Rackspace’s AMD GPU-as-a-Service framework sets a conditional 30-megawatt footprint but obligates AMD to no service quantity or deployment.

Rackspace Technology, listed on Nasdaq under RXT, disclosed a definitive GPU-as-a-Service master agreement with AMD. The agreement creates a framework for phased deployments of AMD Instinct accelerators—including MI355X, MI350P, and future successors—and AMD EPYC central processing units in Rackspace data centers.

Rackspace agreed to dedicate, maintain, and make available an aggregate initial footprint of 30 megawatts, but only where AMD products are fit for purpose and financing, operational, and legal conditions are satisfied. Rackspace said fulfilling that obligation would require substantial capital expenditure, financing, customer demand, power, cooling, networking, and operational execution, none of which is assured on the anticipated schedule or at the anticipated cost.

The master agreement does not require AMD to buy any quantity of services, approve a deployment, or enter future commercial arrangements. Every deployment requires separate agreement on pricing, term, and financial parameters, and AMD has no obligation to accept any particular deployment under the framework.

If a deployment is approved, AMD agreed to purchase residual unsold capacity subject to delivery and service-availability requirements and a deployment-specific aggregate cap. AMD also received a right of first refusal before Rackspace sells GPU-as-a-Service capacity to third parties below a specified price threshold; failure by Rackspace to satisfy its footprint obligation could result in the loss of certain agreement benefits.