When you start making the same mistakes over and over. As long as they learn new things with each iteration it doesn't matter if they blow up 5 or 50 ships. The

When you start making the same mistakes over and over. As long as they learn new things with each iteration it doesn't matter if they blow up 5 or 50 ships. The actual material cost of these prototypes is probably in the low 6 figures. The biggest cost is the labor, but by iterating rapidly they are learning new manufacturing techniques, so if they weren't blowing up these prototypes they would be stacking them in a field or tearing them down anyways, since the value is in the process, not the final product, at this point. Also, there are only an estimated 300 employees building Starships (although there are probably more working on engineering), so on-site labor costs are probably about a million a week. Since they can pop out a new iteration every few weeks they cost is probably only about a few million per ship. That's a lot of money in some industries, but it's a rounding error in rocketry. They could build dozens or even hundreds of prototypes for less than the cost of a conventional disposable rocket, like SLS. The idea that the cheapest way to build a rocket is to blow up dozens of prototypes is a bit counterintuitive, but I believe it's the future. Organizations that move more slowly to avoid failures will fall further and further behind, since the tortoise only wins if the hare falls asleep, which SpaceX shows no signs of doing. Civil engineers are able to use mathematical models to develop bridges because we've been building bridges for thousands of years, and they can build in huge safety factors. Fully reusable rockets have never been built before and can't afford the luxury of large safety factors due to the Rocket Equation, so testing to failure is the fastest way to learn, and when almost all of the cost is labor, fastest = cheapest.