Singapore – Amazon Web Services (AWS) and NVIDIA have expanded their long-standing strategic partnership, with plans to deploy two million additional NVIDIA GPUs across AWS’s global infrastructure.
The expansion builds on 16 years of joint work between the two companies and extends their collaboration across AI factories, CPUs, networking, open models, data processing, and robotics.
According to the companies, the move is designed to deliver co-engineered solutions that help customers scale AI development at unprecedented speed.
Notably, the expansion sits alongside Amazon’s own custom silicon, meaning customers retain the flexibility to choose between NVIDIA GPUs, AWS Trainium chips, or a combination of both, depending on their workload requirements.
AWS and NVIDIA have collaborated for nearly two decades, a partnership that produced the world’s first GPU-accelerated cloud instance. As it stands today, AWS offers what the companies describe as the widest range of NVIDIA GPU solutions available to customers.
As demand for AI infrastructure accelerates, the two companies say they are taking their collaboration further. AI workloads are scaling rapidly, spanning model training and deployment through to data processing and the powering of intelligent applications.
Customers, meanwhile, are increasingly moving from pilot projects to full production across agentic AI, scientific discovery, enterprise automation, and robotics, driving demand for infrastructure capable of keeping pace.
“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, optimising 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”, he continued.
“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”, Huan continued.
Turning to the scale of the investment, AWS had already announced plans at NVIDIA GTC 2026 to add more than one million NVIDIA GPUs from 2026 onward. Since then, however, demand has reportedly exceeded those initial projections.
As a result, AWS now plans to deploy a further two million NVIDIA GPUs between 2027 and 2028 across its global infrastructure, including its AI factories. This additional capacity is intended to support workloads ranging from agentic AI and scientific discovery to enterprise automation and physical AI.
Customers are said to already be seeing tangible outcomes from the partnership, including faster drug discovery and more efficient fraud detection systems.
In parallel, AWS and NVIDIA are also working on advanced networking technology aimed at connecting GPUs more efficiently for large-scale AI training.
On the CPU front, the two companies are bringing NVIDIA Vera CPU-based infrastructure to AWS, offering customers an additional option for agentic AI workloads that require high-performance CPU compute alongside accelerated infrastructure. This, the companies say, reflects AWS’s broader approach of offering the widest possible range of compute choices rather than a single standardised solution.
Vera has been built to handle the CPU-intensive work behind agentic AI and reinforcement learning, including code execution, tool use, sandboxing, analytics, data pipelines, and orchestration.
Functioning both as a host CPU for accelerated systems and as a standalone CPU for AI factory workloads, Vera is designed to keep GPUs supplied with data, agents responsive, and training loops running.
Elsewhere, at re:Invent 2025, AWS confirmed support for NVIDIA’s NVLink Fusion high-speed chip interconnect technology within its next-generation Trainium chips.
Building on this, Amazon’s Annapurna Labs will now work with NVIDIA’s new custom high-bandwidth memory (NVHBM) technology, developed in partnership with memory suppliers, which is expected to give Trainium access to faster, more power-efficient memory.
Together, NVHBM and NVLink Fusion will allow Annapurna Labs to draw on NVIDIA’s custom memory technology and scale-up architecture, enhancing performance and efficiency for AI workloads while integrating Trainium and GPUs within a shared rack-scale architecture.
Security remains a key focus of the expanded partnership, particularly for government use. AWS and NVIDIA plan to build AI factories for the US government, delivering NVIDIA’s AI stack—including 100,000 GPUs—on AWS’s secure infrastructure for federal and national-security workloads classified at Impact Level 6 (IL6) and above, among the highest government security classifications.
Beyond these new commitments, the companies point to existing technical integrations already delivering results for customers today.
On security and reliability, all NVIDIA GPU-based and Trainium-based EC2 instances, including those using NVLink Fusion, are built on the AWS Nitro System and connected through Elastic Fabric Adapter (EFA), which the companies say helps preserve security, reliability, and network performance at scale.
For open models, NVIDIA’s Nemotron family is available on Amazon Bedrock as fully managed, serverless models, as well as on Amazon SageMaker for customers who prefer to deploy and fine-tune models on their own infrastructure.
Data processing has also seen gains, with GPU-accelerated processing on Amazon EMR using NVIDIA cuDF delivering up to 3.7 times faster processing speeds and 30% better price-performance for Apache Spark workloads compared with CPU-based configurations. GPU-accelerated vector indexing on Amazon OpenSearch Service, meanwhile, delivers up to nine times faster indexing at a quarter of the cost.
Finally, in robotics, Amazon Robotics is integrating NVIDIA’s full-stack physical AI platform—comprising Jetson, Omniverse, and Isaac—to accelerate next-generation warehouse automation through simulation, synthetic data generation, and real-world validation.
