TechCrunch Startup News Flapping Airplanes and the promise of research-driven AI; plus, Upwind raises $250M at $1.5B valuation to continue building ‘runtime’ cloud security
Jan 29, 2026
A deep dive into a new AI lab aiming to train large models with far less data. Discussion of research-first versus compute-first approaches and why long-term bets matter. The rise of a runtime cloud security company to a $1.5B valuation and how it detects and remediates in real time. Tales of early doubt, slow sales, and a $250M funding milestone.
AI Snips
Chapters
Transcript
Episode notes
Research-First Over Compute-First
- Flapping Airplanes pursues a research-first path that favors long-term breakthroughs over raw scale.
- Sequoia's David Kahn frames this as expanding the search space rather than doubling down on compute-only bets.
Temporal Spread Of Bets Matters
- A compute-first approach prioritizes cluster scale and short-term wins while a research-first approach spreads bets over five to ten years.
- Research-first accepts many long-shot projects to collectively increase the chance of major breakthroughs.
Founders' Exit Shaped Security Thesis
- Upwind's founders previously built Spot.io and sold it to NetApp for about $450 million in 2020.
- That operational experience shaped their belief they understood cloud environments better than many security incumbents.