Our AI Hypercomputer underpins our Cloud customers’ most demanding AI workloads. Its hardware and software layers are optimized to deliver more intelligence per dollar for training and inference.

Today at Google Cloud Next 25, we introduced updates throughout the AI Hypercomputer stack:

• AI-optimized hardware: Our new seventh-generation TPU, Ironwood, is designed specifically for thinking and inferential AI models. Ironwood offers five times more peak compute capacity and six times the high-bandwidth memory (HBM) capacity compared to the prior-generation TPU.
• Software advances for inference: Updates to our AI Hypercomputer’s software layer help developers optimize compute resources, while speeding up AI workloads. These advances are shortening the time between training and inference.
• Flexible consumption options: There are more ways for businesses to control costs with flexible consumption models in Dynamic Workload Scheduler.

Learn more about these AI infrastructure updates on the Google Cloud blog.