Key Takeaways

  • Felicis generated its largest historic returns by backing infrastructure layers, including Shopify for digital storefronts, Adyen for global payments, and Credit Karma for consumer credit data.
  • Application layer companies like Lovable and Replit race for citizen developers, but both rely on Supabase as their database, concentrating revenue into the shared foundation.
  • Senkut targets categories with proven terminal value, pointing to historical database giants like Oracle, Snowflake, and ClickHouse to justify underwriting lower layer risk.
  • Autonomous agents expand data consumption rather than eliminating it, ensuring database platforms capture downstream activity regardless of whether human users churn.

The Winner Takes the Rake

Venture investors spend billions guessing which front-end application wins the next consumer or enterprise cycle. Aydin Senkut built Felicis across two decades by doing the opposite: buying the rails beneath the fight. “Virtually all of our major fund maker companies have been foundational layer infra companies,” (play from 18:19) Senkut said. “Shopify was commerce infrastructure, Adyen was payment infrastructure, Credit Karma was credit score infrastructure.” (play from 18:25)

Front-end software cycles move fast and burn cash. Consumer preferences turn on a dime, switching costs for citizen developer platforms stay low, and churn remains a constant drag on enterprise value. When capital floods into consumer-facing tools, backing the shared foundation lets an allocator capture the upside of the wave without picking the individual survivor.

Consider the recent rush into citizen development. Platforms like Lovable and Replit saw fast revenue spikes as non-engineers started generating software. Backing either one requires taking direct product and retention risk. Felicis bypassed that choice by funding their data layer instead.

“There are some successful companies that have grown revenues, Lovable and Replit, chasing the citizen developers,” (play from 17:39) Senkut noted. “Even though we have not backed them, virtually all of them use Supabase as the underlying database. We didn't have to make one of those bets. We still won because the company we backed that is the underlying infrastructure layer is benefiting from the sprints of these companies and the growth spurts.” (play from 17:51)

The Category Zip Code

Underwriting infrastructure requires patience, but it removes product fad risk. Senkut argues that the market category itself dictates the ceiling of an outcome. Some software sectors top out at single-digit billions, while base infrastructure compounds for decades.

“One of the reasons why we had confidence is if you look at the history of tech, some of the most valuable companies are database companies, whether it's Oracle or Snowflake, ClickHouse,” (play from 43:33) Senkut said. “The zip code and category matters as much.” (play from 43:44) In generative AI, Felicis applied the same logic to foundational models and media generation pipelines like Runway, treating them as base platforms that feed downstream applications.

The shift toward autonomous AI agents reinforces this dynamic rather than breaking it. While agentic workflows threaten to automate seat-based software seats and disrupt traditional SaaS pricing, they cannot bypass storage, state, and query layers. “The agents, even if they're not humans, they're still going to need an underlying database,” (play from 44:24) Senkut observed. As software creation costs drop toward zero and autonomous bots spin up millions of ephemeral applications, compute and database query volumes compound, funneling enterprise spend straight back to the foundational layer.

Why It Matters

Private equity and growth allocators face severe margin compression in seat-based SaaS as generative tools lower the technical barrier to entry for competing applications. Senkut's portfolio track record signals that enterprise defensibility has migrated down the stack into data persistence and operational plumbing. Capital allocators evaluating software assets must underwrite database and rails infrastructure, where pricing power survives application churn and autonomous agent proliferation.