Konstantine Buhler, the early-stage AI investor at Sequoia
From predictions to creation: How AI's 3-wave evolution is reshaping technology investment and the future of value creation
I asked Konstantine Buhler, partner at Sequoia Capital, what was the hardest thing as an investor. His answer? Predicting people.
« Investing is a psychological career. You are playing a mind simulation, making predictions about the future and trying to exert influence on a lot of variables. By definition, no one can be perfect at it. »
I had the honor to host Konstantine for an Icons dinner. Here are 8 take-aways from our discussion.
AI evolution in 3 waves: predictive AI (2000-2010) creating trillions in market value through recommendations, descriptive AI (2010-2020) worth hundreds of billions through classification systems, and now generative AI (2020-2030) which may create the most value long-term.
DeepSeek's success follows historical patterns: The efficiency improvements we're seeing with DeepSeek mirror what happened in computer vision, where costs dropped dramatically (from $10 cents per image identification to millions of calls for $1).
Foundation Model economics: There will be diminishing economic returns for the marginal improvements between frontier models and free/open models, challenging the business model of companies solely focused on building better foundation models.
Infrastructure demand remains strong: Despite efficiency gains, GPU consumption for training will continue to grow substantially, supporting the data center investment thesis.
Models are not sticky: Users will readily switch to better AI models, meaning value isn't inherently in the models themselves.
Inference market consolidation: The inference market will likely consolidate to 3-4 major players who can compete on cost with hyperscalers like AWS and Azure.
AI Agents as transformative force: AI agents (systems that make predictions and act on them) represent the next evolution, with examples like Xpow demonstrating superhuman capabilities in cybersecurity testing.
Value creation framework: The most promising investments combine application-layer solutions for real problems with data accumulation advantages, creating compounding benefits over time.
I would like to thank Konstantine for sharing valuable insights on AI evolution and investment frameworks.
Follow me on LinkedIn for more!
