Amazon is deploying over a gigawatt of new AI-dedicated data centers packed with its custom Trainium2 chips to anchor client Anthropic, betting that hardware-software co-design can overcome raw performance gaps with Nvidia. This partnership, central to AWS's cloud resurgence, leverages Trainium's memory bandwidth efficiency for Anthropic's reinforcement learning ambitions at unprecedented scale. The move signals a strategic pivot where hyperscalers and AI labs converge on custom silicon to escape the GPU scarcity trap.