IN Brief:
- AMD and Rackspace Technology have signed a definitive agreement for 30MW of AMD AI compute deployment.
- The rollout will use AMD Instinct GPUs and AMD EPYC CPUs across Rackspace’s infrastructure footprint.
- Enterprise AI infrastructure is placing new pressure on processors, power delivery, data-centre capacity, and workload governance.
AMD and Rackspace Technology have signed a definitive agreement for a phased deployment of 30MW of dedicated AMD AI compute.
The infrastructure will use AMD Instinct GPUs alongside AMD EPYC CPUs across Rackspace’s data-centre footprint. Deployment is expected to begin in late 2026 and continue through 2028, forming part of Rackspace’s Enterprise AI Cloud architecture for regulated and governed enterprise workloads.
The agreement builds on a memorandum of understanding announced in May and gives Rackspace a defined hardware platform for its AI cloud strategy. At full deployment, the compute footprint will provide dedicated AMD capacity for customers running controlled inference, clinical AI, and operationally accountable AI services.
The processor decision is only one part of the system. AI infrastructure requires accelerators, CPUs, memory, networking, power delivery, cooling, orchestration software, monitoring, and compliance controls to operate as one environment. A 30MW deployment is therefore a hardware, facilities, and workload-management commitment rather than a simple server procurement exercise.
AMD’s Instinct GPU line targets accelerated AI and high-performance workloads, while EPYC CPUs provide general-purpose compute for orchestration, data processing, smaller models, embeddings, retrieval, and supporting services. In production enterprise AI, the balance between accelerator and CPU resources affects latency, cost, memory use, and governance.
Rackspace’s focus on governed AI infrastructure reflects the move from experimental AI pilots toward production services. Enterprises are asking for systems that support access controls, data management, workload isolation, compliance, auditability, availability, and cost predictability. The physical compute platform has to serve those requirements rather than simply provide peak training or inference numbers.
The same upstream pressure can be seen in CSA Catapult’s move to become the UK’s Semiconductor Catapult, where AI hardware deployment, chip validation, and commercialisation are being drawn into national capability planning. AI compute is no longer only a hyperscale issue; it is becoming an industrial infrastructure question shaped by silicon access, verification, deployment capacity, and energy availability.
Europe’s wider debate around chips and AI sovereignty also links semiconductors, cloud services, data centres, and energy digitalisation into one infrastructure agenda. The AMD and Rackspace agreement is a commercial deployment rather than a sovereignty programme, but it reflects the same dependency chain between processors, power, capacity, and trust.
Power and cooling are now decisive constraints. A 30MW AI compute rollout requires electrical infrastructure, thermal design, airflow or liquid-cooling strategy, power conversion, switchgear, backup systems, and monitoring. Accelerator availability alone cannot deliver AI capacity if facilities cannot provide and remove energy safely and efficiently.
The manufacturing impact reaches well beyond processor vendors. AI server-board demand is already lifting requirements for dense PCB assemblies, advanced placement, high-current power distribution, connectors, memory, thermal interfaces, and test capacity. ASMPT’s comments on AI server-board demand illustrate how accelerator deployments are flowing back into electronics manufacturing equipment and SMT capacity.
Enterprise AI procurement is also changing. Many organisations are not looking only for raw accelerator access; they need managed environments that can handle regulated data, operational controls, service commitments, and integration with existing IT estates. Providers able to combine hardware capacity with governance may hold an advantage as AI systems move from experimentation into routine operations.
The agreement strengthens AMD’s role in AI infrastructure built from mixed CPU and GPU resources, while giving Rackspace a clearer hardware foundation for its managed enterprise AI strategy. It also reinforces a wider shift in electronics: processors, power systems, data-centre engineering, and governance are now tightly connected parts of the same infrastructure market.
