Amazon is deploying more of its own Trainium accelerators than Nvidia GPUs in AWS, aiming to reduce reliance on third-party chips and optimize performance for large language models.

AWS has a $225 billion backlog for Trainium compute capacity, signaling strong customer demand and securing multi-year AI infrastructure commitments.

Alphabet is similarly designing in-house AI accelerators to power its cloud services and AI projects, seeking to challenge Nvidia's dominance and lower hardware costs.