Podcast, Interview
Sequoia Partner, David Cahn on Who Wins in AI, Defence & The New $0–$100M Playbook
- AI compute providers face cyclicality and lower valuation multiples due to commodity pricing, whereas consumers of compute benefit from overproduction-driven cost reductions, improved gross margins, and potential price deflation in the COGS for goods.
- The "year of the data center" narrative is expected to dominate in 2025 via the AI power trade, as power emerges as the primary constraint over capital, though construction delays are anticipated following the summer 2024 "shovel ready" phase.
- Construction capabilities are projected to become a competitive moat with timelines extending to two years, creating winners and losers based on the ability to manage supply chain complexity, particularly when vendors are simultaneously targeted by all players.
- Capital deployment trends favor compute producers due to high capital intensity, even as the narrative shifts toward consumers, while the gap between $150 billion in NVIDIA chip investments and $840 billion in revenue questions remains unresolved by summer 2025.
- Talent acquisition costs have escalated unexpectedly, with recent graduates potentially receiving $50 to $100 million packages and brand names commanding $1 billion, though the actual 1% probability increase in success for such hires may be negligible compared to macro variables.
- Venture capital cannot "make" companies succeed; instead, success relies on product-market fit and founder quality, with overcapitalization leading to rapid hiring and internal distortion fields rather than genuine capability.
- Model providers are expected to pursue vertical integration into chips, power, and infrastructure over the next 12 months to ensure durability, despite initial false predictions regarding Meta's performance in that specific 12-month horizon.
- Monopoly profits are unlikely in the AI era due to massive global competition and the expectation that economic profit above the cost of capital will remain at 1% of global GDP, accruing to working people rather than companies.
- The AI bubble is predicted to unwind via an equity correction rather than a credit crisis, potentially impacting MAG-7 stocks, which mirror the Japan market in the 90s regarding market concentration.
- Demand for AI in sectors like legal is expected to correct downward from 100% adoption rates next year and the year after, as current demand estimates are viewed as overestimated before long-term adoption realizes.
- Gross margins for AI companies are expected to rise over time as compute costs decline, enabling firms to enter the "$100 million revenue" club quickly, particularly those with established product-market fit.
- Founders who have endured significant struggle are expected to outperform those who succeeded quickly, while companies that overfeed engineers with billions are predicted to underestimate productivity.
- Defense represents a "next AI" opportunity with a 50-year catch-up period, expected to consolidate into a few national champions like Anduril and Keleva rather than a broad SaaS-like ecosystem.
- Voice interfaces are considered wildly undervalued and are projected to evolve into relationship-based interactions within 10 years, marking a 50-year generational change for the world.
- AGI timelines are debated, with experts like Sutton and Sutskever projecting 20 to 30 years, while others like Karpathy suggest a decade of agents, and the general view is that timelines are underestimated by youth and overestimated by aggressive voices.
- Investment banking and consulting sectors are expected to lose talent to AI companies as the memetic algorithm guiding career choices breaks down due to the AI cataclysm.
- The mainstream media is predicted to fully adopt the narrative linking AI's physicality to GDP growth within one year, with AI becoming one of the biggest contributors to US GDP growth driven by infrastructure construction.
- Economic profit expectations suggest AI will affect 5% of GDP in the fullness of time, but not generate $4 trillion in economic profit as monopolistic margins are overestimated.
- The market is currently pricing AI implementation on a short-term timeline that contradicts the long-term reality, with the bubble expected to persist until incentives change in an uncoordinated fashion.