BRK429 Breakout Tue 17 Nov 15:15 PST
The evolution of Azure infrastructure for AI models, apps, and agents
Narayan Annamalai, Aung Oo, Max Uritsky
Summary
From the recording / Microsoft AI summary
Microsoft said Cobalt 200 VMs are generally available, with guest RDMA in preview. AKS agent sandboxes and Application Gateway for Containers are in public preview, and Azure Files for agent workspaces is 'coming soon'. Cache work, including model-weight and context caches, is aimed at inference cost. The percentage and latency figures were Microsoft's demo results and carried no status label.
Session description
Microsoft's catalogue description
Agentic AI is reshaping AI infrastructure. As workloads move from training to production, success depends on a balanced foundation of CPUs, GPUs, storage, networking, and operational intelligence. In this session, learn how Azure IaaS helps optimize the full AI stack, from training and inferencing to autonomous agents, using infrastructure intelligence to recommend, right-size, and continuously optimize resources for better performance, efficiency, and innovation.