Amazon is dramatically expanding its partnership with Nvidia, adding another 2 million of the chipmaker's GPUs to Amazon Web Services data centers as demand for artificial intelligence computing continues to accelerate.
The expanded agreement comes only five months after Amazon committed to deploying more than 1 million Nvidia GPUs across AWS infrastructure. The latest order effectively triples the scale of the planned deployment and highlights how rapidly cloud providers are increasing their investment in AI infrastructure.
The additional chips will include Nvidia's Blackwell Ultra, Rubin, and Rubin Ultra GPUs, with deployments expected across AWS data centers during 2027 and 2028.
Financial terms were not disclosed, but an order involving millions of high-end AI processors could represent tens of billions of dollars in hardware and infrastructure spending.
Amazon and Nvidia announced the expanded partnership during Nvidia's latest quarterly earnings update.
Under the agreement, AWS will add another 2 million Nvidia GPUs to its infrastructure starting in the third quarter, significantly expanding the computing capacity available to businesses, AI laboratories, startups, and government customers using Amazon's cloud platform.
Amazon had already announced plans earlier in 2026 to deploy more than 1 million Nvidia GPUs. According to Nvidia, demand for that computing capacity exceeded the companies' previous expectations, prompting the much larger expansion.
The new agreement shows how difficult it remains for major cloud providers to keep pace with demand for AI infrastructure.
Training increasingly sophisticated models requires enormous amounts of computing power. Running those models for millions of users also creates significant and recurring demand for GPUs.
For AWS, securing additional Nvidia hardware provides more capacity to serve customers building and operating AI applications.
The expanded deployment will include several generations of Nvidia's newest AI processors.
Amazon plans to use Blackwell Ultra GPUs as well as Nvidia's upcoming Rubin and Rubin Ultra processors.
Rubin represents Nvidia's next major GPU architecture following Blackwell and is expected to become an important part of the company's next generation of data center infrastructure.
The inclusion of Rubin chips in Amazon's agreement suggests AWS is planning its AI infrastructure several years in advance rather than simply responding to current demand.
Cloud computing companies increasingly need to secure processors, networking hardware, memory, energy, and data center capacity long before customers actually use the infrastructure.
Amazon's agreement gives the company access to Nvidia's current technology while also reserving capacity for future generations of AI hardware.
The expanded relationship is about more than Amazon purchasing additional graphics processors.
Nvidia plans to integrate a broader collection of its technology across AWS, including networking systems, CPUs, AI models, data processing software, and robotics platforms.
Networking has become particularly important as AI systems grow larger.
Modern AI data centers often connect thousands or even hundreds of thousands of accelerators into massive computing clusters. The performance of those systems depends not only on individual GPUs but also on the networking technology that allows them to communicate efficiently.
By integrating more of Nvidia's hardware and software stack, AWS could offer customers infrastructure designed around Nvidia technology from the processor level through networking and software.
The arrangement potentially gives Nvidia a larger role in how AWS builds its future AI infrastructure.
The enormous Nvidia order comes despite Amazon investing heavily in processors designed to reduce its reliance on outside chip suppliers.
AWS has developed Trainium chips for AI training and Inferentia processors for AI inference workloads.
Trainium competes directly with Nvidia GPUs for some machine learning applications. Amazon has also indicated that it is exploring opportunities to sell Trainium processors to companies that want to operate them inside their own data centers.
Amazon's custom semiconductor efforts extend beyond AI accelerators.
Its Graviton family of Arm-based CPUs has become an important part of AWS and competes with traditional server processors from companies such as Intel and AMD.
Amazon has said its custom chip business has reached a $25 billion annualized revenue run rate, demonstrating that its internal semiconductor strategy has already become a significant operation.
The latest Nvidia agreement suggests Amazon believes custom chips and Nvidia GPUs can coexist rather than requiring the company to choose one strategy.
Amazon's decision to dramatically increase its Nvidia purchases demonstrates the chipmaker's continuing influence over the AI industry.
Technology companies have spent years developing alternatives to Nvidia GPUs, partly because of their cost and partly because companies want greater control over critical infrastructure.
Google has developed its Tensor Processing Units, Amazon has Trainium, and other major technology companies are designing or exploring custom AI accelerators.
Yet demand for Nvidia hardware remains exceptionally strong.
One reason is Nvidia's mature software ecosystem. Developers have spent years building applications around the company's CUDA platform and related software tools.
Nvidia also sells a broader infrastructure stack that includes networking equipment, CPUs, software libraries, AI models, and complete computing systems.
That combination makes replacing Nvidia more complicated than simply designing a competing processor.
Amazon's latest order demonstrates that even companies investing billions of dollars into custom silicon continue to depend heavily on Nvidia for large portions of their AI infrastructure.
The partnership will also introduce Nvidia's Vera CPUs into Amazon's cloud infrastructure.
Nvidia plans to send an unspecified number of Vera processors to AWS, with some integrated alongside Rubin GPUs and others operating as standalone CPUs.
