Amazon.com, Inc. (AMZN) shares opened under pressure as Amazon Web Services (AWS) announced an expanded partnership with Nvidia covering millions of GPUs, even as the company continues to develop its own internal AI chips. AMZN traded at $256.32, down 1.52%, after falling from the $259 to $260 range. The stock tested the $256 support level while selling pressure continued.

Amazon Expands Nvidia GPU Deployment

AWS plans to deploy 2 million additional Nvidia GPUs between 2027 and 2028. The deployment will include Blackwell Ultra, Rubin, and Rubin Ultra systems across AWS infrastructure. The plan follows Amazon's commitment to install 1 million Nvidia chips starting in 2026. The commitments were announced at Nvidia's GTC conference in Washington, D.C., where Nvidia disclosed more than $500 billion in Blackwell and Rubin GPU bookings from U.S. companies, including AWS, Microsoft, Meta, and Oracle.

Taken together, the two commitments would give AWS access to millions of Nvidia GPUs spanning multiple hardware generations. The company continues its Nvidia relationship, and AWS keeps adding Nvidia systems to its infrastructure. The expansion reflects AWS's continued use of Nvidia hardware for computing workloads as demand for cloud computing capacity continues to grow, and the expanded agreement places Nvidia hardware at the center of AWS capacity growth. Rival clouds are moving in the same direction: Microsoft Azure and Google Cloud also purchase Nvidia GPUs at scale as the largest cloud providers compete to add capacity for AI training and inference. Alongside its Nvidia purchases, Amazon continues developing its own processors through Annapurna Labs.

AWS Broadens the Nvidia Collaboration

The GPU purchases are only one part of the deepened relationship. AWS and Nvidia also plan to build dedicated AI factories for the U.S. government using 100,000 GPUs. Those facilities will use Nvidia hardware and are intended to support government computing requirements. The plan aligns with the White House's 2025 AI Action Plan, which called for expanding federal AI infrastructure and computing.

AWS will additionally add Nvidia's Nemotron models to its Bedrock and SageMaker services, and Amazon Robotics will use Nvidia's physical AI platform for warehouse automation — a step that extends the partnership into robotics as AWS expands its use of Nvidia computing systems.

Collectively, the projects cover cloud services, government infrastructure, software tools, and warehouse operations. The partnership extends beyond GPU purchases to include software and physical computing applications, with Nvidia providing hardware and technology while AWS integrates those capabilities across several Amazon services. The companies are broadening their relationship as AWS increases computing capacity.

Amazon Maintains Its Own Chip Strategy

Despite the larger Nvidia GPU commitment, Amazon continues developing its Trainium processors through Annapurna Labs. Amazon acquired the Israel-based chip design firm in 2015, and the unit has since produced the Graviton processors that run many AWS cloud servers. Amazon has invested in Trainium as part of its effort to develop internal hardware, and AWS said Annapurna Labs will work with Nvidia to extend NVLink Fusion to Trainium chips. The work will also add high-bandwidth memory to those Trainium systems. Trainium already anchors a major AWS build-out: Project Rainier, the compute cluster created with Anthropic using hundreds of thousands of Trainium2 chips, went live in 2025 to train and run Anthropic's models.

The approach lets Amazon combine its internally designed processors with Nvidia connectivity technology, while AWS continues deploying Nvidia GPUs for workloads that require large computing capacity. The strategy places both Nvidia hardware and Amazon-designed chips within AWS's infrastructure plans, giving the company two sources of AI compute capacity as it scales.

Amazon's latest plans combine external GPU supply with continued internal chip development. The company will add 2 million Nvidia GPUs while advancing Trainium technology through Annapurna Labs. As a result, AWS is expanding its infrastructure while maintaining its own processor technology. Because the Nvidia deliveries are scheduled from 2026 into 2028, the added capacity arrives in stages rather than all at once, and how workloads divide between Nvidia GPUs and Trainium chips will become clearer as the systems come online.

Source: Blockonomi