Summary
- Paris and Singapore join existing AI factory locations in Texas and Riyadh.
- The deployments complement a separately opened Munich computing environment.
- NTT has not disclosed installed IT power or cooling specifications for Paris.
NTT DATA has added GPU computing infrastructure in Paris and Singapore to its international AI Factory programme, extending the locations where customers can test workload performance and operational requirements before larger production deployments.
The two environments join facilities in Plano, Texas, and Riyadh, Saudi Arabia, forming a four-location programme across major regions. They also complement an earlier Nvidia GPU installation opened at NTT DATA’s Munich 2 data centre during October.
The facilities bring together bringing together computing hardware, networking, software, security and operational services. Customers can test workload behaviour under representative infrastructure conditions and assess how an intended deployment might perform as its scale changes.
The Paris installation gives European organisations another location for evaluating accelerator workloads. The company has not disclosed the number of GPUs deployed, installed IT power, site electrical connection capacity or cooling configuration. Its opening cannot therefore be used to calculate a new European megawatt total.
The distinction between operating test infrastructure and future production capacity is important. A customer may use a GPU environment to assess software and hardware requirements without yet committing to a dedicated data centre installation or a long-term computing contract.
Testing network and operating constraints
AI workloads have different infrastructure demands depending on whether they train models, serve inference requests or process data. Memory capacity, interconnect throughput and storage performance can determine whether accelerator systems spend their time computing or waiting for information.
Training large models involves frequent data exchange between processors. If the network cannot maintain the required throughput, adding more accelerators may deliver less benefit than expected. Representative testing helps identify such bottlenecks before a company invests in a larger cluster.
Inference workloads can have different latency and utilisation characteristics. The chosen hardware and the location of an application in relation to users and data stores can influence operating costs and response performance.
The environments allow customers to examine architecture, performance, workload placement and cost. The assessments also cover security and governance requirements, which become more difficult to change once systems are running in production.
Power and cooling requirements follow the hardware selected. High-density GPU racks can require substantial electrical distribution and thermal management infrastructure. However, test results achieved in one installation do not guarantee that a much larger deployment will have identical efficiency or performance.
Paris extends a separate European presence
The company’s Munich facility opened earlier in October and provides a distinct European testing location. Paris adds geographic choice rather than replacing the German installation. The two deployments should not be described as a single campus or as evidence of an undisclosed European capacity programme.
Location may influence workload placement, network routes and data handling requirements. A European facility can support local testing, although contractual arrangements, administrative controls and the design of connected services still determine where information travels.
Enterprise customers also need to understand how systems are monitored, supported and recovered after faults. Hardware availability is only one element of resilience: operational procedures, network design, backup arrangements and security controls determine service continuity.
Available company information does not establish whether the Paris environment has new dedicated cooling or electrical plant, or whether existing infrastructure accommodates the GPU installation. Such information would be required to compare its physical impact with the earlier Munich deployment.
With Paris and Singapore now included, the programme provides a wider set of locations for evaluating computing configurations. The confirmed development is available testing infrastructure; subsequent production capacity will depend on customer demand, project-level investment and the supporting electrical and mechanical systems.
