The Twenty Minute VCSarah Tavel: Will Foundation Models Be Commoditised? | E1149
At a glance
WHAT IT’S REALLY ABOUT
Sarah Tavel on AI, Venture Discipline, and Foundation Model Oligopolies
- Benchmark partner Sarah Tavel discusses how AI is reshaping both the infrastructure and application layers, arguing that most long‑term value will accrue to applications that “sell the work” rather than just improve productivity. She predicts foundation models will likely be controlled by a small oligopoly due to escalating compute and power requirements, making frontier models predominantly closed source, with Meta’s LLaMA as a possible exception. Tavel explains why many first‑wave AI startups are just thin wrappers around models and how newer companies must own more of the workflow and outcome to be defensible against incumbents. Beyond AI, she walks through Benchmark’s partnership model, views on pricing and dilution, board quality, and why strong “why now” timing and founder ambition matter more than ever in hyper‑competitive markets.
IDEAS WORTH REMEMBERING
5 ideasAI enables startups to “sell the work,” not just productivity tools.
Instead of charging per seat for small productivity gains, AI companies can package and deliver entire outputs (e.g., HR ops, translation, sales workflows), effectively selling a 90–95% productivity improvement and pricing against headcount rather than marginal efficiency.
Most first‑wave AI apps are too dependent on foundation models to be defensible.
If 90% of an app’s value comes directly from an OpenAI API, incumbents can quickly replicate it and bundle it into existing products, so newer winners must own more of the workflow, data integration, and end‑to‑end experience to avoid being commoditized wrappers.
Foundation models are likely headed toward an oligopoly, not broad commoditization.
Training each successive frontier model is increasingly compute and power constrained and 10x more expensive, which favors a small number of players able to fund massive infrastructure, keeping frontier models mostly closed source except for special cases like Meta’s LLaMA.
“Why now” is a critical and often underweighted determinant of startup success.
Strong timing acts like a current that can push a company forward despite mistakes, whereas weak or purely conceptual ‘why now’ stories leave founders paddling hard against powerful incumbents or countervailing consumer behaviors (e.g., BeReal vs. TikTok attention gravity).
The best AI application opportunities compound as models improve, rather than get steamrolled.
Tavel aligns with the view that if a 100x better model doesn’t massively enhance your product, the model provider will likely outcompete you; good AI apps should become more capable, less human‑dependent, and higher margin as underlying models advance.
WORDS WORTH SAVING
5 quotesIf you want a model that's on the frontier, you're gonna be closed source.
— Sarah Tavel
AI enables a very different unit of work that you sell, which is doing the work.
— Sarah Tavel
When you have a strong why now, it’s like this strong current that just pushes the company forward.
— Sarah Tavel
If you like everything but the price, you pay the price.
— Sarah Tavel
The model of partnership with big platforms is more about scaling the GP than it is the founder.
— Sarah Tavel
High quality AI-generated summary created from speaker-labeled transcript.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome