Microformats, semantic HTML tags, and non-rendered meta-data can be found all over the place. Especially in eCommerce.
What hasn't caught on is the formal language of the semantic web. And for that my hypothesis is that it is so formal and so strict that it produces a barrier of entry that is too high for non-academics. RDF has a bit of a leaning curve compared to competing technologies. Furthermore, FOAF (Friend of a Friend) last time I looked at it had a lot of outdated domain specific tags. For example, there is a tag for ICQ (an instant messenger practically no one uses anymore but was very popular in the 90s) but no tags for new technology rather than, say, having generic tags that can be used for any messenger system and using attributes.
Summary:
- High barrier of entry
- Easier to use alternatives
- It hasn't really failed, it's just taken a form that is more friendly for day-to-day
Edit: Also, if you go the Semantic Web Meetup in Cambridge there are a lot of talks of the medical industry, database industry, and things like that which show it is alive https://www.meetup.com/The-Cambridge-Semantic-Web-Meetup-Gro...
Many companies are using RDF-based technology and are not talkative about it because they see it as a secret weapon. I find out about these by going to conferences, cold calling, etc.
People are slowly starting to understand about JSON-LD, maybe around 2025 people are going to realize that it's a problem that vanilla JSON lacks decimal math. (Even COBOL programmers know if you do money math with floats you will sooner or later cut somebody a check for the wrong amount) JSON-LD unobtrusively adds the "good parts" of XML (decimal math, date and time types,...) to JSON.
As for the model of "Linked Data" the deep problem is that the model of "I publish data, throw it over the wall, and my responsibility is done" does not work. As much as people like to deny it, there is a science of quality, and it involves forming feedback loops that just don't exist in the likes of DBpedia, Wikidata, etc. See the last part of this essay
http://ontology2.com/the-book/data-lakes-ponds-and-droplets....
Because most of the time, you already have your data in a structured format of some kind. So if you need to share data between your own systems, you use the underlying data, not the presentation layer. And your end users see the UI created by your markup, so they don't really care either. Semantic markup really helps other people ingest your data... so unless you intend for your data to be shared, it infrequently makes a ton of business sense to invest effort into making things semantic.
That being said, if you DO intend to share your data, and want it exposed and usable, then it does make sense to work at it. But in the grand scheme of things... where does most web content fall?
also, you need to decide on a format, and there are all sorts of micro formats to choose from. You would have to get everyone on board.
Writing HTML from the old days was pretty easy, and most browsers could handle bad HTML. I am not sure if the same could be said about bad micro formats.