NTT Global Data Centres has deployed an Nvidia Blackwell GPU cluster at its Munich 2 campus, a 6,300 sqm (67,813 sq ft) site near Unterschleißheim outside Munich that offers 14MW of critical IT load across two buildings.

NTT calls the deployment AI Factory Munich. It's positioned as something more than a demo lab: customers can run their own AI applications and data against the infrastructure, security, governance and operational conditions they'd face in production — before spending on a broad rollout of their own.

What does the cluster actually give customers?

The setup provides a "real-world environment" for testing how AI applications perform at scale, according to the company. That's a meaningful distinction for practitioners. A typical innovation lab proves a workload runs; this cluster tests whether it runs the way it will need to when security, governance and data requirements come into play.

The hardware stack breaks down like this:

• Nvidia Blackwell GPUs, interconnected via Nvidia Spectrum-X Ethernet
• Servers and storage from Dell's AI Factory
• Platform work from Red Hat and model work with Mistral AI

Anne-Sophie Lotgering, CEO Europe at NTT, confirmed the partner split in a LinkedIn post: Nvidia and Dell supplied the infrastructure, while Red Hat and Mistral AI worked with NTT on the platform and the model.

Why test before you buy?

For enterprises weighing six- and seven-figure GPU investments, the pitch is straightforward — de-risk the commitment. Graham Clarke, managing director for global private and sovereign AI at NTT Data, put it in terms any buyer will recognize.

"Every leader I speak to wants AI they own, control and trust. At NTT Data's AI factory Munich, they can run their own applications and data in a production data center, on their terms, before they commit to deployment," Clarke said. "It's the difference between a slide and a decision."

That framing lands with practitioners who've sat through AI pilots that looked fine in a lab and fell apart under production constraints. Running workloads inside an operating data center — with 14MW of capacity behind it — closes part of that gap before a euro is spent on dedicated infrastructure.

The sovereign angle

The location matters as much as the hardware. A Bavarian data center running customer data on customer terms feeds directly into European demand for private and sovereign AI — workloads that stay under the owner's control rather than a hyperscaler's. Clarke's portfolio title, global private and sovereign AI, signals where NTT sees the market heading.

For data center operators and enterprise AI teams, the move adds another large-scale test bed in the German market — one backed by Nvidia's current-generation Blackwell silicon and a Dell-Red Hat-Mistral software path. Whether it converts testers into deployment customers is the open question, but the facility gives them a place to find out with real numbers instead of slideware.