Amazon has significantly deepened its strategic partnership with Nvidia, announcing a massive expansion in its procurement of high-performance graphics processing units (GPUs). The cloud giant has committed to integrating an additional 2 million Nvidia GPUs into its Amazon Web Services (AWS) data centers, with deployments scheduled for 2027 and 2028. This move effectively triples the company’s previous order, reflecting a rapid escalation in the infrastructure requirements needed to power the next generation of artificial intelligence models.

The expanded agreement encompasses Nvidia’s most advanced hardware, including the Blackwell Ultra, Rubin, and Rubin Ultra GPU architectures. Beyond raw processing power, the collaboration will see the integration of Nvidia’s comprehensive software and networking ecosystem across AWS. This includes the deployment of Vera CPUs, robotics platforms like Isaac and Omniverse, and the hosting of Nvidia’s Nemotron open models on Amazon Bedrock and SageMaker. The decision to scale up is driven by surging demand from a diverse array of clients, ranging from emerging AI startups and research labs to large-scale enterprise organizations and government entities.

This development is particularly noteworthy as it occurs alongside Amazon’s ongoing efforts to develop its own proprietary silicon, such as the Trainium AI chips and Graviton CPUs. While Amazon continues to build out its internal hardware capabilities to reduce reliance on third-party suppliers, the sheer scale of the new Nvidia order underscores the chipmaker’s continued dominance in the AI sector. By securing this massive supply, Amazon aims to maintain its competitive edge in the cloud infrastructure market, ensuring it has the necessary compute capacity to support the increasingly complex workloads of its global customer base.

Key Takeaways

• Amazon has tripled its order of Nvidia GPUs, committing to 2 million additional units for AWS data centers through 2028.
• The partnership extends beyond hardware to include Nvidia’s robotics platforms, software stacks, and open-source AI models.
• Despite Amazon’s investment in its own custom silicon, the company remains heavily reliant on Nvidia to meet the explosive demand for AI compute power.

Editor’s Analysis & Impact

The massive expansion of the Amazon-Nvidia partnership signals a critical phase in the AI arms race: the transition from experimental development to large-scale industrial deployment. By securing such a significant portion of Nvidia’s future supply, Amazon is effectively hedging against potential hardware shortages while signaling to the market that the demand for AI compute is far from peaking. While Amazon’s parallel investment in custom silicon like Trainium suggests a long-term strategy to diversify its supply chain, the immediate reliance on Nvidia highlights the current ‘winner-take-all’ dynamic in high-end AI hardware. The broader implication is that cloud providers are now locked into a cycle of massive capital expenditure, betting that the ‘profitable tokens’ generated by AI will eventually justify the hundreds of billions of dollars being poured into data center infrastructure.

Frequently Asked Questions

Q: Why is Amazon buying more Nvidia chips if they are developing their own?
A: While Amazon is developing custom chips like Trainium and Graviton, Nvidia’s hardware remains the industry standard for high-performance AI training and inference. Amazon is utilizing both strategies to ensure it can meet the massive, immediate demand from its cloud customers.

Q: What specific Nvidia technologies are being integrated into AWS?
A: The partnership includes Nvidia’s Blackwell Ultra and Rubin GPUs, Vera CPUs, the Omniverse simulation platform, the Isaac robotics platform, and the hosting of Nemotron open models on Amazon Bedrock.