AI has become a cornerstone of Europe’s technological competitiveness and strategic autonomy. From smart cities and industrial decarbonization to healthcare and financial services, AI is reshaping organizations and the broader European economy.
The European Commission’s AI Continent Action Plan and the Cloud and AI Development Act place advanced, sustainable data center infrastructure at the heart of this ambition, including the goal of at least tripling EU data center capacity within the next five to seven years to support frontier AI and sovereign cloud workloads.
As AI initiatives accelerate and scale across the EU, a critical question is how to design and operate next-generation data centers that meet Europe’s unique constraints and opportunities. AI infrastructure is fundamentally different from traditional IT: it operates at greater scale, higher density and with significantly increased energy demands, while also facing strict EU energy-efficiency and sustainability rules. AI workloads cannot simply be layered onto existing data center architectures; they require a new class of infrastructure designed from the ground up to deliver performance, efficiency and scalability, while complying with European requirements on renewable energy use, waste-heat reuse and transparent energy reporting.
Integrated rack-scale solutions
Traditionally, data centers have been built as collections of individual servers, which are optimized independently and connected into clusters. To handle the demands of scaling AI-ready infrastructure, the industry is increasingly exploring tightly integrated, rack-scale architectures in which compute, memory, networking and software are co-designed as a unified system. These platforms bring together general-purpose processors, AI accelerators, high-speed networking and an increasingly open software stack in a cohesive architecture designed specifically for AI workloads.
This level of integration means AI performance is no longer limited by compute alone. Increasingly, the bottleneck lies in how efficiently data can be moved across the system. As models grow in size and complexity, memory bandwidth, interconnect efficiency and system-level orchestration become critical factors. In short, moving data may consume more power than the computation itself. That is why rack-scale solutions are designed to optimize performance per watt and performance per byte moved, rather than raw performance alone. These metrics are essential to scalable AI.
At the same time, compute density is increasing dramatically. In certain AI workloads, a single rack can deliver levels of AI performance that previously required multiple server clusters. This shift is transforming the data center itself, with the rack becoming the fundamental building block of AI infrastructure. By integrating processors, accelerators, networking and software at the rack level, rack-scale architecture enables organizations to deploy AI infrastructure more efficiently, scale more predictably and optimize performance across the entire system. This approach also addresses one of the most pressing challenges for AI today: energy.
Optimising AI architecture
Energy availability is rapidly becoming a potential constraint in AI data center deployment. As facilities scale towards gigawatt levels, digital initiatives and physical infrastructure must be developed alongside national and regional energy systems. At the same time, compute density is increasing dramatically, requiring new approaches to cooling, power delivery and system architecture.
This is where architectural balance becomes essential. AI performance is no longer defined by a single component, such as an accelerator. Instead, it depends on the seamless integration of processors, accelerators, networking, memory and software. Meeting the power and performance demands of AI depends on assigning the right compute engine to the right task, whether that means a CPU, GPU, specialized accelerator or a combination of these working together in a balanced system. GPUs and other highly parallel accelerators play a critical role in workloads such as training large language models and high-throughput AI inference at scale. CPUs remain essential for orchestrating data movement, managing workflows and handling general-purpose and latency-sensitive tasks.
Equally important is energy efficiency at the component level. Performance per watt and performance per byte moved are becoming defining metrics for power-optimized AI solutions. Hardware manufacturers such as AMD are responding with more efficient processors, accelerators, memory systems and interconnects, while system designers are optimizing the complete platform rather than focusing on any single component. This enables organizations to scale AI while limiting the impact on energy demand, cooling requirements and total cost of operation.
Towards an open AI future
When it comes to AI acceleration, the future is open. Developers need software ecosystems such as AMD ROCm™ that provide accessible tools, libraries, frameworks and performance-optimization capabilities, while allowing them to choose hardware according to workload, availability, cost and strategic requirements. Open standards and portable software can help reduce lock-in and keep infrastructure investments adaptable as technologies evolve.
To realize its true potential, AI must become pervasive from the data center and cloud to the network edge and endpoint devices. We continue to see rapid progress in generative AI and agentic AI, but there is also physical AI, bringing real-time intelligence to real-world systems. AI is moving beyond digital environments and becoming embedded directly into machines and end-user devices, from AI PCs to industrial systems and robots. To enable this future, AI infrastructure must be designed not only for scale, but also for distribution, making open software ecosystems, interoperable interfaces and efficient deployment models increasingly important.
The question increasingly centers on how AI will transform industries and economies, and how we build the infrastructure to support AI at scale while balancing performance, flexibility and compute- and energy-efficiency requirements. For Europe, this represents a significant opportunity: to develop AI infrastructure that is powerful and sustainable, while remaining open, resilient and adaptable to a changing technology landscape.
This is a contributed article by Rob de Bruin, Head of Channel Benelux AMD. For more information about AMD, you can go here.