NVIDIA and AWS Launch EC2 G7 Instances with Blackwell GPUs and GPU-Accelerated Vector Search
Summary
- • AWS launches EC2 G7 instances powered by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs, delivering 4.6x AI inference and 2.1x graphics performance over prior G6 instances
- • NVIDIA cuVS library now makes GPU-accelerated vector indexing the default in Amazon OpenSearch Serverless — 10x faster at 25% of CPU-only cost
- • AWS earns NVIDIA Exemplar Cloud status for the GB300, certifying peak-optimized training performance for frontier model workloads
- • G7 instances support up to 8 GPUs, 256GB GPU memory, 700 Gbps EFA networking, and 7.6TB local NVMe storage across multiple configurations including bare metal (coming soon)
Details
EC2 G7: NVIDIA RTX PRO 4500 Blackwell GPUs
New AWS instance type built on NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs; targets AI inference, graphics, spatial computing, video workflows, and GPU-accelerated data analytics workloads.
4.6x AI inference, 2.1x graphics vs G6
G7 delivers up to 4.6x AI inference performance and up to 2.1x graphics performance compared with prior G6 instances; also brings faster GPU-accelerated data analytics on Amazon EMR.
Up to 8 GPUs, 256GB memory, 700 Gbps EFA, 7.6TB NVMe
Available in 1-, 2-, 4-, and 8-GPU configurations plus bare metal (coming soon); accessible via Deep Learning AMIs, EMR, EKS, ECS, and SageMaker AI (coming soon).
cuVS: GPU vector search as OpenSearch Serverless default
NVIDIA cuVS library makes GPU-powered vector indexing the default compute choice for all Amazon OpenSearch Serverless vector collections — removing opt-in friction for RAG, semantic search, recommendation, and agentic AI applications.
10x faster vector indexing at 25% of CPU-only cost
cuVS-powered GPU vector indexing delivers 10x speed improvement at a quarter of the cost compared to CPU-only OpenSearch builds, according to NVIDIA.
AWS earns NVIDIA Exemplar Cloud status for GB300
AWS achieves NVIDIA's Exemplar Cloud certification for the GB300, guaranteeing customers peak-optimized performance for training workloads and positioning AWS as a premier destination for large-scale model training.
Product Launch = new instance types and capabilities; Stat = performance benchmarks; Infrastructure = hardware specs and configurations; New Tech = new default behaviors changing how services work; Partnership = certifications and joint designations.
What This Means
NVIDIA and AWS's joint announcements mark a significant expansion of GPU-native infrastructure across multiple layers of the AI production stack. By bringing Blackwell GPUs to general-purpose EC2 instances and defaulting to GPU-accelerated vector search in OpenSearch Serverless, the partnership removes the need for enterprises to specially engineer GPU access into their AI pipelines. The GB300 Exemplar Cloud designation further cements AWS as a leading destination for frontier model training. Together, these moves accelerate the normalization of GPU-native infrastructure as the standard — not the exception — for enterprise AI production workloads, and they deepen the NVIDIA-AWS partnership at a time when cloud providers are competing intensely for AI workload share.