A Guth Labs publication ↗
Back to News

Rackspace plans roughly 2 MW of CPU capacity for Blacksmith’s build-and-test workloads

Rackspace and Blacksmith announced a multiyear agreement on November 10 for dedicated AMD EPYC compute and storage representing approximately 2 MW of capacity in Northern Virginia. Deployment is expected to begin in early 2027. The contract targets the build-and-test workloads behind software development, rather than a new GPU training cluster. Rackspace announcement

The companies connect the expansion to increased use of AI coding tools: generating changes faster can increase the amount of code that needs to be built and checked before it is merged. Blacksmith intends to use Rackspace’s data-center and infrastructure operations while continuing to focus on its continuous-integration platform.

The announcement is a forward-looking capacity commitment. It does not establish that the full 2 MW is operating today or provide an independent benchmark for the future deployment. Rackspace identifies delivery, component procurement, financing and customer demand among the factors that could affect the plan.

Blacksmith’s current product website helps explain the underlying workload. It offers infrastructure for running GitHub Actions, with caching and retained Docker layers intended to reduce repeated work between builds. The company also describes tools for finding slow or failing jobs, searching continuous-integration logs and showing failed-test logs alongside pull requests. Blacksmith product overview

Those features illustrate why compute alone does not determine how quickly a change reaches production. Build pipelines also depend on data movement, cache reuse, test behavior and developers’ ability to understand failures. Blacksmith’s advertised speed and cost improvements remain vendor claims; they should not be treated as measured results for infrastructure that has yet to deploy.

The deal offers a useful view of AI-related infrastructure demand outside model execution. As an editorial interpretation, more automated code creation can shift pressure downstream into conventional CPU-heavy tasks. That does not prove that every engineering team needs more hardware, or that faster test execution guarantees safer software.

For readers tracking capacity announcements, the important distinctions are the workload, the approximate scale and the start date: build-and-test services, around 2 MW, and an expected early-2027 deployment. Actual utilization and customer results will be more informative than treating the announcement as capacity already delivered.

Sources and citations

Revision history

Published .