AWS to Deploy 2 Million NVIDIA GPUs by 2028
The cloud giant doubles down on AI infrastructure with massive GPU expansion, new CPU options, and secure government AI factories.
Amazon Web Services is committing to a massive expansion of its AI infrastructure partnership with NVIDIA, planning to deploy 2 million additional GPUs across its global data centers during 2027 and 2028. The move represents a significant escalation from the 1 million GPU deployment AWS announced at NVIDIA GTC 2026, reflecting demand that has exceeded initial forecasts.
The expanded collaboration between the two companies spans multiple technical fronts beyond raw compute capacity. AWS will integrate NVIDIA's Vera CPUs into its cloud offerings, provide secure AI infrastructure for federal agencies, and connect its own Trainium chips with NVIDIA's NVLink Fusion interconnect technology.
Why it matters
This partnership expansion signals the scale of enterprise AI adoption and the infrastructure required to support it. By combining NVIDIA's GPU leadership with AWS's cloud reach, the companies are positioning themselves to capture demand from organizations moving AI workloads from experimentation to production. The federal government component addresses a critical gap in secure, high-performance AI infrastructure for national security applications.
New CPU options for agentic AI
AWS plans to offer NVIDIA Vera CPU-based infrastructure specifically designed for agentic AI workloads. These processors handle the CPU-intensive tasks that support AI agents, including code execution, tool use, sandboxing, analytics, and orchestration. Vera can function both as a host CPU for accelerated systems and as standalone compute for AI factory operations, keeping GPU resources fully utilized while maintaining responsive agent performance.
The addition of Vera CPUs gives AWS customers another configuration option alongside NVIDIA GPUs and AWS's own Trainium chips, reinforcing the cloud provider's strategy of offering diverse compute choices rather than a single solution.
Secure AI infrastructure for government
AWS and NVIDIA will construct AI factories specifically for U.S. government agencies, deploying NVIDIA's complete AI stack including 100,000 GPUs on AWS infrastructure certified for Impact Level 6 and higher classifications. These security designations cover federal and national security workloads requiring the highest levels of protection.
The government-focused infrastructure addresses growing demand from agencies seeking to deploy AI capabilities while maintaining strict security and compliance requirements.
Trainium integration advances
In a notable technical development, AWS announced that its next-generation Trainium chips will support NVIDIA's NVLink Fusion interconnect technology. Amazon's Annapurna Labs will also work with NVIDIA's custom high-bandwidth memory technology, potentially giving Trainium chips access to faster, more power-efficient memory options.
This integration allows Trainium and NVIDIA GPUs to operate within a unified rack-scale architecture, providing customers with flexibility in how they configure AI training and inference workloads.
Existing integrations
The companies highlighted several technical integrations already delivering results for customers. GPU-accelerated data processing on Amazon EMR with NVIDIA cuDF provides up to 3.7 times faster processing speeds for Apache Spark workloads compared to CPU configurations. GPU-accelerated vector indexing on Amazon OpenSearch Service delivers up to nine times faster indexing at one-quarter the cost.
Amazon Robotics is integrating NVIDIA's physical AI platform, including Jetson, Omniverse, and Isaac technologies, to advance warehouse automation through simulation and synthetic data generation.
The details were first reported by AWS in an official announcement about the expanded partnership.
This is an original analysis by the Omega editorial team. Source reporting: The Verge.
Want systems like this working for your business?
Book a Call