AWS grants $110M in Trainium credits to 34 universities for Responsible AI research

Amazon Web Services announced 34 recipients from 30 universities for its Build on Trainium program, a $110 million compute credit initiative supporting novel AI research on AWS's purpose-built Trainium chips. The Fall 2025 call for proposals centered on Responsible AI, inviting work on AI safety and alignment, multilingual language models, representation engineering, sustainability, small language models, and synthetic data generation. Each project was selected based on scientific quality and potential societal impact. Recipients include researchers from UIUC, UCLA, CMU, MIT, Georgia Tech, UCSD, and University of Washington, among others.

Awardees gain access to more than 700 Amazon public datasets, AWS promotional credits for AI/ML services, and dedicated Amazon research contacts for consultation. They also benefit from Trainium-specific resources like tutorials and hands-on sessions. Notable projects include UIUC's topology-aware parallelization for trillion-parameter mixture-of-experts models running on up to 1,024 Trainium chips, and University of Washington's inference-optimization framework to improve token efficiency for portable, high-performance LLM inference. Other work targets differentially private synthetic data, machine unlearning, multimodal scam detection, and efficient sparse/quantized LLMs. This effort underscores AWS's commitment to democratizing AI research by removing compute barriers for academia.