Madrona Ventures raises $770M to bet on AI application companies, not foundation models

Summary

Madrona Ventures, the 30-year-old Seattle venture firm and early backer of Amazon and Snowflake, has raised $770 million across two new funds focused on AI application companies rather than frontier models. The firm closed $495 million for early-stage investments across roughly 30 companies and $275 million for later-stage Series B and C rounds, targeting about a dozen companies in that bracket.

The capital haul exceeds Madrona's previous round of $690 million but trails recent mega-funds from Thrive Capital, General Catalyst, and Andreessen Horowitz. The firm's strategy reflects a deliberate narrowing: it is avoiding the expensive, winner-take-most race to build foundation models and instead backing startups that differentiate by customizing large models for specific customer needs and specialized data.

Runway AI, a video generation and editing company that Madrona invested in during 2022, exemplifies the thesis. Runway does not compete on building frontier models; instead it layers AI capabilities on top of existing infrastructure to create differentiated tools—comparable to a next-generation Adobe Premiere rather than a standalone model play. The company's website directs users to web-based video editing features, not APIs, and early tools like its rotoscoping capability, which uses AI to isolate video layers without green screens or manual drawing, show how the strategy focuses on solving specific user problems.

Madrona is also screening for founder discipline. The firm passes on researchers primarily interested in retaining board control for entire founding teams over building sustainable businesses, or those focused more on model capability than product-market fit. This disciplined capital allocation reflects how mid-sized venture firms are carving out positions as AI-focused funding intensifies and more researchers launch startups.

Emerging patterns in consumer AI apps will likely shape where capital lands. Products that become more personalized and tailored over time—storing user preferences and learning patterns that competitors lack access to—may create durable switching costs and walled gardens. The interaction model itself is shifting: successful apps may stop requiring users to prompt them and instead proactively ask questions based on context, inverting the engagement dynamic from passive tool to active assistant. Consumer AI apps chasing obvious ideas like AI girlfriends or dating bots may struggle; breakout products will likely embed AI invisibly, making the technology invisible until critical mass is reached.