AMD introduces Helios, a rack-scale AI architecture for large enterprises in EMEA

AMD has unveiled a new rack-scale AI architecture called Helios, specifically targeting large enterprises in Europe, the Middle East and Africa. The architecture is designed to meet the more demanding requirements that AI workloads place on data centres, requirements that traditional infrastructure struggles to handle.

Classic data centre setups are built on general-purpose compute. In such environments, compute, memory and networking are assembled step by step and optimised after the fact. For standard business applications that works well, but large-scale AI workloads, such as model training or extensive inference, require a different approach. The bandwidth, latency and memory capacity needed exceed what an incremental build-up can deliver.

With Helios, AMD opts for integration at the rack level: compute, memory and interconnects are designed and deployed as a single coherent system, rather than as separate components combined after the fact.

What Helios does differently from a technical standpoint

The core principle of the Helios architecture is that the rack is treated as the unit of design, not the individual server. By aligning interconnects, memory bandwidth and compute capacity at the rack level, AI workloads can be distributed more efficiently across the available hardware. This is relevant both for the training phase of large models and for inference at scale, where multiple requests are handled simultaneously.

Traditional architectures run into bottlenecks in node-to-node communication when handling these kinds of tasks. The latency and bandwidth between GPUs or AI accelerators are key factors in how quickly a model can train or respond. Helios aims to structurally reduce those bottlenecks by designing the hardware as a system from the outset.

AMD positions Helios explicitly as an enterprise solution, meaning it is aimed not primarily at hyperscalers building their own infrastructure, but at large organisations that want to run AI capacity in their own data centres or procure it through co-location facilities.

Why EMEA is the focus region

AMD is targeting the EMEA market specifically with Helios. Large European enterprises in sectors such as financial services, manufacturing and the public sector have been investing heavily in their own AI infrastructure in recent years. In part because they prefer not to store data in US cloud environments for regulatory or strategic reasons, and in part because the costs of cloud-based AI compute add up quickly under intensive use.

The European AI Act, which requires organisations to be transparent and maintain control over how AI systems operate and where data is processed, reinforces that preference for infrastructure under their own management. For AMD, this creates an opportunity: organisations that want to run AI workloads on-premise or in regional co-location facilities need infrastructure built for exactly that purpose.

Competitive position relative to Nvidia and Intel

In the AI infrastructure market, Nvidia remains the dominant player, largely due to its H100 and H200 GPUs and the associated CUDA software ecosystem. AMD is working to gain ground with its Instinct accelerators and the ROCm software stack, but adoption has been slower than the company had hoped. In that context, Helios also represents an ecosystem strategy: by offering a complete rack-scale system rather than individual GPUs, AMD aims to lower the barrier to entry for enterprise customers.

Intel is competing in the same space with its Gaudi accelerators, but faces similar challenges in taking market share from Nvidia. For enterprise customers looking to diversify or avoid full dependency on a single vendor, AMD and Intel at least offer an alternative.

Availability and partners

Based on available source material, AMD has not yet released detailed information on exact pricing, delivery partners in the Netherlands or Belgium, or a concrete rollout schedule for the EMEA region. The announcement of Helios appears at this stage to be focused on positioning the architecture with large enterprises and their infrastructure partners. Further details on availability are expected to be communicated through AMD's partner network and direct enterprise channels at a later stage.

Implications for the European AI infrastructure market

The introduction of Helios fits into a broader trend in which hardware manufacturers offer complete AI system solutions rather than individual components. For European enterprises looking to build their own AI capacity, this expands the range of integrated alternatives to cloud-based solutions.

For Dutch and European founders and technology companies that procure or resell AI infrastructure, it is relevant that competitive architectures are entering the market alongside Nvidia-based solutions. Greater supply in principle increases negotiating leverage and reduces dependence on a single vendor. Whether Helios actually achieves adoption at a European scale will depend heavily on how well AMD develops the software ecosystem around ROCm and which system integrators and co-location partners embrace the architecture.