Amazon And NVIDIA Team Up To Build The First Cloud AI Supercomputer With Grace Hopper

NVIDIA and Amazon Web Services officially announced an expansion of their existing partnership in supercomputing infrastructure, and services and software for generative AI at this year’s AWS re:Invent. According to the two companies this collaboration will “bring together the best of NVIDIA and AWS technologies that are ideal for training foundation models and building generative AI applications.”

Through this partnership, AWS will become the first cloud provider to utilize NVIDIA GH200 Grace Hopper Superchips with the new multi-node NVLink technology. It will be available in Amazon Elastic Compute Cloud (Amazon EC2) instances, which will make thousands of GH200 Superchips available at scale. This will enable customers to run large-scale AI/ML workloads.

NVIDIA DGX Cloud, the company’s AI-training-as-a-service, will also now run on AWS. This new service will be able to run large language models that can go past 1 trillion parameters, says NVIDIA. Customers will also receive access to integrated NVIDIA AI Enterprise software, and will even be able to reach out to NVIDIA AI experts.


Moreover, NVIDIA and AWS also plan on designing the fastest GPU-powered AI supercomputer in the world. This effort has been dubbed Project Ceiba. The supercomputer will feature 16,384 NVIDIA GH200 Superchips and will be able to process 65 exaflops of AI. NVIDIA says it will be using the supercomputer “for research and development to advance AI for LLMs, graphics and simulation, digital biology, robotics, self-driving cars, Earth-2 climate prediction and more.”

This strengthening partnership is good news for generative AI. It will leverage the companies’ core competencies to bring better offerings to companies and researchers working in the AI space.