Blacksmith: Fixing the CI Bottleneck AI Coding Agents Created
Blacksmith hit a $1B valuation by running CI/CD builds on gaming-grade hardware instead of hyperscaler clouds. Here's how and why.
What does Blacksmith actually do?
Blacksmith builds infrastructure that runs continuous integration (CI) jobs, the automated checks that confirm new code doesn’t break anything, much faster than standard cloud providers. Instead of renting generic compute from AWS, GCP, or Azure, Blacksmith runs CI workloads on gaming-grade hardware tuned for fast single-threaded performance, the same trait that makes a gaming PC good at running Call of Duty. The company says this approach cut wait times for at least one customer from 15-20 minutes to under 15 seconds, and it raised a Series C at a $1 billion valuation led by Sequoia.
TL;DR
- Blacksmith reworks CI infrastructure around the observation that CI workloads behave more like desktop applications or video games than like typical cloud server workloads, favoring fast single-thread execution over parallelism.
- The founders came from outside DevOps, with backgrounds in database engineering and search infrastructure rather than platform engineering, which the team credits with letting them see the problem differently.
- AI coding agents created the bottleneck Blacksmith is built for, since developers and agents now generate code far faster than teams can validate it, shifting the constraint from writing software to testing it.
- The company started on rented gaming servers from a German data center provider known for hosting Minecraft and Call of Duty servers, then grew into direct megawatt-scale power agreements with data centers.
- Blacksmith raised a $66 million Series C at a $1 billion valuation led by Sequoia, following earlier rounds including a Y Combinator SAFE and a Series A from the same investor.
- Customer growth has been steep: Blacksmith says usage grew 20x in the past year, and the team expanded from about five people to roughly 30, including 12 engineers.
- A major reliability incident in Phoenix, where a power surge knocked out cooling systems at a key data center, pushed the company toward wider geographic distribution of its infrastructure.
Why did AI coding tools create a CI bottleneck?
Agentic coding tools let developers produce code at a much higher volume than before. But generating code quickly only matters if a team can confirm it works and ship it. That confirmation step runs through CI: a pipeline of automated tests and checks that validate every change before it merges. When code generation speeds up by an order of magnitude but CI pipelines stay just as slow, the pipeline becomes the limiting factor. Developers end up waiting on test runs and build queues rather than waiting on their own typing speed.
Blacksmith’s pitch is that this shift makes CI speed a direct lever on how fast an engineering org can actually ship, not just a backend chore. The company frames its mission as closing the gap between “speed of thought” and “speed of shipping.”
How does Blacksmith make CI faster?
The core technical insight is about workload shape. Hyperscalers like AWS, GCP, and Azure are optimized for workloads that split well across many parallel threads and many tenants sharing resources efficiently. CI jobs don’t look like that. A single CI job, compiling code, running a test suite, is mostly a long chain of sequential operations where raw single-thread speed matters more than parallel throughput. That’s closer to how a video game runs on a gaming PC than how a typical cloud web service runs.
Blacksmith built its infrastructure around that distinction, running CI jobs on hardware chosen for strong single-threaded performance rather than the dense, parallel-optimized hardware hyperscalers emphasize. The company also focuses heavily on caching: reusing artifacts from previous builds instead of redoing work, and physically locating cached data close to the compute nodes that need it to cut retrieval latency. Internally, Blacksmith describes a lot of custom software for managing fleet growth and observability across the stack, paired with a deliberately narrow product surface area so a small engineering team can support a large and growing customer base.
How did Blacksmith get started?
The company’s three founders were friends from the University of Waterloo who had all worked as engineers at different companies and independently run into the same complaint: CI infrastructure was universally slow and painful to manage. None of them had prior backgrounds in DevOps or platform engineering. Two of the founders had worked on database engineering, and the third had worked on search infrastructure. That outsider position, according to the team, is part of why they approached the problem differently than people steeped in traditional CI tooling.
Without shadow IT.
Blacksmith applied to Y Combinator (on the founders’ second attempt) and got an initial investment via a SAFE. Early infrastructure ran on rented servers from a German data center services company best known for hosting Minecraft and Call of Duty servers, an unconventional choice that one early investor described as surprising precisely because nobody else in CI infrastructure was trying it. As the company scaled, it moved from renting individual machines to negotiating large-scale agreements with data center providers, effectively buying power at the megawatt level rather than buying machines one at a time.
Who is actually using Blacksmith, and why did they switch?
Early customers were small, fast-moving companies similar to Blacksmith itself: teams that valued CI speed highly and had little internal bureaucracy slowing down adoption of a new vendor. Without a dedicated sales team, the company says it landed five- and six-figure contracts largely because switching over was simple, described as a one-click migration of an existing CI setup.
One customer example cited is LangChain, which had been running CI on Google Cloud Platform. On GCP, developers reportedly waited 15 to 20 minutes just to acquire compute for a CI job. After moving to Blacksmith, that wait dropped to roughly 10 to 15 seconds, which the company attributes to exposing what it calls an effectively infinite pool of compute for CI jobs specifically. Engineers, as one team member put it, are a discerning audience that tries lots of new tools and isn’t automatically loyal, so retention depends on consistently outperforming alternatives rather than on contracts or lock-in.
Is Blacksmith worth the attention it’s getting?
The company’s growth numbers, a reported 20x increase in usage over the past year, a jump from roughly five employees to about 30, and a billion-dollar Series C valuation, suggest real market pull rather than hype alone. The underlying thesis also lines up with a broader shift: as AI coding agents push code volume up, any team not also speeding up validation and testing is going to feel that mismatch directly in developer wait times.
That said, Blacksmith operates in a space with real operational risk. The company has already dealt with at least one significant outage, a power surge in Phoenix that took down both primary and backup cooling systems at one of its major data centers, which it says drove home the need for broader geographic distribution of infrastructure. Keeping up with sudden usage spikes, which the team says happen every time a major new LLM ships and developers rush to test it, remains an ongoing operational challenge given the lead time required to procure and install physical machines.
Frequently Asked Questions
What is CI/CD and why does it matter for AI coding?
CI (continuous integration) is the automated process of testing and validating code changes before they merge or ship. As AI coding agents let developers generate code much faster, CI has become the main bottleneck between writing code and actually shipping it, since teams still need to verify nothing is broken.
How is Blacksmith different from AWS or GCP for CI?
Blacksmith runs CI workloads on hardware optimized for fast single-threaded performance, similar to what gaming PCs use, rather than the parallel-optimized infrastructure hyperscalers favor. It also focuses on caching build artifacts close to compute nodes to skip redundant work.
How much funding has Blacksmith raised?
Blacksmith raised a $66 million Series C led by Sequoia at a $1 billion valuation. Sequoia had previously backed the company through an initial Y Combinator SAFE and a Series A round.
Everyone else built a construction worker.
We built the contractor.
One file at a time.
UI, API, database, deploy.
What results have Blacksmith customers reported?
LangChain, an example cited by the company, reportedly saw CI compute wait times drop from 15-20 minutes on GCP to 10-15 seconds after switching to Blacksmith.
What is Blacksmith working on next?
The company mentions building toward autonomous QA, gathering artifacts like screenshots and video snippets from test runs to give developers confidence in code changes without manually reviewing everything, aimed at letting them merge code faster as code volume keeps rising.
