Amazon EC2 G7 instances go GA with NVIDIA RTX PRO 4500 Blackwell GPUs

AWS made Amazon EC2 G7 instances generally available, powered by NVIDIA's RTX PRO 4500 Blackwell Server Edition GPUs. AWS benchmarks claim up to 4.6x AI inference performance and 2.1x graphics performance over the prior G6 generation, with the instances aimed at inference serving, graphics rendering, and analytics workloads rather than large-scale training.

The inference focus is strategically pointed: as a separate report noted, Nvidia is tightening its grip specifically on the AI inference market — the revenue-generating phase of running deployed models — and G7 is AWS reselling that capability to customers at scale. For teams running production generative AI, the generational jump matters because inference cost is the recurring bill that determines whether an AI feature is economically viable.

The launch arrived in a dense AWS Summit week alongside SageMaker AI inference observability (token performance, GPU health, autoscaling visibility on a CloudWatch dashboard), faster ECS auto-scaling via 20-second high-resolution metrics, and new JumpStart models including Mistral's Ministral-3-14B. Collectively these signal AWS doubling down on the unglamorous-but-critical production-inference layer.

The tension worth watching: AWS is simultaneously NVIDIA's largest customer (reselling Blackwell via G7) and a would-be rival, with Bloomberg reporting Amazon is in talks to sell its own custom AI chips to outside companies. G7's NVIDIA dependency underscores how far Amazon still is from displacing the GPU incumbent in the segment it most wants to win.