Amazon Web Services (AWS) has established plans to incorporate an additional two million Nvidia Graphics Processing Units (GPUs) between 2027 and 2028. This acquisition aims to significantly increase the computational capacity available for various artificial intelligence (AI) applications.
These components will be integrated into the company's global infrastructure, allocated in facilities known as 'AI factories,' which are data centers built specifically to handle AI demands.
This move is part of a continuous expansion of collaboration between AWS and Nvidia, which also covers new processors, chip communication systems, AI models, and robotics-focused technologies.
Capacity Expansion and AI Demand
The new capacity will include GPUs belonging to the Blackwell Ultra, Rubin, and Rubin Ultra families. According to both companies, the demand for AI infrastructure has exceeded initial projections. Previously, during the GTC 2026 event, AWS had announced intentions to add over one million Nvidia GPUs starting in 2026, and now plans to add another two million units in the following two years.
This infrastructure will support a variety of applications, ranging from autonomous AI systems to scientific research, corporate process optimization, and AI systems that interact directly with the physical environment.
New Components and Performance
Additionally, AWS intends to increase its offering of Blackwell-based infrastructure. Among the newly acquired items is the Nvidia RTX PRO 4500 Blackwell Server Edition, which will be implemented in Amazon EC2 G7 instances. Nvidia reports that these instances provide 4.6 times better performance in AI inference and 2.1 times better graphics performance compared to the previous generation's G6 instances.
This expansion reflects a transformation in how companies use AI; customers are migrating models from testing stages to production environments, scaling up their AI applications.
Advanced Use Cases and Robotics
Cited application examples include AI systems that execute multi-stage tasks, advances in scientific discovery, automation of business workflows, and robotics. The partnership also extends to so-called physical AI, a field dedicated to enabling AI systems to perceive and act upon the real world.
In this context, Amazon Robotics will utilize technologies provided by Nvidia, such as the Jetson, Omniverse, and Isaac platforms. This combination enables work with both on-site processing and simulations and the development of robotic systems.
Nvidia considers physical AI one of the next major growth areas for its software and chips. The Rubin platform itself was designed to support, among other uses, agentic AI applications and systems requiring vast amounts of processing.
Government Projects
The expansion also includes initiatives directed at the United States government. Nvidia and AWS plan to develop AI factories for government workloads, including infrastructure equipped with 100,000 Nvidia GPUs within secure AWS environments. This structure will be dedicated to governmental and national security needs that require high processing capacity and rigorous security standards.
With this expansion, AWS and Nvidia aim to provide robust infrastructure for a new generation of AI systems, which require not only greater processing power but also high-speed networks, specialized storage, and efficient integration between various hardware and software components.