Amazon and Nvidia are dramatically escalating their partnership as Amazon Web Services (AWS) prepares to deploy 2 million Nvidia GPU chips in its data centers. These chips—including Nvidia’s next-generation Blackwell Ultra, Rubin, and Rubin Ultra series—are scheduled to roll out to AWS locations throughout 2027 and 2028 in an effort to increase AI computing capabilities.
Surging AI Needs Accelerate Amazon-Nvidia Deal
This announcement was made public on Wednesday during Nvidia’s quarterly earnings call, shortly after Amazon revealed plans to implement over 1 million Nvidia GPUs in AWS infrastructure beginning this year. Nvidia stated that demand has far outpaced initial expectations, prompting the companies to expand their original agreement.
While both Amazon and Nvidia kept the precise costs of the new arrangement under wraps, analysts project the total value of this multi-year GPU expansion could reach the tens of billions of dollars due to the pricey nature of advanced chips.
The deal covers more than just the installation of additional Nvidia chips. It features deeper integration of Nvidia technologies into AWS, spanning networking equipment, open AI models, CPUs, data management software, and the entire robotics stack.
Amazon Pursues Dual Track with Nvidia and In-House AI Hardware
Alongside this expanded Nvidia partnership, Amazon is ramping up its own AI chip projects to lessen dependence on external suppliers and build a competitive edge. AWS continues to enhance its Trainium processors—designed to rival Nvidia’s H100 and Blackwell offerings for deep learning—and is currently considering selling Trainium chips to outside organizations for data center usage. Amazon is also developing the Graviton CPU line, based on Arm architecture, to challenge top-tier products from Intel and AMD.
According to the company, its custom chip initiative has been scaling rapidly, boasting a $25 billion annualized revenue run rate. AI-focused clients such as Anthropic and OpenAI have committed a total of $225 billion to AWS’s custom silicon offerings.
Nvidia Expands Hardware and Prepares for Future Cloud Solutions
Despite growing competition, Nvidia is still regarded as “the GOAT” of AI silicon. In the upcoming third quarter, Nvidia will start providing AWS with the newly committed 2 million GPUs, along with a yet-to-be-specified count of Vera CPUs. These Vera chips will be delivered both as stand-alone processors and integrated with the Rubin GPU lineup, according to Nvidia CFO Colette Kress. CEO Jensen Huang unveiled a potential $200 billion TAM (total addressable market) for these new chips, indicating a major expansion opportunity sector-wide.
Nvidia has already begun distributing Vera processors to key partners, such as Oracle and SpaceXAI. Kress stated major AI research labs and leading hyperscale cloud providers will soon be receiving them as well.
Expanded Robotics and Enterprise Integration
Nvidia’s full physical AI platform is coming to Amazon’s robotics division as part of this partnership. The deployment will include Omniverse for simulation, Cosmos for world modeling, Isaac for robotics development, and Jetson for computation. This initiative follows the launch of a newly introduced entry-level Jetson hardware this week, aimed at broadening access to edge AI and robotics solutions.
Enterprise customers on AWS will soon be able to work with the Nvidia Nemotron open model family directly through Amazon Bedrock, AWS’s proprietary foundation model service, and via the established SageMaker machine learning suite.
Nvidia Achieves Historic Revenue Amid Accelerating AI Adoption
Nvidia reported $96.2 billion in revenue for its second fiscal quarter, surpassing all major analyst estimates. The data center division played a huge role, bringing in $89 billion and achieving a 117% annual growth rate. With the Rubin GPU ramping up production and shipment, Nvidia projects $108 billion in revenue for the upcoming third quarter.
To ensure future capacity, Nvidia has expanded its future supply and manufacturing commitments for data center projects to $279 billion (up from $119 billion previously). This figure covers $92 billion of anticipated spend for the remainder of the fiscal year and an expected $87 billion for fiscal 2028.
During the earnings call, CEO Jensen Huang remarked, “AI is now doing productive and useful work. AI is generating profitable tokens… If we had more compute, we could generate more profitable tokens, which results in more profit for all of the services.”
AI Infrastructure Spending: The Next Phase of Profitable Growth?
With competition among Amazon, Nvidia, and industry peers mounting, investors and analysts are monitoring whether these ambitious AI infrastructure investments will deliver continued growth and robust returns as the market matures.