Key Takeaways
- Energy costs are a major bottleneck: Electricity prices in Germany (34–35 cents/kWh) are dramatically higher than in Nordic countries (~2 cents/kWh), pushing companies to build AI infrastructure elsewhere.
- Containerized data centers eliminate the need for traditional buildings: Up to 90% of infrastructure is pre-fabricated in factories, drastically reducing on-site installation time versus the 24 months needed for a conventional data center.
- Liquid cooling and software optimization drive efficiency: Direct Liquid Cooling (DLC) extends hardware lifespan and reduces cooling costs, while deployment management software can save at least 20% in energy costs.
- Germany may accelerate by 2027: Despite current delays, projects like the Schwarz Group’s three green-powered data centers signal a potential turning point.
High Electricity Prices Stall German AI Infrastructure Expansion
Germany is falling behind in the global race to build AI infrastructure, and the reasons are as much economic as they are technical. Darren Cox, General Manager for Europe at KAYTUS, points to electricity prices as the primary culprit: in Germany, a kilowatt-hour costs between 34 and 35 cents, while the same kilowatt-hour can be obtained for around two cents in Nordic countries. This staggering disparity is leading many German companies to site their AI data centers in the Nordics rather than domestically.
“The German market has been lagging behind global development in the expansion of AI infrastructure for some time, mainly due to a lack of affordable energy.”
Despite this headwind, construction continues in Germany. The Schwarz Group, for instance, is currently building three data centers designed to run on green power and feature state-of-the-art energy efficiency. Depending on the location, these facilities are expected to come online within the next six to twelve months. Cox expects development in Germany to accelerate noticeably by 2027, driven by a growing number of similar projects.
Containerized Data Centers: Speed and Flexibility Without a Building
When asked how the demand for faster deployment can be met, Cox points to container-based AI data centers. In this model, up to 90% of the infrastructure—including networking, cooling, and power systems—is pre-fabricated in factories, which significantly shortens on-site installation time. A conventional data center can take up to 24 months to build; a containerized solution is delivered as a turnkey package.
“The customer doesn’t even need a building. Essentially, they just need electricity.”
This model allows infrastructure to be deployed at locations where electricity is available but a traditional data center building does not exist. A diesel generator may serve as a temporary backup, but the key requirement is simply an adequate power supply. This approach is particularly attractive for companies looking to stand up AI capacity quickly and avoid the long lead times and capital tie-ups associated with conventional construction.
Liquid Cooling and Software: Twin Pillars of Efficiency
Direct Liquid Cooling (DLC) plays a central role in KAYTUS’s strategy. Unlike traditional air cooling, DLC cools not only CPUs but also GPUs, memory, and other components. This delivers two benefits: lower cooling costs and longer hardware lifespans. Cox cites the old rule of thumb that every additional 10 degrees of temperature above the baseline can reduce a component’s operational lifespan by up to 30%. Keeping systems cooler translates directly into reliability and cost savings.
On the software side, deployment management platforms optimize GPU load balancing and power supply control. Cox reports that these optimizations yield energy savings of at least 20%, with an additional 10% reduction in power demand for liquid-cooled installations. At scales approaching one gigawatt, these percentages translate into substantial financial impact. The platform “MotusAI,” based on Kubernetes, supports deployment, multi-tenant billing, and management of heterogeneous AI infrastructure, and is open to any AI agent — it is not locked to a specific vendor.
Growing Convergence of HPC and AI
Cox also observes a continuing convergence of High Performance Computing (HPC) and AI workloads. Many classical HPC applications are already GPU-accelerated, and the trend points toward both workload types eventually running on the same infrastructure. This would eliminate the need for separate environments, streamlining operations and reducing costs for organizations that require both.
Looking ahead, Cox identifies flash memory prices as a persistent challenge driving infrastructure costs upward, though new semiconductor factories may ease the situation over time. Meanwhile, GPU innovation cycles have compressed dramatically—from two-to-three-year intervals to less than twelve months—requiring infrastructure and engineering teams to keep pace with an accelerating technology curve.
Conclusion: A Market in Transition
Germany’s AI infrastructure landscape is defined by tension: high energy costs push investment to the Nordics, while domestic projects like the Schwarz Group’s green-powered facilities and the government’s National Data Center Strategy signal a commitment to catching up. Containerized and liquid-cooled solutions from providers like KAYTUS offer a practical path to rapid deployment, bypassing the need for traditional buildings and lengthy construction timelines. Whether Germany can close the gap by 2027 will depend on its ability to make energy affordable and available at scale—and on how quickly companies adopt modular, pre-fabricated approaches to data center construction.
References
[^1]: Paula Breukel (2026-08-14). “„Der Kunde braucht nicht einmal ein Gebäude”“. DataCenter-Insider. Retrieved 2026-08-21.
[^2]: “Energieeffizienz im Rechenzentrum“. DataCenter-Insider. Retrieved 2026-08-21.
[^3]: Paula Breukel (2026-08-20). “Aus Serverwärme von Schwarz Digits wird Fernwärme für Lübbenau“. Dev-Insider. Retrieved 2026-08-21.
