StoryTechnical

Fixing AI Coding's Bottleneck

Y Combinator

Blacksmith is a CI/CD infrastructure company that accelerated software validation by using gaming CPUs and custom cloud infrastructure optimized for CI workloads. Founded by engineers from Waterloo who identified that CI was universally slow across their previous employers, the company raised $66M at a $1B valuation and now serves customers like Langchain, reducing CI wait times from 15-20 minutes to 10-15 seconds.

Summary

Blacksmith addresses a critical bottleneck in modern software development: as AI coding tools enable developers to write code faster than ever, the ability to validate and ship that code has become the limiting factor. The company was founded by three friends from University of Waterloo who worked as infrastructure engineers at various companies and consistently observed that CI management was painful and slow across the board.

The founding insight came from recognizing that CI workload characteristics differ fundamentally from the workloads that hyperscalers like AWS, GCP, and Azure optimize for. CI jobs prioritize single-thread execution speed rather than parallel throughput, similar to desktop gaming applications. This led the founders to an unconventional solution: using gaming CPUs (initially renting servers from a German data center company known for hosting Minecraft and Call of Duty servers) rather than traditional server infrastructure. Over time, Blacksmith scaled from renting individual machines to negotiating megawatt-level power agreements directly with data center providers.

The company's initial go-to-market strategy was grassroots—asking engineer friends at other companies for logos to build credibility, then cold-calling CTOs and VPs of Engineering. Early customers were small, fast-moving companies with minimal organizational friction. Blacksmith's product design prioritizes ease of use (one-click migration) and performance, enabling five-to-six-figure contracts even without a dedicated sales team. A notable customer win was Langchain, which reduced CI job acquisition wait times from 15-20 minutes on GCP to 10-15 seconds on Blacksmith by leveraging its effectively infinite compute pool.

As of the transcript, Blacksmith operates a team of 30 people (12 engineers) managing many orders of magnitude more infrastructure than the initial small cluster. The company maintains focus through ruthless product discipline, keeping surface area small and only scaling features with team size. However, the company faces scaling challenges: each new LLM release drives massive spikes in CI usage as developers rush to test and ship, requiring creative technical solutions around job placement algorithms, packing efficiency, and network topology. Geographic distribution became critical after a power surge in Phoenix knocked out primary and backup cooling systems, affecting customer job startup times.

Looking forward, Blacksmith is developing autonomous QA capabilities to gather confidence artifacts (screenshots, video snippets) that help developers trust their code changes. The broader vision is to remove validation as a bottleneck and allow developers to ship software at the speed of thought.

Key Insights

  • The bottleneck in modern software development has shifted from code writing to code validation, as agentic coding tools enable developers to generate code 100x faster but cannot ship at that speed without fast CI systems
  • Blacksmith identified that CI workloads have fundamentally different characteristics from hyperscaler workloads—they prioritize single-thread execution speed like gaming applications rather than parallel throughput, requiring different hardware and infrastructure design
  • By moving cache artifacts close to compute nodes, Blacksmith significantly reduces redundant work and enables major performance gains—LangChain achieved a reduction from 15-20 minute CI wait times to 10-15 seconds
  • Blacksmith maintains scaling efficiency by being ruthless about keeping product surface area small and manageable, scaling features only with team size rather than accommodating bespoke customer requirements, enabling a team of 12 engineers to manage many orders of magnitude more infrastructure
  • Each new LLM release creates massive spikes in CI usage as developers rush to test and ship code, forcing Blacksmith to innovate continuously with better job placement algorithms, packing efficiency, and network topologies to keep up with demand

Topics

CI/CD Infrastructure OptimizationGaming CPU Architecture for CI WorkloadsDeveloper Velocity and Code ValidationScaling Infrastructure and Geographic DistributionAutonomous QA and Developer ConfidenceGo-to-Market Strategy for Infrastructure ProductsAgentic AI and Code Generation Bottlenecks

Transcript

[0:00] Every time a developer writes code, they have to run a bunch of checks to make sure that nothing is broken. That process is called CI and Blacksmith writes infrastructure to run those really, really fast. We've grown 20x in in the last year. We recently raised our series C. This was done at a billion dollar valuation led by Seoia and we raised about 66 million. We were outsiders. We were not platform engineers. We didn't work in DevOps. In fact, none of us had written a CI pipeline before. [0:30] >> When I think about why we were one of the only ones to to approach the problem this way and to think of it this way,…

Full transcript available for MurmurCast members

Sign Up to Access

More from Y Combinator

Get AI summaries like this delivered to your inbox daily

Get AI summaries delivered to your inbox

MurmurCast summarizes your YouTube channels, podcasts, and newsletters into one daily email digest.