AWS and NVIDIA Announce Deployment of 2 Million Next-Gen GPUs in Massive Cloud Expansion

Amazon Web Services and NVIDIA are expanding their infrastructure partnership by deploying 2 million additional next-generation GPUs across AWS data centers in 2027 and 2028. This triples Amazon's previous commitment and broadens the collaboration to include CPUs, robotics, and secure government facilities.[1][2][3]

The scale of the deployment reflects a shift in the artificial intelligence industry. As frontier labs and enterprises move from pilot programs to production-scale systems, the demand for compute power has consistently outpaced initial forecasts.[2][5]

The hardware mechanism driving this expansion centers on NVIDIA's upcoming architectures. The 2 million new GPUs will primarily consist of the Blackwell Ultra, Rubin, and Rubin Ultra platforms, which represent the next generation of accelerated computing.[4][6]

These processors are designed to handle the massive data throughput required for agentic AI—systems that can plan and execute multi-step tasks autonomously—and physical AI, which powers robotics and autonomous machines interacting with the real world.[2][6]

Beyond graphics processing units, the agreement introduces NVIDIA's Vera CPUs to the AWS ecosystem. Vera is purpose-built for AI agent workloads, offering an alternative for tasks that require high-performance central processing alongside accelerated infrastructure.[1][2]

The integration extends deeply into the networking layer. The companies are combining NVIDIA's NVLink Fusion technology with custom high-bandwidth memory (NVHBM) to allow GPUs and Amazon's own custom chips to communicate more efficiently within a single rack-scale architecture.[2][5]

A significant portion of the new capacity is earmarked for public sector use. AWS and NVIDIA plan to construct dedicated AI factories for the United States government, deploying 100,000 GPUs on highly secure infrastructure.[2][6]

These government facilities are designed to handle workloads classified at Impact Level 6 (IL6) and above, which includes highly sensitive national security data. This move addresses the growing need for sovereign AI capabilities that remain strictly isolated from public cloud networks.[2][5]

In the realm of physical AI, the partnership brings NVIDIA's robotics stack—including the Omniverse, Isaac, and Jetson platforms—directly into Amazon Robotics.[2][6]

This integration allows Amazon to simulate, train, and deploy next-generation warehouse robots in virtual environments before they ever touch a physical factory floor, accelerating the development of automated logistics.[5]

On the software side, AWS will expand support for NVIDIA's Nemotron family of open models, making them available as fully managed services on Amazon Bedrock and SageMaker to give developers more choices in foundation models.[2]

The expansion also highlights a dual-track silicon strategy at Amazon. While committing heavily to NVIDIA hardware, AWS continues to scale its own custom AI chips, Trainium and Graviton.[3][4]

Amazon's custom chip business has reportedly crossed a $25 billion annualized revenue run rate, backed by massive financial commitments from leading AI labs like Anthropic.[1][3]

By investing in both NVIDIA's ecosystem and its own proprietary silicon, Amazon is hedging its bets, ensuring it can meet customer demand regardless of which hardware architecture ultimately dominates the market.[3]

What remains unproven is the long-term economic viability of these massive infrastructure investments. The industry is currently testing whether the returns generated by AI applications can justify the tens of billions of dollars spent on data centers.[4]

While some executives claim that customers are seeing rapid returns on invested capital, the broader market is still waiting to see if enterprise adoption will scale fast enough to absorb the incoming wave of compute capacity in 2027 and 2028.[4][5]