AI21 Labs speeds up AI workloads with Google Cloud

AI21 Labs says switching to Google Cloud's AI Hypercomputer dropped its wait times for key AI training jobs from 72 hours to just 12, an 83% reduction. Previously, researchers relied on manual Slack requests and a patchwork of GPU resources, which led to delays and scheduling headaches. Now, a shared Kubernetes cluster pools thousands of high-end NVIDIA GPUs to manage workloads automatically, eliminating manual interventions.

• Wait times dropped from 72 to 12 hours
• Manual scheduling cut from 20 times a week to none
• Cluster pools NVIDIA H100 and H200 GPU instances
• High-priority jobs can now access resources faster