Nebul has chosen Datacenter United’s data center in Machelen for the expansion of its new AI SuperCluster. The planned deployment includes more than 1,000 NVIDIA GB300 GPUs, with server infrastructure from Lenovo. This gives concrete technological substance to the previously announced multi-megawatt agreement in Machelen. The project demonstrates what is needed to support large-scale AI computing power: from high-density power distribution and advanced liquid cooling to reliable connectivity and operational continuity.
From announced capacity to a concrete AI deployment
In September, Datacenter United announced a binding agreement for several megawatts of AI capacity in Machelen. With the announcement that Nebul is a client, it is now also clear what technological application lies behind this infrastructure development.
Nebul, a European provider of AI and cloud services, is expanding its infrastructure with a new SuperCluster in Belgium. More than 1,000 NVIDIA GB300 GPUs will collectively provide the computing power required for intensive AI workloads, including the training and deployment of advanced AI models.
The scale of this deployment makes it immediately clear that large-scale AI requires more than just powerful processors. The physical infrastructure must also be designed to reliably support such systems.
For Datacenter United, this means developing a specialised high-density environment in which electrical power, liquid cooling and connectivity are closely coordinated.
Why the NVIDIA GB300 places different demands on data centre infrastructure
The NVIDIA GB300 is part of a new generation of AI technology that concentrates an exceptional amount of computing power within a limited physical space.
One example of this is the NVIDIA GB300 NVL72 reference architecture. It combines 72 Blackwell Ultra GPUs and 36 Grace CPUs within a single rack system. Via NVIDIA NVLink, the GPUs can exchange data at high speed and collaborate on complex AI computations.
That concentration of computing power presents specific infrastructure challenges.
According to NVIDIA’s technical reference architecture, a fully loaded GB300 NVL72 rack can require up to approximately 142 kW of electrical power. This is considerably higher than in many traditional server environments.
This is not solely a matter of ensuring sufficient power is available. The power must be distributed safely, whilst the heat generated by the systems must be continuously dissipated. Floor load, available space, monitoring and operational continuity also play a role. As power density increases, these factors become increasingly interdependent.
The challenge, therefore, is not just to accommodate more computing power, but to tailor the entire infrastructure to its technical requirements.
Liquid cooling as an essential component of high-density AI
One of the main challenges with modern AI hardware is the amount of heat generated within a confined space.
Traditional data centre environments rely largely on air cooling. However, at very high power densities, it is becoming increasingly difficult to dissipate the entire heat load through air circulation alone. That is why modern AI systems are increasingly making use of liquid cooling.
In direct-to-chip liquid cooling, coolant is channelled via heat spreaders past heat-generating components such as GPUs and CPUs. The absorbed heat is then transferred via a fluid circuit to the underlying cooling infrastructure. This allows a large proportion of the heat to be dissipated closer to the source, which helps to support high power densities.
For Nebul’s SuperCluster, Datacenter United is developing a specialised high-density environment in Machelen with advanced liquid-cooling capabilities. This development goes beyond simply installing additional cooling technology. Electrical systems, heat dissipation and operational processes must function as a single integrated system and be tailored to the requirements of the AI hardware.
Energy efficiency also remains important in this context. Higher power densities allow for more intensive use of the available space, but at the same time place greater demands on the design, monitoring and management of the infrastructure.
The technology behind Nebul’s SuperCluster
| AI cloud platform | Nebul |
| GPU technology | NVIDIA GB300 Blackwell Ultra |
| Planned deployment | More than 1,000 GPUs |
| Server infrastructure | Lenovo |
| Data centre location | DC Machelen, Belgium |
| Infrastructure partner | Datacenter United |
| Infrastructure development | High-density, advanced liquid cooling and carrier-neutral connectivity |
Technical context: the NVIDIA GB300 NVL72 reference architecture combines 72 Blackwell Ultra GPUs and 36 Grace CPUs within a single liquid-cooled rack system, with a maximum power requirement of approximately 142 kW per rack. These specifications illustrate the technical requirements of the technology. They do not necessarily describe the exact hardware configuration that Nebul will install in Machelen.
Why Machelen? Capacity, connectivity and continuity
In addition to computing power and cooling, location, connectivity and operational reliability are also key factors in the development of large-scale AI infrastructure.
DC Machelen is a power-intensive data centre site near Brussels, with capacity for larger infrastructure projects. Datacenter United is investing in the further development of the cooling infrastructure there to support specialised high-density applications.
Connectivity also plays an important role.
Within an AI SuperCluster, GPUs and servers must be able to exchange data with one another at very high speeds. This internal communication is supported by specialised network technology. In addition, the AI cloud platform requires reliable external connections to customers, other networks and cloud environments.
This is where Datacenter United’s carrier-neutral infrastructure plays a key role. It enables connections with various network providers to be organised flexibly, tailored to the customer’s technical and operational requirements. Continuity remains a fundamental prerequisite in this regard. Power supply, cooling and connectivity must function reliably, supported by monitoring, maintenance and appropriate operational facilities.
Four partners, complementary expertise
The development of Nebul’s SuperCluster brings together various specialised technologies and layers of infrastructure.
- Nebul develops and operates the AI cloud platform that gives European organisations access to specialised computing power.
- NVIDIA supplies the GPU technology that enables the processing of intensive AI workloads.
- Lenovo provides the server infrastructure into which this technology is integrated.
- Datacenter United develops and manages the physical data centre environment in which the systems are housed, powered, cooled and connected.
This collaboration makes it clear that the performance of modern AI systems is not determined solely by the processors. The various components of the technological infrastructure must be closely integrated in order to reliably deliver the intended computing power at scale.
A concrete step towards AI infrastructure in Belgium
Nebul’s decision to choose Machelen means that the growing demand for specialised AI infrastructure is being met on Belgian soil.
This collaboration demonstrates how international AI cloud providers may opt for Belgian data centre infrastructure when the necessary conditions regarding power, cooling, connectivity and scalability are in place or are being developed.
For Datacenter United, the project represents a further step in realising its investments in high-density infrastructure. It therefore not only expands the available AI computing capacity in Europe, but also serves as a concrete example of how specialised data centre infrastructure enables new technological applications in Belgium.
Want to find out more about the infrastructure behind AI?
Large-scale AI applications place new demands on data centre infrastructure. Find out how Datacenter United supports organisations with specialised solutions for high-density computing.
Discover our AI and high-density infrastructure
Does your organisation have any specific infrastructure requirements? If so, please get in touch with our team here.