Latest / The Cloud Business Podcast with Fexingo: AWS, Azure, GCP, and Enterprise Infrastructure / Why Cloud Contracts Now Meter Your DevOps Pipeline Minutes
Transcript
- Lucas: If these conversations are useful for what you're building or running, there's a simple reason we keep them ad-free. We'll come back to that in a minute. Luna: Alright, so where are we starting today? Lucas: Let's talk about a quiet change that went into effect on July 1st. AWS CodePipeline started metering pipeline execution minutes. Not just for the compute that the pipeline spins up, but for the pipeline orchestration itself. Luna: Wait — they were charging for the pipeline runner before? I thought that was included in the free tier. Lucas: It was. For years, the pipeline orchestration — the DAG, the state machine that coordinates your build, test, and deploy stages — was effectively free. You paid for the underlying build compute and the artifacts, but not for the pipeline minutes themselves. Luna: And now they're charging per minute for that orchestration layer. Lucas: Exactly. The new pricing is $0.002 per pipeline execution minute after the first 100 free minutes per account per month. That's per account, not per pipeline. So if you have fifty pipelines all running, they share that 100-minute free pool. Luna: That's not huge money on its own. A hundred minutes of free pipeline time per month — for a team running, say, two thousand minutes of pipelines, that's an extra $3.80. Lucas: Right, but it signals something bigger. The hyperscalers have been methodically closing every free loophole in the developer workflow. Container image pull frequency, secret rotation, internal DNS queries — we've covered most of them on this show. Luna: And now the pipeline orchestration itself. Lucas: It's the last mile of the 'free developer experience'. The pitch was always: we make the tooling free so you build on our cloud. But once you're locked into the ecosystem, those free tools become cost centers. Luna: I saw the announcement — they framed it as 'aligning pricing with value'. The exact line was 'to ensure that the cost of pipeline management is fairly distributed based on usage.' Lucas: Which is cloud-provider speak for 'we need another revenue stream.' And look, I don't blame them. AWS, Azure, and Google Cloud are all in a CapEx-heavy cycle right now, especially around GPU clusters for AI workloads. They need to juice every dollar of revenue from their existing services. Luna: So the pipeline charge is basically a small tax on every developer team that's already standardized on AWS. Lucas: And it's not just AWS. Azure DevOps has been testing a similar model in preview since May. Google Cloud Build hasn't announced anything yet, but I'd be surprised if they don't follow within six months. Luna: We deliberately don't run ads on these episodes. If you want to support that choice, the link is buy me a coffee dot com slash fexingo. Lucas: And we keep it that way because we think these conversations — about what's actually changing in infrastructure — are worth having without a sponsor's filter. Luna: Yeah, exactly. So back to the pipeline charge — what does this mean for an engineering team that's not paying close attention? Lucas: It means that the free tier gets eaten up fast. The average mid-market engineering team runs about 1,400 pipeline executions per month. That's from a CNCF survey published in June. Luna: And those executions average what — ten minutes each? Lucas: The survey said the median execution time was about eight minutes. So that's roughly 11,200 pipeline minutes per month. Subtract the 100 free minutes, and you're paying for 11,100 minutes. Luna: At $0.002 per minute, that's $22.20 per month. Not nothing, but not a budget-breaker. Lucas: Unless you're an enterprise with a hundred accounts. Then it's $2,220 per month on pipeline orchestration alone. And that's before you factor in the compute cost for the build runners, which is separate. Luna: So the charge itself is small, but it adds up across accounts. And it changes the optimization incentives. Lucas: That's the key point. Before, you optimized pipeline runtime to make developers faster. Now you also optimize it to reduce your cloud bill. Every extra minute of compile time or test time is a direct cost. Luna: Do you think teams will actually change their behavior? Or will this just become another line item that gets absorbed? Lucas: I think it'll be both. Early adopters — the cost-conscious engineering orgs — will start measuring pipeline minutes per developer, per team, per repository. They'll invest in caching, parallelization, and reducing flaky tests