AMD recently introduced the “Helios” rack-scale AI architecture, which changes the way large enterprises approach infrastructure readiness compared with traditional data center models. Traditional data center architectures were built for general-purpose computing, where compute, memory, and networking are assembled and optimized incrementally over time. AI at scale changes those requirements: training and large-scale inference demand tightly integrated systems that balance accelerators, memory bandwidth, CPUs, and networking from the outset to run efficiently. “Helios” reflects that shift by delivering rack-level integration from day one, bringing together AMD Instinct™ GPUs, AMD EPYC™ CPUs, AMD Pensando™ networking, and AMD ROCm™ software into a purpose-built AI system rather than a collection of discrete components.
For large enterprises in EMEA, this model changes infrastructure readiness. Instead of stitching together bespoke architectures, “Helios” enables deployment of a consistent, scalable, open rack-scale platform that is production-ready from the start. It provides a foundation that can scale from single racks to multi-rack clusters while avoiding proprietary lock-in, supporting AI deployments built for production, not just pilots. This becomes even more relevant as the EU plans to set up AI gigafactories, where deployment speed and repeatable system integration matter as much as raw compute. In environments measured in hundreds of megawatts and industrial-scale clusters, a rack-scale architecture like “Helios” helps reduce integration time by starting with a balanced system design.
Flexibility and Open Standards
AMD has consistently emphasized open standards and rack-scale system design, and this approach helps EMEA enterprises maintain flexibility and avoid lock-in as they expand AI infrastructure at scale. AI is evolving rapidly, and models, frameworks, and enterprise requirements continue to change, which makes rigid infrastructure a long-term risk. That is why a focus on open standards is central to our approach. Through the AMD ROCm open software platform, enterprises can run widely adopted frameworks such as PyTorch and TensorFlow, giving them the freedom to use the models and tools that best fit their needs as workloads evolve.
At the system level, rack-scale design reinforces that flexibility as deployments grow. With “Helios,” expansion is based on extending a consistent, open system foundation instead of re-architecting infrastructure at each new phase. This simplifies deployment, accelerates time to scale, and preserves interoperability across vendors and environments. The EU’s gigafactory plans also signal the scale enterprises are planning for, with expectations of roughly 100,000 AI chips per gigafactory. As that level of capacity comes online, the ability to scale consistently from rack to multi-rack deployments, without re-architecting the platform, becomes a practical advantage.
Sovereign AI Independence
As EMEA expands AI-ready data center capacity, system-level architecture becomes as important as raw megawatts in enabling sovereign AI capabilities and long-term infrastructure independence. AI capability is not defined by capacity alone, but by how that capacity is designed, integrated, and evolved over time. Large-scale AI infrastructure must support scalability, operational control, and long-term flexibility so that models, frameworks, and workloads can change without forcing repeated architectural resets. This is where open, rack-scale system design plays a critical role.
By building on open standards and delivering tightly integrated rack-level architectures, “Helios” enables long-term optionality rather than dependence on a single proprietary ecosystem. It applies system-level thinking to AI infrastructure, supporting platforms designed to scale consistently, operate efficiently, and remain adaptable over time.
Talent and Ecosystem Growth
As AI moves from pilot programs into operational deployment, EMEA’s competitive advantage is increasingly defined by the depth and scalability of its talent ecosystem, alongside growing AI-ready infrastructure. EMEA has a large, future-ready pool of engineers, researchers, and developers, supported by a strong open-source culture and increasing alignment between industry, academia, and national AI initiatives. As enterprises move from experimentation to production, the countries that lead will be those that can combine technical talent with platforms that allow skills, models, and innovation to scale reliably.
AMD is positioning itself to support that trajectory by investing not only in rack-scale AI infrastructure like “Helios,” but also in EMEA’s AI workforce and developer ecosystem. Through initiatives such as providing 100,000 GPU hours on the AMD Developer Cloud to researchers, academia, and startups, hosting large-scale hands-on workshops and hackathons, and building curriculum and Centers of Excellence through the AMD AI Engage platform, AMD is helping ensure that talent has access to real compute, open tools, and production-grade environments. With dozens of gigafactory proposals spanning many potential locations and operators, portability and repeatability become essential product requirements. “Helios” and AMD ROCm are designed to provide a consistent rack-scale foundation and open software environment that can be deployed across different data center designs as programs scale. Together with open standards and scalable system architectures, this approach helps innovations developed in EMEA move from research and pilots into operational AI deployments at enterprise and national scale.
This article was submitted by AMD.