Amazon Web Services and Nvidia said Aug. 26 that AWS plans to deploy 2 million additional Nvidia graphics processing units across its global infrastructure in 2027 and 2028, a major expansion of the computing capacity available to AI developers, enterprises and government customers.
The new commitment follows AWS’ March plan to add more than 1 million Nvidia GPUs beginning this year. Together, the programs put more than 3 million Nvidia accelerators on AWS’ deployment roadmap while extending the companies’ relationship beyond chips into networking, CPUs, software, open models and robotics.
AWS expands its Nvidia capacity for 2027 and 2028
The additional fleet will include Nvidia’s Blackwell Ultra, Rubin and Rubin Ultra GPUs, according to the companies’ joint announcement. AWS said demand for the capacity announced in March had exceeded expectations. The companies did not disclose financial terms or provide a detailed regional deployment schedule.
The agreement also calls for AWS to introduce infrastructure based on Nvidia’s Vera central processing units and to connect its custom Trainium accelerators with Nvidia’s NVLink Fusion technology. Nvidia’s Nemotron open models will remain available through Amazon Bedrock and SageMaker, while Nvidia software will be integrated into Amazon’s data-processing, search and robotics systems.
TechCrunch reported that the expansion was announced alongside Nvidia’s quarterly results. Nvidia said in its quarterly report that fiscal second-quarter revenue reached $96.2 billion, up 106% from a year earlier, while data-center revenue rose 117% to $89 billion. The chipmaker forecast $108 billion in revenue for the current quarter.
The deal widens AWS’ infrastructure strategy
The scale of the Nvidia deployment does not mean Amazon is abandoning its own silicon. AWS continues to develop Trainium chips for AI workloads and Graviton CPUs for general-purpose computing. Instead, the agreement shows a dual strategy: expand access to Nvidia hardware that customers already demand while using Amazon-designed processors to compete on price, performance and supply flexibility.
That balance matters for enterprise buyers planning multiyear AI investments. More GPU capacity can reduce access constraints, but deeper concentration around one supplier increases exposure to Nvidia’s product cycle, pricing and manufacturing network. AWS’ own chips give Amazon an alternative, while the Nvidia partnership helps it serve customers that have standardized on Nvidia’s software and hardware ecosystem.
The companies also plan to build AI infrastructure for the U.S. government, including secure AWS systems using 100,000 Nvidia GPUs for federal and national-security workloads. Amazon Robotics will adopt Nvidia’s physical-AI platform, connecting the cloud agreement to warehouse automation and next-generation robots.
Nvidia’s record quarter underscores the demand
Nvidia’s results provide financial context for the deal. The company’s revenue more than doubled from a year earlier, and its data-center business accounted for most of total sales. Reuters reported that Nvidia’s third-quarter forecast exceeded Wall Street estimates, signaling that demand for AI infrastructure remains strong despite concerns about the pace and cost of data-center investment.
For AWS, the commercial test will be turning that capital-intensive expansion into durable cloud demand. The announced deployment runs through 2028, giving enterprises a clearer view of future capacity but leaving questions about pricing, availability and utilization unanswered.
The strategic message is already clear: Amazon is willing to deepen its reliance on Nvidia even as it builds competing chips. For executives, that combination suggests the AI infrastructure market is moving toward multiple processor options inside the same cloud rather than a clean shift away from Nvidia’s platform.