the program. “The Build on Trainium initiative allows our faculty and students large-scale access to AWS Trainium with an open programming model, significantly expanding our research capacity,” said Todd C. Mowry, CMU professor of computer science.
Amazon is also facilitating training for future AI experts by providing hands-on access to advanced computing hardware like Trainium. According to Christopher Fletcher, an associate professor at UC Berkeley, Trainium’s architecture is uniquely adaptable for research, offering low-level access for optimization. This flexibility is further enhanced by AWS’s Neuron Kernel Interface (NKI), a programming interface that allows researchers direct access to the chip’s instructions to build optimized compute kernels for new model operations.
As part of the program, grant recipients will connect with AWS’s Neuron Data Science community and other resources to advance their work. Amazon aims to foster open collaboration, encouraging researchers to publish their findings and contribute to open-source machine learning libraries, laying the groundwork for future AI advancements.
