Oracle announced plans to bring over 18 zettaFLOPS of AI infrastructure to its cloud offerings by the latter half of next year, leveraging both Nvidia and AMD technologies. This ambitious deployment will feature a staggering 800,000 Nvidia GPUs reaching up to 16 zettaFLOPS of AI performance. These GPUs, integrated into the Oracle Cloud Infrastructure’s Zettascale10 offering, will utilize Nvidia’s Spectrum-X Ethernet switching platform to enhance connectivity and performance. Nvidia will not only supply the hardware but also bolster Oracle’s AI capabilities through various cloud services. Meanwhile, AMD joins the venture with plans to deploy 50,000 of its MI450X accelerators, capable of delivering impressive compute power using the innovative Ultra Accelerator Link (UALink) technology in their Helios racks. AMD’s expansion indicates growing momentum in the AI marketplace as it targets more significant computing milestones in the upcoming years. Following successful initial rack deployments at Oracle, the MI450X racks will provide 2.9 exaFLOPS of sparse compute power and enormous memory bandwidth. The advancements promise Oracle’s infrastructure will be primed to meet burgeoning AI demands. Notably, OpenAI is poised to benefit significantly from Oracle’s expanded capabilities, as both Nvidia and AMD have secured investment deals dependent on large-scale accelerator deployments at OpenAI’s partnering data centers.