A Faster Build Queue Still Needs a Release Decision
Rackspace and Blacksmith plan to add capacity for building and testing code. Software teams can use faster feedback to make better release decisions if they measure the wait that capacity actually removes.

Inside this article
01Infrastructure targets a specific wait ↓02Track the first result someone can use ↓03Check how much work survives to matter ↓Event / source date: 10 November 2026 · Rackspace and Blacksmith agreement announcement. Deployment was expected to begin in early 2027. · Published 11 NOVEMBER 2026
Infrastructure targets a specific wait
On 10 November, Rackspace and Blacksmith announced a multiyear agreement for dedicated computing and storage in Northern Virginia. Rackspace says it plans to supply AMD EPYC capacity representing approximately two megawatts, with deployment expected to start in early 2027. Blacksmith supplies infrastructure for building and testing code. The companies argue that AI-assisted code production is increasing pressure on this stage. Their announcement describes planned capacity. It provides a reason for software teams to examine their own queues before assuming that faster machines will produce earlier releases.
Continuous integration, or CI, runs checks when code changes are proposed. A proposed change can wait for a runner, occupy it during setup and testing, and then wait for someone to act on the result. Those intervals have different owners. The value of additional capacity depends on which interval is consuming the useful time.
Track the first result someone can use
Imagine a hypothetical team that submits a change at 10:00. An eligible runner becomes available at 10:12. A test fails at 10:19 and points to an actual regression. At 10:31 the engineer submits a correction, which is checked and reviewed before release. Reducing the original queue to one minute would make the diagnosis available earlier. The size of the subsequent release improvement would still depend on when the engineer can respond and whether the reviewers are ready.
This team could record submission, runner start, first actionable result and approval as separate events. It could separate ordinary test failures from failures in the infrastructure itself. A test failure with a reproducible case directs a repair. A machine that disappears before it produces output gives the engineer a different problem. Combining them in a single failure count hides the spending decision.
Check how much work survives to matter
A new commit can supersede a previous one while checks are still queued. If a team keeps executing every obsolete check, some of the apparent demand will belong to versions nobody intends to release. Capacity can be made more useful by deciding which work may be cancelled and which results must be retained. A cancelled check should remain distinguishable from a pass, especially where policy requires a particular verification before merging.
The team can also separate short feedback checks from a long test that a release actually requires. If both jobs compete for the same runners and both have the same scheduling rules, an otherwise fast check may wait behind a large workload. A useful benchmark gives the two classes separate wait distributions. It also records the exceptions that make a shared pool preferable, such as rare jobs with special equipment requirements.
Purchase a measured change in service
The announced power figure describes a planned infrastructure commitment. A customer needs a service expectation expressed in terms of its work. For example, what share of important builds can receive an eligible runner within a chosen interval at the team's busiest hour? Which workloads can use the new capacity? What happens when an image fetch, dependency server or storage operation rather than the CPU becomes the delay?
In a hypothetical evaluation, a team could select two similar repositories, establish their normal queue and runtime patterns, and move one first. It would track the upper end of the wait distribution as well as the median. That offers a way to see whether the improvement reaches the developers who previously waited longest. It would also examine the cost per useful result, including retries, abandoned runs and the time spent investigating infrastructure failures.
Preserve the release decision as demand rises
A build that passes the tests provides evidence about the checks that ran. The authority to expose a change to customers can depend on a separate question about the deployment environment. Suppose a proposed migration passes a test database exercise. The release owner may still need to know whether customer writes can continue during the change and what recovery means if the schema begins to diverge. CYPAPER's analysis of software rollback after data changes examines that separate responsibility.
Faster queues can help here by delivering more time before the release deadline. The team could spend part of that time on a realistic rehearsal or a second look at a surprising result. If the saved time is immediately filled with additional changes, review can become the next overloaded stage. This is the concrete software-delivery version of measuring the whole AI workflow. The useful output is an authorized change that can be operated and recovered.
Assign an owner to the result
Before changing infrastructure, agree on which queue problem the purchase should fix. A platform engineer can own runner eligibility and scheduling. A service team can own test relevance and recurring flakiness. A release owner can explain which checks are required and what evidence is needed for the final decision. If the change in capacity helps one measure while leaving another stalled, this division makes the next intervention clear.
The Blacksmith agreement shows a supplier making a planned commitment to the build and test stage. An engineering team should translate that announcement into its own measurable requirement. It can then see whether it bought faster diagnostics, shorter queues or more headroom for difficult checks, and judge release speed using the complete path from proposed change to live service.
Gabriel Nicolaev — CyGuru
Sources & further reading
Rackspace announcement of the Blacksmith agreement ↗
Hands on a keyboard in front of several monitors showing code in a dark workspace. Illustrative photograph. Photo by Jakub Żerdzicki on Unsplash.
Another angle on this subject
Continue through related ideas, decisions and consequences.
Human BehaviourUseful Feedback Gives the Receiver a Decision They Can Make
Feedback becomes actionable when it identifies an observed issue, explains its consequence and leaves a realistic route to a better result.
Money & BusinessA Rate Increase Reaches a Business Through Its Contract Calendar
The Bank of England raised Bank Rate to 4 per cent. A business still needs to identify which commitments change, when they reset and what cash decisions follow.
Money & BusinessA Business Succession Plan Must Transfer Judgment, Not Just a Title
An operating essay on succession, concentrated knowledge and the decisions a business needs to make when its most experienced person steps away.
