Why Do Startups Always Buy the Smartest AI Model?
Martin Casado says two competing startups never pick a dumber model to get ahead. In a competitive equilibrium you always pay 10 times more for the smartest model.
He leaves open whether tasks will saturate so a smarter model is no longer needed, and he thinks that is very hard while companies are actually competing. Some tasks may saturate anyway. The real test is whether software ships faster, gets better, and wins in the marketplace, and that pressure pushes teams to the more intelligent model. Progress can still slow when companies are not spending more tokens, so they either increase prices or go broad. In his experience pricing holds, and the competitive equilibrium still means everyone wants the smartest model. Use is spreading past code into general knowledge work.
Martin Casado on Frontier Model Economics, AI's Slowdown, and the Cursor Deal ·