The next AI pilot has to prove operations and jurisdiction. NTT DATA opens a production test floor in Munich
NTT DATA is opening an AI Factory in Munich where organizations can validate AI architectures on production-grade infrastructure before committing. It shows what a serious pilot now has to prove.
Listen to this article · 4 min
AI-generated narration of the full article.
There is a meeting that happens in almost every enterprise AI program. The pilot went well: the model answered questions accurately, the demo impressed the steering committee. Then the infrastructure team asks what it will cost at full load, the security team asks where the data will sit, and someone from risk asks how the governance controls were tested. The answers are some version of “we haven’t tried that yet.”
The pilot proved the wrong thing. It proved the model works, which was rarely the open question.
A facility built around that gap
On October 7, NTT DATA announced the opening of an AI Factory in Munich aimed at exactly that meeting. NTT DATA describes it as the first facility of its kind in its global network: a production-grade environment where AI solutions can be designed, tested and validated before deployment. According to its own description, a visit includes a tour of the data center environment, with GPU compute clusters, enterprise storage and networking, and real-time visibility into utilization, job queues and performance; a walkthrough of MLOps orchestration, training and fine-tuning environments, deployment pipelines and AI governance tooling; live use cases in sectors including banking, automotive and healthcare; and sovereign and private AI architectures. Each engagement, by request and tailored, ends with a signed benchmark report, an architecture recommendation and agreed next steps.
The facility is in Munich, and NTT DATA has not announced an equivalent in other regions. It is also a sales and advisory tool and should be read as one. What makes it interesting is the demand it answers.
What each kind of test can prove
DemoProduction
- Laptop demoThe model answers well
- Vendor cloud sandboxQuality on your task
- Production-grade test floorLoad, cost, operations, governance
- Own productionResidency, jurisdiction, your team running it
Four questions a demo never answers
Does it run at our scale? Throughput, latency and cost under realistic load, on the hardware class you would actually buy. We argued that private AI is now sized by workload, not GPU count; a pilot should produce the workload numbers.
Can we operate it? Model updates, monitoring, incident response and capacity planning, run by your team rather than the vendor’s experts.
Does governance work in practice? Access control, logging, evaluation and approval flows exercised end to end, not described on a slide.
Does it satisfy our jurisdiction? Where data and models reside, who can reach them and under which law. In Europe that means data residency and, for many uses, the EU AI Act; we covered how sovereignty now reaches the data center floor.
Using a test floor without being sold by it
Bring your own success criteria: a vendor-designed benchmark proves what the vendor chose to measure. Read a signed benchmark report as evidence for the procurement file, after checking that the configuration tested matches the one you would buy. And have someone independent review the architecture recommendation, since it comes from a provider that also sells the implementation.
The bar for an AI proof of concept is moving from “the model works” to “the system works here, under our rules.” Whether the test runs in Munich, in a partner’s lab or on your own cluster, the pilot that matters is the one that answers the questions from that meeting before anyone has to ask them.