In the case of SingularityNET, AI agents are accessible and composable. They can interact with other agents, or act in unison to form larger, more complex agents. The market incentivizes development and maturity of both the system and the individual agents. This is a killer feature if you ask me. There will be a great number of already-useful agents available upon launch of SingularityNET, to kick-start the service's ability to provide tangible value.
However, you're right that most of these projects fail to deliver any value. Many projects are riding the hype-train, and many more are outright scams. However, the example I named is none of those.
Edit: I should add that a big problem of centralized markets is that nobody wants to put all of their eggs in one basket. Take Second Life for example: In SL, your digital avatar and assets are siloed into that world. You cannot transfer them to the next great virtual world. This is conceptually similar to the standardization debates. Anyone can make their own protocol, but if there are profit motives behind one, the industry will be reluctant to adopt. Standardization takes a lot of time, trust, and debate. And for good reason!
With decentralized marketplaces, standards are not quite as important. An implementation can be as fluid as an app, and the cost of replacing one interface/implementation with another is much lower than the cost of standardization. The internal workings of the marketplace itself can be altered with a democratic vote.
This is all theoretical, we have yet to see these ideas produce real returns. But the killer feature is that people have more incentives to invest their resources toward adopting the platform, and that itself is a tangible value, as long as the adoption of the platform itself creates value in other ways.