Amazon Web Services is preparing for another major expansion of its AI computing infrastructure, committing to deploy an additional 2 million Nvidia GPUs across its global cloud footprint. The deployment will span 2027 and 2028, extending an already aggressive buildout that reflects how quickly demand for accelerated computing has moved beyond earlier capacity forecasts. The agreement also broadens the relationship between AWS and Nvidia beyond GPUs, covering CPUs, networking, open models, data processing, and robotics. For AWS, the move positions Nvidia infrastructure as a deeper part of its long-term strategy for serving frontier AI developers, enterprises, and government customers.

The new GPU commitment includes Nvidia’s Blackwell Ultra, Rubin, and Rubin Ultra platforms, giving AWS a pipeline that stretches across multiple generations of accelerator technology. AWS had previously said during Nvidia GTC 2026 that it planned to bring more than one million Nvidia GPUs online across its platform during 2026. Demand has since exceeded that expectation, pushing the companies toward a substantially larger deployment schedule. The scale of the latest commitment also suggests that hyperscale cloud providers are planning capacity further ahead as AI training, inference, and agentic workloads continue to consume increasingly large amounts of compute.

AWS expands beyond GPUs into the full AI stack

The agreement reaches into the processor layer as AWS prepares to introduce Nvidia Vera CPU-based infrastructure on its cloud platform. The companies also plan to extend Nvidia NVLink Fusion with custom Nvidia high-bandwidth memory, creating another path for tightly integrated compute architectures as AI systems become more demanding. AWS will build AI factories for the US government using 100,000 GPUs on secure AWS infrastructure, adding a significant government-focused component to the collaboration. The broader arrangement points toward infrastructure designed around complete AI systems rather than isolated accelerator deployments.

AWS is also adding Nvidia’s Nemotron models to its platform, widening the software choices available to customers building AI applications. The partnership will cover data processing and open-model capabilities alongside the underlying compute, networking, and security layers. Amazon Robotics will further work with Nvidia to adopt the company’s physical AI platform, connecting cloud-scale AI infrastructure with robotics systems operating in physical environments. At the same time, the partnership is becoming a platform strategy that links model development, infrastructure, deployment, and physical AI rather than treating each area as a separate investment.

Blackwell capacity expands across EC2

AWS is also increasing Blackwell capacity through Nvidia RTX Pro 4500 Blackwell Server Edition GPUs for its EC2 G7 instances. AWS says the instances deliver 4.6 times the AI inference performance and 2.1 times the graphics performance of previous-generation G6 instances. The move gives customers another option for workloads that require a combination of accelerated inference and graphics capabilities without relying solely on the highest-end training systems. It also shows how AWS is expanding the Nvidia portfolio across different performance tiers as cloud demand becomes more varied.

The infrastructure push comes as Nvidia continues to report exceptional demand for its computing platforms. Nvidia reported $96.2 billion in revenue for the second quarter of fiscal 2026, representing an 18 percent increase from the previous quarter and a 106 percent increase from the same period a year earlier. Both GAAP and non-GAAP gross margins reached 75 percent during the quarter, underscoring the financial strength behind the accelerator market. However, AWS’s decision to add two million more GPUs indicates that customer demand remains a central constraint even as Nvidia scales production and introduces new generations.

AWS keeps older Nvidia systems in service

AWS’s strategy also includes extending the useful life of existing Nvidia hardware rather than treating each new generation as an immediate replacement cycle. CEO Matt Garman has previously said AWS is still operating six-year-old Nvidia A100 servers and has “never retired an A100” server. That approach gives AWS another lever for managing capacity as newer Blackwell and Rubin systems enter the fleet. It also highlights the importance of total compute availability, where older accelerators can continue serving suitable workloads while newer systems handle more demanding AI applications.

“Nvidia and AWS have built one of the great growth engines of the AI era, and demand is running ahead of every forecast,” said Jensen Huang, founder and CEO of Nvidia. “For 16 years, we have scaled Nvidia computing in the cloud together. Now, we are expanding our partnership across the full stack — GPUs, CPUs, networking, open models and software — to make agentic and physical AI real at an unprecedented pace and scale that only AWS and Nvidia can deliver. This expansion reflects customers’ demand for Nvidia’s platform on AWS.”

Cloud AI capacity becomes a strategic battleground

The latest commitment changes the scale at which AWS is planning its Nvidia relationship. Two million additional GPUs represent not just a hardware purchase but a multi-year bet on sustained demand for accelerated computing across commercial AI, inference, government systems, robotics, and emerging agentic workloads. The inclusion of CPUs, networking, memory, models, and physical AI suggests that AWS wants customers to consume an increasingly integrated Nvidia technology stack through its cloud. That strategy could strengthen AWS’s position among organizations that want Nvidia’s newest platforms without building and operating the corresponding infrastructure themselves.

“Customers want the freedom to choose the best tools for their AI workloads, and they want confidence that everything works seamlessly together,” said Matt Garman, CEO of AWS. “That’s why we’ve invested deeply with Nvidia to make AWS the best place to run Nvidia AI technologies, optimizing performance across our infrastructure from networking and security to deployment. This expanded collaboration gives frontier labs, enterprises and governments even more ways to build and deploy AI on AWS.”

The AWS-Nvidia agreement ultimately reflects a broader shift in the economics of cloud infrastructure, where access to GPUs increasingly determines how quickly customers can train, deploy, and scale AI systems. AWS is responding by committing capacity years ahead while supporting several generations of Nvidia accelerators and keeping older hardware productive inside its fleet. The strategy also gives Nvidia a deeper route into cloud deployments across compute, software, networking, and physical AI. With two million more GPUs scheduled for 2027 and 2028, the companies are effectively betting that the current AI infrastructure race has much further to run.