One day in 2002, rocket engineers Tom Mueller and John Garvey were working on an advanced engine at the Reaction Research Society’s test site in the Mojave Desert. As was his habit, Elon Musk came out to kibitz about rockets and to observe the action; at that point, the three men were developing the concepts for what would be SpaceX’s first launch vehicle. Garvey and Mueller mounted their design on the test stand—a secure steel framework bolted into a concrete foundation that would prevent the engine from taking off willy-nilly into the desert scrub.

After taking up position a safe distance away, the observers ignited the engine. It took just a hundred milliseconds for the prototype to explode, taking the test stand with it in a burst of flame as they looked on.

“Elon, you better get used to this,” Garvey warned the prospective space entrepreneur. Rocketry is hard, he told Musk, and it comes with many expensive setbacks. They’d need to delay further testing for weeks, until they rebuilt the stand. Musk simply turned to Mueller and said, “Tom, make sure we build two stands.”

Garvey saw Musk’s response as a good approach to the risks that come with building rockets. At a big contractor or at NASA, the response might have been: We can’t have a failure like this, so let’s spend time and money ensuring that the prototype doesn’t explode during testing. Traditionally, these kinds of precautions take time and can defeat the purpose of testing itself.

When he was still at McDonnell Douglas, Garvey had worked on a project to develop lightweight, high-pressure vessels made of carbon fiber that could hold ultra-chilled rocket propellants. He pitched the company on building a dozen small prototypes to test on cheap rockets for initial evaluations of different designs. Instead, the project managers chose to make one enormous tank the way NASA wanted it made. This technology deviled the space industry for years as engineers sought to safely cut weight from the vital plumbing of the rocket; tank failures led to the cancellation of entire rocket programs. Garvey, working with a company called Microcosm, flew the first high-pressure composite tank containing chilled liquid oxygen in one of his rockets, out in the desert.

Aversion to failure desperately reduced innovation in the aerospace industry. But a different approach was common in Silicon Valley. Software engineers rejected the so-called waterfall style of project management common in more industrial settings, where product requirements are outlined, developed, tested, and implemented in rigid succession. Instead, under rubrics like “agile” engineering, developers would gradually build out the software, testing it as they went and updating requirements in the face of challenges. This is the origin of the “fail fast” ethos associated with risk-taking digital entrepreneurs: once you figure out what doesn’t work, it’s easier to figure out what does.

These approaches, however, don’t always lend themselves easily to physical manufacturing, where the cost of materials and machining adds up faster than the hours of software engineers making virtual products. But many in the aerospace industry saw endless requirements and a lack of experimentation as problems that needed fixing. That was especially true as the amount of software inside the rocket, and used to build it, increased. For Garvey, Musk’s ability to account for failure — indeed, to expect it — showed a healthier attitude toward building rockets than that of his former managers at McDonnell Douglas.

NASA and the prime contractors had come by their risk aversion honestly: it had everything to do with flying humans in space. NASA gift shops sell T-shirts emblazoned with the phrase “Failure is not an option.”

The agency’s critics argue that this attitude has permeated well beyond a refusal to abandon their comrades in their hour of need. The expectation of perfection reflects an institution that sets the highest standards, which the US space agency sets out to be. But, drifting down to technical departments that depend on delivering success to win funding from Congress, an aversion to mistakes becomes an obstacle to taking the risks that bring real innovation. “The most expensive way to run a program is doing so in a fashion that ensures you’ll never break a piece of hardware,” Mike Griffin, a veteran space engineer who consulted with Musk, told me.

And, as any investor will tell you, without risk there is little reward.

“I thought maybe we had a 10 percent chance of doing anything — of even getting a rocket to orbit,” Musk would say of his company when it opened its doors in April 2002. Luke Nosek, who helped build PayPal with Musk and later joined the rocket company’s board of directors, underscored the doubts of the time: “So many of his friends advised him not to do SpaceX.” Yet Musk’s pitch to become a rocket maker was similar to those for his other companies.

“Just as DARPA served as the initial impetus for the internet and underwrote a lot of the costs of developing the internet in the beginning, it may be the case that NASA has essentially done the same thing by spending the money to build sort of fundamental technologies,” Musk told a classroom full of budding entrepreneurs at Stanford a year after SpaceX’s inception. “Once we can bring the sort of commercial, free enterprise sector into it, then we can see the dramatic acceleration that we saw in the internet.”

Rocket Billionaires: Elon Musk, Jeff Bezos and the New Space Race is available March 20.

Excerpted from Rocket Billionaires by Tim Fernholz. Copyright © 2018 by Tim Fernholz. Reprinted by permission of Houghton Mifflin Harcourt Publishing Company. All rights reserved.