Conversion versus new construction of data centers for the use of AI

NDC-GARBE 01/10/2024

The introduction of applications and services that use artificial intelligence (AI) presents developers and data center operators with new challenges. Above all, they need an infrastructure that offers far more computing power for the solutions. This demand leads to the use of powerful, compact AI servers and confronts companies with the question: rebuild or build new data centers for AI applications?

However, the increase in packing density generates enormous heat. The capacity of traditional air cooling is no longer sufficient for the new load. Liquid cooling, which ensures more efficient heat dissipation, is therefore in demand. This development is forcing the industry to fundamentally rethink the design of data centers and adapt them as necessary. What exactly this means is explained below. The focus is on the question of when the conversion of individual data centers is more advisable than a new build.

New overall architecture of the buildings

There is currently no uniform standard for the design of AI data centers. But in future, developers and operators will have to take the integration of liquid cooling for the AI servers into account when planning the buildings. In most cases, it requires the installation of a separate water circuit, which has a significant impact on the overall architecture. In addition, there is another significant change: unlike before, IT infrastructure, cooling and supply technology will increasingly be closely interlinked. Furthermore, preparing for liquid cooling requires all rack configurations to be adapted.

Finding the right strategy for converting to AI-optimized data centers

The solution that meets all requirements does not lie exclusively in the construction of new data centers. Modernizing existing buildings is a possible alternative. If a separate water circuit already exists, implementation is quick and straightforward. Otherwise, the buildings must first be gutted and then structurally adapted. Compared to a new build, which often costs more time and money, this makes perfect sense. When converting, some of the permits are already available. This reduces the bureaucratic effort and speeds up the project.

However, if the initial conditions are not right, the only option is a new building. Above all, developers and operators of data centers must be able to make the necessary changes quickly and easily. An overall architecture in the style of a hall, for example, offers plenty of space for free design after gutting. Compared to a new building, the overall cost must remain within reasonable limits.

Future-proof infrastructure from NDC-GARBE

NDC-GARBE has been relying on the Green IT Cube for ten years now. The patented concept for energy-efficient and environmentally friendly high-performance computing (HPC) in data centers already meets the requirements for AI applications and services, including with regard to liquid cooling in the server room. Although the concept dates back to a time when AI was not yet a major topic, it is proving to be future-proof. The Green IT Cube therefore provides developers and operators of data centers with a suitable solution for the new requirements.

Over the next few years, the industry will face the major task of optimizing infrastructures for the growing AI workloads. As a data center development-as-a-service expert, NDC-GARBE is available to you at any time for consulting, design, concept development and the construction and approval process. Would you like to talk to experts about converting or building new data centers for AI applications or do you need comprehensive support in implementing an AI data center? Then please get in touch with the NDC-GARBE team.