Amazon Expands Nvidia Partnership With 2 Million More GPUs for AWS Data Centers

Amazon will add 2 million Nvidia GPUs to its AWS data centers over the next 2 years, extending a partnership that has grown sharply since an earlier commitment to deploy more than 1 million chips. The new hardware will include Nvidia’s Blackwell Ultra, Rubin, and Rubin Ultra GPUs, with deliveries planned for 2027 and 2028.

The announcement, made during Nvidia’s quarterly earnings call, marks a major expansion of Amazon’s AI infrastructure spending. Financial terms weren’t disclosed, but the size of the order places its value in the tens of billions of dollars based on GPU pricing. Nvidia said demand has exceeded the expectations behind the companies’ earlier agreement.

The additional capacity will support customers running AI model training and inference through Amazon Web Services. Startups, large enterprises, AI research labs, and governments are all contributing to what the companies described as surging demand for computing power.

The order also gives Amazon a larger supply of the processors that dominate advanced AI workloads. Nvidia’s GPUs handle parallel calculations needed for training and serving models, while AWS provides the data centers, networking, software, and commercial access that turn those chips into cloud services. The scale of the expansion shows that customer demand is shaping infrastructure plans years before the chips arrive.

Amazon and Nvidia are broadening their relationship across the data center stack, rather than limiting it to processor purchases. Nvidia’s networking systems, CPUs, data-processing software, open models, and robotics platform will also be integrated into AWS infrastructure.

That wider arrangement matters because large AI clusters depend on more than accelerator chips. Networking equipment links thousands of GPUs into a single computing system, and software determines how efficiently customers can schedule workloads and move data. For AWS, deeper Nvidia integration can make the platform easier to sell to organizations already building around Nvidia technology. For Nvidia, AWS becomes a larger channel for its hardware and related products.

Amazon’s Nvidia deal sits alongside a separate effort to reduce its dependence on outside processors. AWS has developed Trainium AI chips for deep-learning workloads and Graviton CPUs for general-purpose cloud computing. The company has also discussed selling Trainium capacity or systems to other data center operators.

Amazon’s custom-chip business has reached a $25 billion annualized revenue run rate, driven by $225 billion in commitments from AI customers including Anthropic and OpenAI, TechCrunch reported. That investment gives Amazon bargaining power and a second infrastructure path, while Nvidia remains the company’s main source of top-tier AI accelerators.

The expanded order reinforces Nvidia’s role as the leading supplier for large-scale AI computing, even as cloud providers design alternatives. Amazon’s willingness to commit to millions of future GPUs indicates that custom silicon hasn’t removed the need for Nvidia’s newest accelerators.

The deal also connects Nvidia more deeply to AWS customers, who can access Nvidia-based systems without purchasing and operating their own data centers. Microsoft and Google face similar pressure to secure enough AI capacity while developing proprietary chips and cloud services. Nvidia’s influence therefore extends from component supply into the software, networking, and cloud infrastructure that enterprises use to deploy models.

Amazon’s immediate task is converting the commitment into usable capacity across AWS data centers. That involves power, cooling, networking, construction, and software deployment as well as chip availability. The long delivery window gives both companies time to prepare facilities for multiple Nvidia generations.

Customers will also determine how much of the capacity becomes revenue. Strong demand from AI labs and enterprises would support the expansion, while Amazon’s Trainium and Graviton programs will shape the mix of workloads hosted on AWS. The agreement places Nvidia hardware and Amazon’s custom silicon on the same platform, creating a competitive relationship that will define AWS infrastructure through the end of the decade.