the paper is focusing on experimentation in the context of commercial startups, but arguably more high-risk and potentially high-utility experimentation may happen in other contexts, such as universities, that tend to be largely government funded.

consider silicon valley. government allocates funding for long-term high risk research programs toward various defence objectives. this funds fundamental research in universities.

some researchers then spin off startups to develop products for defence applications -- so the government takes the risk to fund any long-term blue-sky research and then is also willing to as the first customer and provider of contracts for any startups that figure out how to industrialise/commercialise the research for defence applications (electronic warfare...).

see e.g. https://steveblank.com/secret-history/

similarly, consider the development of tcp/ip. pretty handy thing. no startup managed to capture the rights to tcp/ip and change rent to all users of the internet, so arguably from the perspective of commercial startups tcp/ip is a failure.