The first data warehouse I worked on was built completely on data marts. Data marts are an anti-pattern, ignoring data normalization principles, creating a single table with repeating data in multiple rows, or having repeating columns. These tables are not normalized to the basics of even first normal form.
The de-normalization is not randomly achieved. Often, query results for producing reports are used as definitions for data marts. The data marts will have many indexes to optimize filtering and ordering.
From a data warehousing perspective, multiple data marts may be joined, producing similar behaviors to cubes or other data mining structures. They are not as flexible as data cubes that can be generated from a data mining tool. They are better suited for a portal, or reporting system, with fixed presentation.
Data Marts are populated with any number of ETL processes. You can populate them by inserting the results of queries against a normalized database. They can be imported from external sources as well.
If you are exploring data warehousing for the first time, Data Marts are an easy way to get your feet wet. You can quickly provide content for your users, allowing you to grow your skills organically, while producing useful results.
Cheers,
Ben