Microsoft Azure CTO Mark Russinovich Showcases Next-Generation AI Innovations Across Cloud, On-Premises, And Edge!

As AI adoption grows, cloud providers are facing increasing pressure to deliver the computing power needed for training and running advanced models. To meet those demands, Microsoft is expanding and redesigning key parts of its Azure infrastructure, from data centers and servers to networking technologies built specifically for large-scale AI workloads.

 

TL;DR

 

• Microsoft launched new Azure infrastructure for AI workloads.
• Azure Boost delivers faster storage and networking performance.
• New Azure technologies also support massive AI GPU clusters.

During a recent Azure session on day 2 of Microsoft Build 2026, Microsoft Azure CTO, Mark Russinovich provided an overview of Microsoft's latest infrastructure innovations aimed at supporting the next generation of AI applications. He noted that some of the technologies discussed are still under development and may evolve before becoming generally available.


A major focus of the session was the company's approach to AI-driven infrastructure design. Russinovich highlighted Microsoft's Fairwater data centers in Atlanta and Wisconsin, emphasizing their scale and ongoing expansion. These facilities are equipped with extensive liquid-cooling systems and high-density cabling infrastructure designed to handle the demands of modern AI computing.

Microsoft also showcased Azure Boost, a specialized offload card that moves storage and networking data processing away from the main CPU. The technology combines DPUs, ARM cores, and FPGA-based acceleration to improve efficiency and free server processors for application workloads. According to Microsoft, the latest Azure Boost generation delivers up to 20 GB/s remote storage throughput, 1 million IOPS, and 400 Gbps networking performance.

The company has also introduced bare metal Azure instances, particularly for OpenAI workloads. These instances provide direct access to physical servers, eliminating virtualization overhead and enabling greater performance for demanding AI applications.
 
To support increasingly large GPU deployments, Microsoft has developed a new networking protocol called Multipath Reliable Connection (MRC). The protocol is designed to improve both resilience and performance across large-scale AI training environments where thousands of GPUs must communicate efficiently.

The announcements show Microsoft's ongoing efforts to strengthen Azure's foundation as AI workloads continue to grow in size, complexity, and infrastructure requirements.