Kafka was investable in 2011 when it was a LinkedIn internal project that Jay Kreps and Jun Rao were about to open-source. Spark was investable in 2012 when it was a UC Berkeley research project with a small community on GitHub. Airflow was investable in 2014 when Maxime Beauchemin built the first version at Airbnb. In each case, the window where an investor with genuine technical judgment could establish a meaningful position was short, and the returns for those who did were compounding and large.
The AI infrastructure generation is in that moment now. The companies that will define the next decade of data infrastructure are being founded in 2022 and 2023, and most of them are open-source projects with small communities and no enterprise revenue. Most institutional investors are not writing checks at this stage. Flintrock is.
Why Series A and later is the wrong entry point
There are good reasons why most infrastructure investors wait until Series A or later. The technical risk is real — many promising open-source projects fail to build a community, find a monetization model, or navigate the transition from developer tool to enterprise product. Waiting for traction metrics reduces the failure rate of individual investments.
But at Series A, the price reflects that traction. The project that was raising at a $10M valuation in 2022 is raising at $80M or $150M in 2024 with the same fundamental technology, just more community evidence. If you have the judgment to evaluate the technology in 2022, you've paid a very high price for waiting.
More importantly, at Series A the best founders have options. The investors who were willing to write a $2M check when nobody else would — those investors get the information advantage, the board seat, the relationship that shapes how the company evolves. The investors who show up at Series A after the community has demonstrated the thesis are buying at full price and competing with everyone.
What changes at Seed evaluation
Seed-stage data infrastructure diligence is a different discipline than Series A infrastructure diligence. You are evaluating architecture decisions before they've been validated by production workloads. You are evaluating founding teams before they've navigated a scaling challenge. You are evaluating market thesis before the market has formed an opinion.
The things that matter at Seed are: technical architecture decisions that will either compound or constrain the project as it grows; API design philosophy that will determine whether developers love or merely tolerate the tool; founding team's depth of domain knowledge versus domain knowledge rented from advisors; early community quality — are the first hundred users practitioners running real workloads, or survey respondents?
These are not things you can evaluate from a deck. They require the kind of technical judgment that comes from having built production data systems at scale. Yuki and I spent our careers inside hyperscale infrastructure. We can read the architecture document of an early-stage data infrastructure company and know, within a conversation, whether the founders understand the failure modes they're going to encounter. Most investors who wait for revenue metrics cannot make that evaluation.
The portfolio as evidence
The Flintrock Fund I portfolio reflects this approach. DuckDB, BentoML, and Prefect were all funded in 2022 when they were small open-source projects with active communities but no enterprise revenue model. Qdrant and Chalk were funded in 2023 when they were beginning to establish enterprise traction but had not yet closed meaningful contracts. In each case, the investment thesis was about technical architecture, community trajectory, and ecosystem fit — not about historical revenue.
This is harder than waiting for traction. It requires more judgment per dollar deployed. But it's the approach that produces the best risk-adjusted returns at the infrastructure layer, and it's the approach that lets us serve founders during the period when having a technically credible investor actually changes outcomes.