Fireside Chat, Interview
Sarah Tavel: Will Foundation Models Be Commoditised? | E1149
- A global oligopoly of frontier AI models is expected to emerge as successive training steps incur exponentially higher costs driven by compute, power, and capital constraints, creating a "winner-take-most" scenario where only large-scale investors can sustain progress.
- The application layer is predicted to capture the majority of value by shifting from selling "software" to selling "work," a shift that expands the total addressable market by 10 to 50 times against headcount costs and offers 95% productivity gains rather than marginal 10% improvements.
- Founders are identified as needing a compelling "why now" narrative driven by technology catalysts or current trends to navigate a competitive landscape where 90% of value for first-wave startups currently resides in the underlying model provider, leaving "wrapper" companies vulnerable to incumbents bundling features.
- While the price of AI services for end customers is expected to decline due to market competition, the cost to train foundation models will continue to rise significantly, likely keeping the absolute frontier models closed-source as open-source business models cannot currently sustain the massive capital requirements.
- Startup success and venture investment strategies are projected to favor companies with unique product dynamics, deep founder motivation, and the ability to escape competition, whereas those relying on platform dependencies or failing to avoid pro-rata dilution traps face high failure risks.
- Specific investment philosophies emphasize partnering with founders who demonstrate vulnerability, urgency, and a desire for partnership rather than platform support, with benchmarking partners avoiding delegation of core work and resisting price-based deal structures to maintain conviction.
- Macro-political factors, including a predicted inevitability of Donald Trump's reelection and a perceived intensification of anti-Semitism, are noted as external crosswinds that will influence the operating environment for businesses and investors.
- Deep specialization in single modalities, such as Eleven Labs in audio, may allow specific AI startups to achieve product parity against generalist models by focusing on workflow integration and customization that broader models cannot easily replicate.
- Incumbents like Microsoft, Adobe, and Notion are expected to leverage existing distribution and workflows to sustain their positions, while startups must avoid being "unbundled" by competitors or failing to raise the tremendous capital required to compete in high-growth AI segments.
- The market is characterized by a dichotomy where "sustaining" technologies benefit incumbents, while disruptive opportunities exist only for those who fundamentally change mental models to sell outcomes and automate specific work products like HR ops, recruiting, and sales.