The move is important because Nvidia has historically been best known for graphics processors rather than general-purpose server CPUs.
CEO Jensen Huang has described CPUs as a major new market opportunity for Nvidia, estimating a potential addressable market worth hundreds of billions of dollars.
If Vera gains adoption among major cloud providers, Nvidia could expand its influence beyond AI accelerators and compete more directly in the broader data center processor market.
AWS could become an important proving ground for that strategy.
Amazon also plans to use Nvidia technology beyond its data centers.
The company is expected to adopt Nvidia's physical AI technology stack for its warehouse robotics operations.
That ecosystem includes Omniverse for simulation and digital twins, Cosmos for world models, Isaac for robotics development, and Jetson computing hardware for robots and edge AI systems.
Amazon operates one of the world's largest fleets of warehouse robots.
Integrating Nvidia's robotics platforms could therefore become a significant test of the chipmaker's ambitions in physical AI, a category Nvidia believes could become another major computing market.
AI models are increasingly moving beyond chatbots and software applications into robots, autonomous machines, manufacturing systems, and other physical environments.
Amazon's logistics network provides Nvidia with an opportunity to deploy those technologies at enormous scale.
The partnership is also expanding at the software level.
AWS plans to make Nvidia's Nemotron family of open AI models available through Amazon Bedrock and SageMaker.
Amazon Bedrock allows companies to access and build applications using foundation models through a managed cloud service, while SageMaker provides tools for developing, training, and deploying machine learning systems.
Adding Nvidia models to these platforms gives AWS customers another collection of models to choose from while extending Nvidia's presence further up the AI technology stack.
Nvidia is increasingly positioning itself as more than a semiconductor manufacturer.
The company now develops AI models, software frameworks, cloud services, networking technology, CPUs, robotics platforms, and complete AI systems.
Its expanded Amazon partnership reflects that broader strategy.
The Amazon announcement arrived alongside another enormous quarter for Nvidia.
The company reported quarterly sales of $96.2 billion, exceeding analyst expectations.
Its data center business generated approximately $89 billion, representing the overwhelming majority of Nvidia's revenue and increasing 117 percent compared with the previous year.
Nvidia expects revenue to climb to about $108 billion in the following quarter as shipments of its next generation Rubin products begin contributing to sales.
The numbers illustrate the extraordinary amount of capital flowing into AI infrastructure.
Technology companies, AI developers, cloud providers, and governments are spending heavily to secure the computing resources required to train and operate increasingly powerful AI systems.
Amazon's additional order indicates that this demand is not slowing enough to reduce the industry's appetite for new GPUs.
Nvidia is also making enormous commitments to ensure it can manufacture enough hardware to meet expected demand.
The company has committed roughly $279 billion toward securing manufacturing and supply capacity for current and future data center projects.
That figure has increased significantly from the previous quarter as Nvidia works to secure advanced manufacturing, memory, packaging, and other components required for AI processors.
Building cutting-edge AI chips involves a complicated global supply chain.
Demand for advanced memory and semiconductor manufacturing capacity can become a bottleneck even when customers are prepared to spend billions of dollars.
Long-term agreements with companies such as Amazon provide Nvidia with greater visibility into future demand, potentially allowing it to make larger manufacturing commitments with more confidence.
Amazon's decision to triple its planned Nvidia GPU deployment highlights the scale of the global race to build AI infrastructure.
AWS competes directly with Microsoft Azure, Google Cloud, Oracle, and a growing collection of specialized AI cloud companies for customers that need large quantities of computing power.
Access to advanced GPUs has therefore become a competitive advantage.
Companies capable of offering large GPU clusters can attract AI laboratories and enterprises developing increasingly computationally intensive models.
At the same time, cloud providers are attempting to control costs by developing their own processors.
Amazon's strategy increasingly appears to involve pursuing both approaches at once. The company can continue developing Trainium and other custom chips while simultaneously purchasing enormous quantities of Nvidia hardware to satisfy immediate customer demand.
The expanded agreement illustrates how deeply connected the largest technology companies have become as the AI infrastructure boom continues.
Amazon wants to reduce its dependence on Nvidia through custom chips, yet customer demand is pushing AWS to buy millions more Nvidia processors.
Nvidia wants to sell GPUs, but it is also moving into areas where Amazon already operates, including CPUs, cloud services, AI models, and enterprise software.
That creates an unusual relationship in which the companies are simultaneously partners and potential competitors.
For now, demand appears strong enough to support both strategies.
If Amazon successfully deploys more than 3 million Nvidia GPUs while continuing to expand its own semiconductor business, AWS could become one of the world's largest platforms for both proprietary and third-party AI computing technology.
For Nvidia, the agreement provides another powerful indication that the industry's largest cloud providers still need massive quantities of its hardware.
The bigger question will be whether the hundreds of billions of dollars now being invested in AI computing capacity ultimately produce enough revenue and productivity to justify the unprecedented infrastructure buildout.
Share your thoughts about this article.
Be the first to post a comment!
