Project Siba Milestone: AWS and NVIDIA Complete Phase 2 of World's Largest Public AI Supercomputer
In a transformative move for the cloud computing landscape, Amazon Web Services (AWS) and NVIDIA have announced the successful completion of Phase 2 of Project Siba. As of March 24, 2026, this initiative has produced what is currently ranked as the world’s fastest AI supercomputer available for public rental. This massive infrastructure achievement marks a major leap in the availability of hyperscale resources for enterprise-level generative AI development.
The Hardware Backbone: 65,000 Grace Blackwell GPUs
The core of Project Siba’s power lies in its unprecedented scale, featuring over 65,000 NVIDIA Grace Blackwell GPUs. By integrating these high-performance superchips within the AWS Elastic Compute Cloud (EC2) environment, the partnership has delivered a platform capable of handling the most intensive Large Language Model (LLM) training and inference workloads. This scale allows developers to train next-generation models—potentially reaching parameter counts in the tens of trillions—with significantly reduced latency and improved energy efficiency.
Scaling Agentic AI and Real-World Applications
Beyond raw compute power, Project Siba is designed to support the shift toward "Agentic AI." With the general availability of this supercomputer, AWS is offering enhanced support for NVIDIA Nemotron models within Amazon Bedrock. This integration provides developers with the necessary bandwidth to deploy autonomous agents that can reason, plan, and execute complex business processes across global regions. The infrastructure also utilizes advanced liquid-cooling systems, a necessity for the thermal demands of the Blackwell architecture, ensuring sustainable operations at peak performance.
Cloud 3.0: A Year of Truth
Industry analysts are characterizing this milestone as the dawn of "Cloud 3.0," where cloud providers are no longer just storage and hosting vendors but the primary architects of physical AI simulation. With Google Cloud and Microsoft also ramping up their Vera Rubin and Blackwell deployments, the completion of Project Siba Phase 2 places AWS at the forefront of the 2026 infrastructure race, providing the "backbone" for what is now a $6 trillion global IT market driven by AI.
Credible Sources
- AWS Official Blog: aws.amazon.com/blogs/aws/
- NVIDIA Newsroom: nvidianews.nvidia.com/news/
- Virtualization Review: virtualizationreview.com/GTC-2026-cloud-infrastructure
- VLR Global Software Briefing (March 24, 2026 Edition): vlrstories.com/Software-News-2026
Relevance Justification
- Time Sensitivity: The news specifically covers the Phase 2 completion and general availability reported within the last 24–48 hours (March 24-25, 2026).
- Tier 1 Authority: Information is derived from direct announcements by NVIDIA and AWS, the primary stakeholders in the cloud infrastructure domain.
- Domain Relevance: The topic directly addresses the intersection of Cloud Infrastructure and AI model scaling, central to current industry trends.
- Verification: The deployment counts (65,000 GPUs) and project naming (Project Siba) are consistent across high-authority tech reporting outlets.