Microsoft Events presents a breakdown of Azure’s latest AI infrastructure at Ignite 2025, detailing performance improvements and new hardware platforms for high-scale workloads.

Azure AI Infrastructure Updates for High-Performance Workloads

Presented by: Microsoft Events, Param Shah, Matt Vegas

This session from Microsoft Ignite 2025 reviews the ongoing transformation of Azure’s AI infrastructure to support the increasing demands of both frontier and enterprise AI workloads. Topics include:

Key Infrastructure Innovations

  • Silicon-to-Systems Optimization: Holistic enhancement of hardware and software integration for AI, from custom silicon to distributed compute systems and networking.
  • Datacenter Advances: Next-gen cooling technologies, power management, and infiniBand networking (Quantum-2) ensure scalability and reliability for massive parallel AI workloads.
  • Security Enhancements: Improved threat mitigation and data integrity across hardware and software layers.
  • NVIDIA Grace Blackwell Superchips: Powering the new ND GB200/GB300 VM families, enabling exascale performance and throughput exceeding 860,000 tokens/sec on LLAMA 70B.
  • NC H100 and NC RTX Blackwell VMs: Purpose-built for robust enterprise-level AI inferencing and training.

Software and Platform Updates

  • Hardware-Software Codesign: Coordinated development between device engineering and software stack for optimal performance in AI scenarios.
  • Agent Mode in Microsoft 365 Apps: Demonstrations include discounted cash flow modeling in Excel and automated content gathering in Word—all powered by coordinated AI agents.
  • Agent Context Transfer: Techniques to move agent state across applications and integrate external models, such as Anthropic for orchestration.

Additional Highlights

  • Major Announcements: Introduction of Work IQ, new features for improved data access methods, and upcoming developer summit details.
  • Interactive Format: Session combines discussion, live demos, and Q&A, allowing attendees to explore Azure’s technical roadmap for AI infrastructure firsthand.

Resources

Speakers: Param Shah, Matt Vegas

For full session content and demos, refer to the on-demand archive on the Microsoft Ignite site.