AMD and Rackspace Technology have signed a definitive agreement for the phased deployment of an initial 30 MW footprint of AMD-based AI compute across Rackspace's global data centers, cementing AMD as a strategic silicon-layer partner for Rackspace's governed enterprise AI stack. The deal, announced on June 16, 2026, operationalizes a memorandum of understanding the two companies unveiled on May 7, 2026.
What the agreement covers
Under the GPU-as-a-service agreement, Rackspace will dedicate, maintain and make available an aggregate of 30 MW of capacity across its footprint exclusively for AMD products, including AMD Instinct GPUs and AMD EPYC CPUs. Deployment begins in late 2026 and continues through 2028, giving regulated enterprises a governed path to accelerated compute without building their own AI infrastructure.
Why governed AI matters
Rackspace is positioning the capacity for regulated workloads where data residency, security and auditability are non-negotiable. The company said healthcare providers have already expressed early interest in accelerated compute for clinical AI and inference at scale, a segment where governed infrastructure can shorten the path from pilot to production.
A widening battle for AI infrastructure
The agreement extends the intensifying competition to supply the silicon and systems behind the AI build-out. AMD's Instinct accelerators are increasingly pitched as an alternative to incumbent GPUs, and locking in dedicated data-center capacity helps guarantee supply for enterprise customers. The move lands alongside a wave of chip-sector deal-making, from Qualcomm's acquisition of AI software firm Modular to record AI-memory demand reported by Micron and national efforts such as South Korea's AI chip drive.
At full deployment, 30 MW of dedicated AMD compute represents meaningful capacity for enterprise AI, and both companies framed the arrangement as a template for repeatable, governed accelerated-compute rollouts.
Reporting based on coverage from AMD and Rackspace Technology (GlobeNewswire), CNBC and HPCwire.