There isn't much real-world RDF data available on the Internet right now. This page lists some SPARQL endpoints and large sources of data. Many other smaller sources are listed at rdfdata.org.

"Live" SPARQL Sources

These sources make their data available over SPARQL, possibly as well as a complete download.

• U.S. Census: 1 billion triples of U.S. census data (disclaimer: I'm the author of that data set)
• GovTrack: Around 10 million triples containing census data for U.S. locations (including lat/long), brief biographical data for all members of Congress, and mainly data for federal legislation and voting records going back five years. (disclaimer: I'm the maintainer of that)
• DBLP Bibliography Database: 15 million triples on computer science bibliographic data.
• DBPedia: Describing 1.6M Wikipedia articles
• my.opera.com: Almost 2.7 million triples of my.opera.com data.
• BBC Backstage from HP Labs: detailed information on the BBC schedules (including digital tv and radio) for the next week, updated every morning.

RDF/XML and N3 Sources

These sources provide raw downloads of large sets of RDF/XML or N3 or otherwise expose a large amount of RDF data.

• DAML: Geographic locations, WordNet dictionary, airports, companies, military information, CIA World Fact Book, etc.
• Class and property definitions for WordNet ontology in OWL format (714k triples)
• OpenCyc (250k triples)
• Uniprot RDF: Protein sequence and annotation data. An enormous amount of data.
• French geography: Départements, Cantons, Communes (590k triples)
• RDF Book Mashup: Wraps several book-related APIs