Project Ceiba is a groundbreaking collaboration between AWS and NVIDIA to construct one of the world's fastest AI supercomputers, hosted exclusively on AWS, to power NVIDIA's internal research and development in AI . The supercomputer is built on NVIDIA's latest GB200 NVL72 architecture—a liquid-cooled, rack-scale system featuring fifth-generation NVLink—scaling to 20,736 NVIDIA Blackwell GPUs connected to 10,368 NVIDIA Grace CPUs . Project Ceiba is capable of processing 414 exaflops of AI compute, making it approximately 375 times more powerful than Frontier (the current world's fastest supercomputer), with a single second of computation equivalent to over 1,660 years of one calculation per second from every human on Earth . Project Ceiba is the first system to leverage the massive scale-out capabilities enabled by fourth-generation AWS Elastic Fabric Adapter (EFA) networking, providing 1,600 Gbps per superchip of low-latency, high-bandwidth throughput . To address the unprecedented power density, AWS pioneered liquid cooling at scale in its data centers, delivering efficient and sustainable high-performance computing solutions . The system incorporates industry-leading security features including the NVIDIA Blackwell GPU architecture's secure communication between GPUs, integrated with AWS Nitro System and EFA technologies, enabling end-to-end encrypted data for generative AI workloads . NVIDIA's research and development teams will use Project Ceiba to drive advancements across large language models, graphics (images, videos, 3D generation), simulation, digital biology, robotics, autonomous vehicles, and climate prediction with NVIDIA Earth-2 . The supercomputer is available through NVIDIA's DGX Cloud architecture, an end-to-end scalable AI platform available on AWS . As of 2026, Project Ceiba has been updated to include GB300s as well as the original GB200 configuration .