The P6-B300 instances provide eight NVIDIA Blackwell Ultra GPUs with 2.1 TB high-bandwidth GPU memory, 6.4 Tbps Elastic Fabric Adapter networking, and 4 TB system memory. Compared to predecessors, these instances double networking bandwidth and increase GPU memory and compute performance by 50%, targeting training and deploying trillion-parameter foundation and language models.
Available in select regions including Seoul, Oregon, and Virginia, the P6-B300 instances enable faster AI workload training and inference with improved token throughput. While AWS notes these capabilities, actual performance gains depend on workload and optimization strategies employed by users.
Source & further reading
Amazon EC2 P6-B300 instances are now available in the Asia Pacific (Seoul) Region ↗Source published 2026-08-20. This report was prepared with AI assistance from the linked primary-source material. Claims and limitations should be checked against the original announcement or research.