AWS expands Graviton4 and Blackwell GPU instances to new regions

Amazon Web Services has widened the geographic footprint of three EC2 instance families built for AI training, graphics workloads, and general-purpose compute, pushing custom silicon and NVIDIA's latest server GPUs closer to customers outside the United States.

The rollout touches I8g storage-optimized instances, G7 graphics instances, and C8g compute-optimized instances, all part of AWS's push to move workloads onto its own Graviton4 chips or NVIDIA's newest hardware rather than older generations.

Storage muscle for AI pre-processing

I8g instances, aimed squarely at AI large language model pre-processing and other storage-heavy training tasks, are now positioned to land in AWS Europe (Paris) and Asia Pacific (Jakarta). They pair Graviton4 processors with third-generation AWS Nitro SSDs, offering up to 1.5 TiB of memory and 45 TB of local NVMe storage across eleven sizes, two of them bare metal.

AWS says the instances cut storage I/O latency by up to 50% and lift real-time storage performance per terabyte by up to 65% compared with the previous-generation I4g line, while providing 100 Gbps of network bandwidth and 60 Gbps dedicated to Amazon EBS.

There's a wrinkle worth flagging: AWS's own announcement describes simultaneous I8g availability in Paris and Jakarta, but outside reporting only traces a related I8ge launch in Paris back to May 2026, paired with Asia-Pacific rollouts in Thailand, Hong Kong, Seoul and Tokyo, not Jakarta. No independent source confirms the Paris-and-Jakarta pairing as described.

Graphics compute reaches Spain, on paper

G7 instances, powered by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs, are listed as available in four regions: US East (N. Virginia and Ohio), US West (Oregon), and now Europe (Spain). AWS credits the line with up to 2.1x the graphics performance of G6 instances, alongside up to 192 vCPUs, 768 GiB of system memory, and 700 Gbps of EFA networking bandwidth, with up to eight GPUs carrying 32 GB of memory apiece.

That Spain listing doesn't fully square with earlier reporting, which had tracked G7 as generally available only in US regions as of mid-July 2026. No independent confirmation of the Spain launch has surfaced in the roughly two weeks since AWS's announcement, leaving the regional claim resting on AWS's own account for now.

Compute-optimized C8g spreads further

The C8g family, also built on Graviton4 and the AWS Nitro System, is expanding to Europe (Paris), Africa (Cape Town), Israel (Tel Aviv), and Canada West (Calgary) starting August 5, 2026. AWS pitches these instances for HPC, batch processing, video encoding, distributed analytics, and CPU-based ML inference.

Graviton4 itself is the headline act here: AWS claims up to 40% faster database performance, 30% faster web application performance, and 45% faster large Java application performance than Graviton3, translating into up to 30% better overall performance and 3x more vCPUs and memory versus the Graviton3-based C7g line.