Blacksmith Hits $1B Valuation as AI Coding Agents Break Software Testing
AI writes code so fast it broke testing pipelines, turning a continuous integration startup into a unicorn in under two years.
Funding · Source: TechCrunch
What happened
Blacksmith just raised a $66 million Series C led by Sequoia. The deal values the continuous integration startup at $1 billion. This comes just two months after a $45 million Series B valued them at $550 million. They doubled their valuation in eight weeks without changing the core product.
The massive jump comes from sheer volume. AI coding agents like Claude Code and Cursor are flooding developer pipelines. A human pushes three or four pull requests a day. An AI agent runs iterative debugging loops that trigger up to 20 automated tests per task. This caused Blacksmith's weekly continuous integration jobs to explode from 5 million in January to over 90 million by October.
This surge broke standard testing infrastructure. Blacksmith handles the load by ditching standard cloud servers. They run jobs on gaming-grade desktop processors like the AMD Ryzen 7950X. They also use local NVMe drives and Firecracker microVMs to isolate jobs in milliseconds. The result is a system that runs twice as fast at up to 60 percent lower cost than GitHub-hosted runners. Annual recurring revenue is up 17 times this year.
Key facts
- $1 billion — Blacksmith's new valuation after the Sequoia-led Series C
- 90 million — Weekly continuous integration jobs run by Blacksmith as of October 2026
- 17x — Growth in Blacksmith's annual recurring revenue this year
- 5.7 GHz — Boost clock speed of the AMD Ryzen desktop chips Blacksmith uses
- 25-fold — Growth in Anthropic's internal CI infrastructure six months after launching Claude Code
Why it matters
The bottleneck in software development just moved. It is no longer about how fast you can write code. It is about how fast you can verify it. AI agents never sleep and never stop submitting pull requests. If your testing pipeline takes 45 minutes to validate a change, your AI agents sit idle. Speeding up continuous integration is now a mandatory requirement to get any real return on investment from AI coding tools. GitHub saw AI agent pull requests jump from 4 million to over 17 million in just six months, causing public platform incidents as infrastructure strained.
Infrastructure spending is shifting permanently. Everyone focuses on the cost of large language model APIs. But continuous integration compute might emerge as the largest line item in your engineering budget. As agents become more autonomous, they run longer planning loops. They generate exponentially more testing work before a human ever sees the code. Anthropic saw its own internal continuous integration infrastructure grow 25-fold in the six months after launching Claude Code.
For builders
Optimize testing for single-thread speed
Most code compilation and testing cannot be parallelized well. Blacksmith proves that high clock speed desktop chips beat expensive server silicon for these tasks. Builders paying for standard cloud runners are losing money on the wrong hardware.
Move caches to local storage
Network-attached storage throttles testing speeds. Pulling dependency trees over a network takes up to half of your total run time. Switching to local NVMe drives cuts this bottleneck and saves compute costs for engineering teams.
Prepare for exponential CI costs
AI agents will inflate your cloud bill through sheer testing volume. Founders must budget for continuous integration scaling alongside their LLM API costs. If you pay per minute on standard platforms, your expenses will multiply rapidly.
My take
AI promised to make software cheap, but it just moved the toll booth. I find it hilarious that we built the most advanced reasoning engines in history, only to get bottlenecked by waiting for code to compile. If you want to build a billion-dollar AI startup today, stop making wrappers and start fixing the boring infrastructure that AI is actively breaking.
Original reporting: TechCrunch. This is my rewrite and opinion.