Perhaps my toughest and most surprising lesson on chip startups is you need to believe the new approach can be 50–100x better than incumbents today on some important metric (performance, cost, efficiency, whatever). That’s a ridiculously high bar. The brutal math: 4–5 years to reach production scale. Incumbents improve ~2x/year, so they’ll be 16–32x better by then. And the startup still needs another ~3x to overcome ecosystem, switching costs, skepticism and inevitable lack of optimizations. 10x better than today sounds extraordinary… and yet isn’t nearly enough. Hardware is hard.