Y Combinator · Built to a Billion
The $1B Company Fixing AI’s Software Bottleneck
Blacksmith built a specialised CI cloud on gaming CPUs and cut developer wait times from 20 minutes to 15 seconds—now valued at $1B after 20× growth in a year.
Tap a timestamp pill below to jump the video to that moment.
The problem
AI writes the code. CI can’t keep up.
Every software team runs a continuous integration pipeline—a suite of tests and checks that must pass before any new code lands in production. For most of computing history, CI was a minor inconvenience: developers wrote code slowly, so a ten-minute wait was tolerable. That bargain has quietly collapsed.
Agentic coding tools—AI assistants that generate, modify, and iterate on code autonomously—have multiplied the volume of commits hitting CI queues. Developers who once pushed a handful of changes a day are now pushing dozens. But CI infrastructure was sized for the old world. The result is that the bottleneck has migrated: the problem is no longer writing software, it’s validating it.
“the bottleneck has become not writing the software but validating it”
“With agentic coding tools, developers are writing more and more code than they ever were. But it’s not very useful to write 100 times the code if you can’t ship it at that speed. So now the bottleneck has become not writing the software but validating it. And that usually means waiting on your CI. So that means making sure that the tests are passing and that you have confidence that what you’re shipping actually works.”
The mismatch runs deeper than just queue length. Standard cloud compute from AWS, GCP, or Azure is optimised for throughput workloads: many threads running in parallel, waiting on network or disk. CI is different. A compiler—the core of most CI jobs—hammers a single CPU core as hard as it can. Hyperscaler VMs, tuned for server workloads, are mediocre at exactly this task. For years, teams accepted that as a fact of life.
How they solve it
Gaming CPUs, gamma-ray caching, and a fleet built from Minecraft servers
Blacksmith’s founding team—three friends from the University of Waterloo, two of them database engineers and one who worked on search infrastructure—came to CI with no DevOps baggage. That naivety turned out to be an asset. Where seasoned platform engineers saw a scheduling problem, the Blacksmith founders saw a hardware mismatch.
The key insight was that CI workloads behave like video games: they want one very fast CPU core, not sixteen mediocre ones. Gaming CPUs—consumer chips sold for high-frame-rate play on a desktop PC—are engineered precisely for high single-thread clock speeds. So Blacksmith’s founders did something that made their early investor do a double-take: they rented racks of gaming servers from a German data centre that had previously hosted Minecraft and Call of Duty game servers, then pointed CI jobs at them.
“CI workloads are a lot similar to…video games…they care about that one single thread”
“What we started to realize was the characteristics of CI workloads are a lot similar to—you know—desktop applications, in particular video games that you would run on your typical gaming PC, in that they don’t care as much about multiple kinds of threads of execution but they care about that one single thread of execution being really, really fast. When we started off, we would rent these servers from this German data center services company that was known for hosting Minecraft servers and Call of Duty game servers.”
Speed alone is not enough when a CI job repeats work already done in a previous run. Blacksmith’s second major lever is caching: keeping build artefacts—compiled libraries, test fixtures, dependency trees—physically close to the compute nodes that need them. Moving that cache layer from a central object store to local NVMe attached to the runner eliminates the network hop that dominates cache-hit latency on commodity clouds.
The business model is deliberately simple. Teams swap their existing GitHub Actions runner config to point at Blacksmith in a single click. There is no complex migration, no agent to install. That low-friction entry point let the founders win their first 200 customers through personal outbound—cold-calling CTOs and VPs of Engineering—without a sales team. LangChain is a recent example: before switching, its developers waited 15 to 20 minutes just to acquire compute before a CI job even started. On Blacksmith, that queue time dropped to 10–15 seconds.
“wait time went down from 15 to 20 minutes to anywhere from 10 to 15 seconds”
“One of the customers that onboarded onto Blacksmith recently was Langchain. Langchain was previously running their CI on GCP. On GCP, their developers would routinely wait 15 to 20 minutes to even acquire compute to run the CI job. On Blacksmith, because we effectively expose an infinite pool of compute, this wait time went down from 15 to 20 minutes to anywhere from 10 to 15 seconds, significantly improving the velocity at which their developers could push code.”
Scaling to the company’s current size—30 people, thousands of servers, megawatts of contracted data-centre power—required the team to stay ruthlessly focused on surface area. Rather than accommodating every bespoke customer request, they kept the product small. That discipline let 12 engineers operate infrastructure that serves hundreds of enterprise engineering teams. The tradeoff surfaces when demand spikes: a power surge in Phoenix knocked out both primary and backup cooling at a major data centre this past summer, underscoring how geographically concentrated the fleet still is. Distributed deployment across regions is now a priority.
Takeaway
The quick version
- AI coding shifts the bottleneck. More code generated means more CI runs—and slower CI is now the limiter on shipping speed, not developer typing speed.
- Matching hardware to workload matters. Hyperscaler VMs optimise for multi-threaded server workloads; CI wants single-thread clock speed. Gaming CPUs are a cheap, high-leverage fit.
- Outsider perspective helped. The founders had never written a CI pipeline, so they questioned assumptions that DevOps insiders had long since stopped examining.
- Next frontier: autonomous QA. Blacksmith is building tooling to gather screenshots and video artefacts that give developers AI-generated confidence in a change—so code can be merged without a human checking every output.
“Part of the answer is just being naive enough in the right way to not kill a good idea before it meets the market.”— Blacksmith co-founder, Y Combinator
If agentic coding tools keep compounding, the gap between code-generation speed and validation speed will only widen—making fast, purpose-built CI infrastructure a bigger structural advantage with every model release.