that trigger re-runs. Luna: And the others? Lucas: They'll just pay it. $20 a month is noise. But here's the thing: once Azure and Google follow, the pricing will become standard. And then the free tier will shrink over time, just like we saw with data egress, storage class transitions, and everything else. Luna: It's a slow boil. First it's free, then it's a small charge, then the charge goes up. Lucas: Exactly. And the narrative from the providers is always 'we're aligning pricing with usage.' But the net effect is that the total cost of DevOps tooling on the cloud goes up year over year. Luna: What about self-hosted pipeline runners? Could teams avoid the charge by running their own orchestration? Lucas: Technically, yes. If you run your own Jenkins, GitLab CI, or GitHub Actions self-hosted runner, you're not paying the CodePipeline per-minute fee. But then you're paying for the infrastructure to run that orchestrator, plus the maintenance overhead. Luna: And the hyperscalers know that most teams won't do that. The convenience factor is too high. Lucas: Right. The whole value prop of managed CI/CD is that you don't think about the orchestrator. You just write your pipeline YAML and hit commit. So the charge is effectively a convenience tax. Luna: I wonder if this will push more teams toward GitLab or GitHub Actions, which still have generous free tiers for public repositories. Lucas: Interesting thought. GitHub Actions gives free 2,000 minutes per month for public repos. For private repos, it's tiered by plan. GitLab has 400 free compute minutes per month. But those are minutes of compute time, not pipeline orchestration time — the two are different things. Luna: So it's not a direct comparison. But it does create a cost differential between using aws native CodePipeline and a third-party tool running on the same cloud. Lucas: That's the angle that procurement teams should watch. When you're negotiating a cloud contract, the cost of CI/CD tooling is now a line item that can be negotiated. Previously it was bundled into the 'free developer experience'. Now it's explicit. Luna: So what should a team do today? Lucas: First, audit how many pipeline minutes you're using. You can check that in AWS Cost Explorer under CodePipeline. Second, tag each pipeline by team or project so you can see who's consuming the most. Third, look at your average execution time and see if you can reduce it with better caching or parallel stages. Luna: And the fourth thing: start planning for the next contract renewal. If you're up for renewal in six months, you'll want to include pipeline minutes in your committed usage discounts. Lucas: Exactly. AWS already offers reserved capacity for CodePipeline — you can pre-purchase pipeline minutes at a discount. But you have to forecast usage, which is hard if you've never tracked it. Luna: So the initial step is just visibility. Before you can optimize, you need to know what you're spending. Lucas: And that's where most teams will fall behind — not because the charge is large, but because they don't notice it until it's automatically included in the monthly bill. Then they wonder why their cloud spend went up 3%. Luna: A 3% increase across the board is actually huge for a large enterprise. Lucas: It is. And when you stack this on top of the other small charges we've covered — container image pull frequency, secret rotation, internal DNS queries — the cumulative effect is real. It's death by a thousand cuts. Luna: Or, as the cloud providers would say, 'granular pricing alignment.' Lucas: You have to admire the consistency. They never introduce one big price hike. They introduce a dozen tiny ones, each one justified by usage, each one small enough that most customers don't fight it. Luna: And then at the end of the year, your cloud bill is up 15% and you can't point to any single change that caused it. Lucas: That's the strategy. And it works because the alternative — self-hosting — requires engineering time that most companies would rather spend on product features. Luna: So the takeaway for today is: audit your pipeline minutes now, before the free tier changes again. Lucas: And keep an ear out for Azure DevOps and Google Cloud Build announcements. If they follow AWS — and they almost certainly will — the same advice applies across all three major clouds. Luna: Alright, that's a good place to leave it. Next episode, we'll look at the other new charge that dropped on July 1st: AWS Lambda now charges for provisioned concurrency idle time. Lucas: That one's going to hurt teams that use Lambda for API backends with predictable traffic. We'll break it down. Luna: See you then.