The project uses nearly 500,000 AWS Trainium2 chips, making it the largest Trainium2-based deployment assembled to date.