Amazon EC2 G7 instances with NVIDIA RTX PRO Blackwell GPUs launch in N. Virginia

AWS launched Amazon EC2 G7 instances in US East (N. Virginia), powered by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs. AWS claims up to 4.6x AI inference performance and 2.1x graphics performance over the previous-generation G6 instances, along with faster GPU-accelerated data analytics. The G-family targets mid-range inference, graphics, and virtual-workstation workloads rather than large-scale training, filling the gap between commodity CPU instances and the high-end P5/P6 training fleets.

The timing is notable given AWS simultaneously raised EC2 Capacity Block prices for its Nvidia P5 (Hopper) and P6 (Blackwell) training instances by roughly 20%, citing tight supply. The G7 launch on the more accessible RTX PRO Blackwell silicon gives customers a cheaper on-ramp to Blackwell-generation performance for inference-heavy workloads without competing for scarce top-tier training capacity.

The broader context is the compute-supply squeeze dominating the industry this week. NVIDIA is even renting back unsold GPU capacity to cloud operators, Meta's own chips hit production in September, and Amazon is planning a $25B+ AI fundraise. Against that backdrop, expanding Blackwell-class inference availability is both a customer-demand response and a way for AWS to monetize Nvidia allocation across a wider range of workloads. For developers, the 4.6x inference uplift matters most for cost-per-token economics — the same dollars-per-task lens driving this week's model launches. The practical question is regional availability and pricing beyond N. Virginia, since single-region launches often lag broader rollout by months.