AWS Triples Its Nvidia GPU Order to 2 Million Chips
News Summary
Amazon Web Services (AWS) and Nvidia announced on August 26, 2026, Eastern Time, an expanded strategic partnership under which AWS will deploy 2 million additional Nvidia GPUs across its global data center infrastructure through 2027 and 2028. The commitment roughly triples the scale of an agreement the two companies struck only five months earlier, when AWS had pledged to bring in just over 1 million GPUs. Nvidia said the jump reflects demand that has "exceeded every forecast" from enterprises, AI labs, startups, and government customers racing to secure compute capacity for large-scale AI workloads.
What Was Ordered
The expanded deal covers Nvidia's next-generation Blackwell Ultra, Rubin, and Rubin Ultra GPU architectures, along with Nvidia Vera CPUs that will be integrated into AWS's infrastructure. Of the 2 million GPUs, 100,000 are earmarked specifically for secure U.S. government AI infrastructure meeting Impact Level 6 classification, a tier used for handling sensitive national security workloads. AWS also confirmed that new EC2 G7 instances will use Nvidia RTX PRO 4500 Blackwell Server Edition GPUs, extending Nvidia's hardware reach into a broader range of AWS's cloud computing lineup. Shipments under the new agreement are expected to begin in the third quarter, with full deployment continuing across 2027 and 2028.
The Scale of the Commitment
While AWS and Nvidia did not disclose the exact dollar value of the agreement, analysts estimate the deal is worth tens of billions of dollars based on typical per-unit GPU pricing at this scale. The announcement follows a blockbuster quarter for Nvidia, which reported data center revenue of about 89 billion U.S. dollars, up 117 percent year over year, within total quarterly revenue of roughly 96.2 billion U.S. dollars. Nvidia guided investors to expect around 108 billion U.S. dollars in revenue for the following quarter, and disclosed that its total supply commitments across cloud customers have grown to roughly 279 billion U.S. dollars, more than double the figure reported the previous quarter.
Executive Statements
Nvidia CEO Jensen Huang said the companies "have built one of the great growth engines of the AI era, and demand is running ahead of every forecast," underscoring how quickly cloud providers are having to revise their infrastructure plans upward. AWS CEO Matt Garman framed the expansion around customer flexibility, saying that "customers want the freedom to choose the best tools for their AI workloads, and they want confidence that everything works seamlessly together." Nvidia CFO Colette Kress has separately pointed to broadening AI adoption beyond the largest tech companies as a driver of sustained chip demand, spanning sovereign AI initiatives, enterprise deployments, and specialized AI infrastructure operators.
Beyond GPUs: A Deeper Technical Integration
The expanded agreement goes beyond raw chip volume. Nvidia's NVLink Fusion interconnect technology, paired with custom high-bandwidth memory, is being integrated directly into AWS's Trainium chip racks, effectively weaving Nvidia's networking architecture into AWS's own custom silicon stack. All Nvidia GPU-based and Trainium-based EC2 instances, including those using NVLink Fusion, will continue to run on AWS's proprietary Nitro System, which handles virtualization, security, and networking functions separately from the main compute silicon. On the software side, Nvidia's open Nemotron AI models are being made available on Amazon Bedrock and SageMaker, while Nvidia's cuDF and cuVS data-processing libraries are being adopted to accelerate workloads on Amazon EMR, with AWS citing up to 3.7 times faster data processing and up to 9 times faster vector indexing with a 25 percent cost reduction in certain benchmarks. The partnership also extends into physical and robotic AI, with Amazon Robotics adopting Nvidia's Jetson, Omniverse, and Isaac platforms for warehouse automation and robotics development.
Context: AWS Still Bets on Its Own Silicon Too
The Nvidia expansion arrives alongside, rather than instead of, Amazon's continued investment in its in-house Trainium chips. AWS CEO Matt Garman and other executives have said that Trainium remains the largest category of AI chips AWS deploys by volume, and that AWS's AI chip and AI services businesses have together reached an annualized revenue run rate of about 25 billion U.S. dollars. Company leadership has described the relationship with Nvidia as a long-term partnership rather than a competitive rivalry, even as Amazon markets Trainium to outside customers including major AI labs. Industry watchers see the dual-track strategy, buying Nvidia GPUs at massive scale while simultaneously scaling proprietary chips, as a hedge against supply constraints and pricing power as global demand for AI computing capacity continues to outpace available supply.
Why It Matters
The scale of the new order highlights how quickly demand for AI infrastructure has grown even beyond what major cloud providers projected just months earlier. For AWS customers, the expanded Nvidia partnership signals more available compute capacity for training and running large AI models, alongside deeper software tooling for data processing and vector search. For the broader technology industry, the deal reinforces Nvidia's central position in the global AI supply chain while also illustrating how large cloud operators are simultaneously deepening ties with Nvidia and building out their own custom silicon to manage costs and dependency risk over the long term.