Microsoft CEO Confirms Azure Will Use AMD Helios and NVIDIA Vera Rubin

Microsoft is expanding Azure’s AI computing capacity with next-generation rack-scale systems from both AMD and NVIDIA. The strategy gives the company access to more hardware while reducing its dependence on a single supplier.

The expansion follows Microsoft’s latest financial results, as FY26 Q4 earnings reached $90 billion, driven partly by rising demand for cloud and AI services.

Azure will rank among the first cloud platforms to deploy systems based on AMD Helios and NVIDIA Vera Rubin.

NVIDIA Vera Rubin combines AI computing and networking

NVIDIA Vera Rubin is a rack-scale AI platform built around Rubin GPUs and Vera CPUs.

The system combines computing, networking, connectivity, and storage hardware in one integrated platform. Its components include NVLink 6 switching, Spectrum-6 networking, ConnectX-9 connectivity, BlueField technology, and Groq 3 hardware.

This design should allow Microsoft to deploy large AI clusters without assembling every part of the infrastructure separately.

AMD Helios could cost significantly more

Helios is AMD’s first complete rack-level platform designed for demanding AI workloads. It combines AMD Instinct MI455X GPUs with sixth-generation EPYC processors.

The system also includes Pensando networking hardware, data processing units, Infinity Fabric connectivity, and AMD’s ROCm software stack.

Futurum estimates that one AMD Helios rack could cost between $5 million and $5.5 million. By comparison, a second-generation NVIDIA Rubin rack could cost between $3.5 million and $4 million.

Based on those estimates, Helios could cost roughly 40% more than NVIDIA’s platform. However, the final value will also depend on performance, power efficiency, software support, availability, and the types of AI workloads Microsoft plans to run.

Microsoft avoids relying on one AI supplier

Microsoft’s decision to deploy hardware from both AMD and NVIDIA reflects a broader hardware-agnostic strategy.

Instead of committing Azure entirely to one chip supplier, Microsoft can choose systems based on performance, cost, supply, and customer requirements. This approach could also strengthen its negotiating position as demand for AI accelerators continues to grow.

Microsoft appears to be applying the same strategy to AI models. Earlier reports suggested the company could consider DeepSeek V4 or a similar model for Copilot Cowork to support complex agent-based tasks at a lower cost.

By combining different hardware platforms and AI models, Microsoft can expand Azure and Copilot without depending entirely on one chipmaker or AI company.

Via Wccftech