Nvidia and Microsoft Build a Full‑Stack AI Platform for Windows, Azure and On‑Premises

The partnership links new Nvidia Windows PCs and deskside systems, Azure model hosting and validated Vera Rubin servers to make it simpler for companies to run agentic AI at scale.

Overview

Nvidia and Microsoft announced the integrated stack at Microsoft Build on Wednesday, June 3, 2026, combining hardware, runtimes, models and datacenter validation into a single developer platform.
Nvidia introduced RTX Spark Windows PCs that deliver about 1 petaflop of on‑device AI with up to 128GB unified memory and full offline capability, and a DGX Station for Windows deskside system that can run models up to roughly 1 trillion parameters.
Microsoft said it will add Nvidia’s open models — including Nemotron 3 Ultra, Nemotron 3.5 ASR, Cosmos 3 and Earth‑2 — to Azure Foundry and that third‑party models such as Anthropic’s Claude are running natively on Nvidia Blackwell infrastructure in Azure.
Nvidia confirmed full‑scale production of its Vera Rubin datacenter platform and Microsoft approved Rubin for Azure deployment, with Nvidia claiming up to 10x inference throughput per megawatt versus prior hardware and analysts expecting a Rubin ramp beginning in Q3.
Wall Street remains broadly bullish with a Strong Buy consensus and an average NVDA target near $309.94 while Goldman Sachs kept a $285 target, and analysts caution that supply, wafer capacity, Windows on ARM compatibility and device thermal or battery constraints are execution risks to watch.