AWS brings NVIDIA Blackwell GPUs to GovCloud for advanced AI training
Amazon has made its EC2 P6-B200 instances, powered by NVIDIA Blackwell GPUs, available in AWS GovCloud (US-East). These instances offer a significant performance boost for AI training and inference, featuring 8 Blackwell GPUs and 1440 GB of high-bandwidth GPU memory. They are designed to scale AI workloads securely within Amazon EC2 UltraClusters.
Key Takeaways
- P6-B200 instances feature 8 NVIDIA Blackwell GPUs and 1440 GB of high-bandwidth memory (HBM3e).
- GPU memory bandwidth is increased by 60% compared to the 5th Gen (P5en) instances.
- Networking performance reaches 3.2 terabits per second via Elastic Fabric Adapter (EFAv4) on the Nitro System.
- Compute is supported by 5th Generation Intel Xeon Processors (Emerald Rapids) for enhanced density.
- Availability expanded to US-East (N. Virginia, Ohio), US-West (Oregon), and both GovCloud US regions.
Why It Matters
The introduction of Blackwell architecture into GovCloud provides federal agencies and highly regulated streaming providers with the compute density required for complex multimodal AI and real-time video analysis. By doubling performance over the H100-based P5en generation, AWS addresses the massive data processing needs of public sector media archives and security-sensitive streaming environments. This capability allows for faster model iteration in environments that require strict compliance with FedRAMP and ITAR regulations. In a competitive landscape, this move reinforces AWS's lead in the government sector as peers Microsoft and Google also scale their respective specialized sovereign cloud clusters. Industry observers should watch for federal adoption rates of the p6-b200.48xlarge size for real-time generative metadata tagging.
Additional Context
The rollout in GovCloud follows a broader capital commitment by Amazon to public sector artificial intelligence. In June 2026, AWS announced an investment of approximately $50 billion to expand its government-specific infrastructure across Top Secret, Secret, and GovCloud regions. This initiative aims to add 1.3 gigawatts of compute capacity to support the sharp increase in federal AI use cases, which reports from Nextgov suggest grew by 105% between 2025 and 2026. This infrastructure serves nearly 7,500 government agencies, supporting missions ranging from fraud detection to cybersecurity and public media accessibility. Technically, the transition to the Blackwell architecture is essential for handling the scale of the most recent foundation models. Independent benchmarks from MLPerf in April 2026 indicate that the B200 GPU provides roughly 2.9 times the throughput of the preceding H200 systems for large-scale inference workloads. This efficiency is critical as inference has overtaken training to account for two-thirds of all AI compute demand in 2026, per Spheron reporting. For video streaming specifically, this shift supports the production-level deployment of real-time translation and background noise removal, as seen in recent implementations by public broadcasters like Austria’s ORF. These agencies use AI to repurpose broadcast notes into radio scripts and social media content, a process that requires the high throughput and low-latency networking provided by the EFAv4 and Nitro-powered P6 instances.
Read full article at aws.amazon.com
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source