This is a hasty phylogenetic tree I drew up showing the major groups of I-M26 (which has been known as I2a1 for the last 3 years or so). This is the group well-known for occurring in around 40% of men in Sardinia. But in the last three years we have discovered several new SNPs that split M26 into more specific groups which occur in different regions, all in westernmost Europe.
The Sardinian M26 seems to be all L160+ which makes an out of Sardinia origin unlikely. Spain and France contain most of the major groups of M26 and I think our M26 ancestors started expanding from somewhere like southern France, up the Atlantic coast to northern France, Britain and Ireland, and probably up the Rhine to Western Germany and Switzerland. A very few M26 made it to Denmark, Norway and Sweden. The single M26 I know of from Iceland has a 37 marker STR haplotype similar to some Irish M26+ L160- people. There are also a very few M26+ people from Jewish families from Poland, Belarus and Odessa, I believe most of these probably go back to Sephardic (Spanish/Portuguese Jewish) roots, and in fact some of the families have a tradition that their paternal ancestors were Sephardic.
Some specifics about the tree: the green triangles show the periods when population have been expanding, the L160+ triangle is the biggest because this group is most common. The dates when the expansions started and when the different groups split is very loosely based on Ken Nordtvedt's work but I didn't attempt to show his calculations very accurately. Not shown is M161 which was discovered before the year 2000 but has never been found in testing at Family Tree DNA or as far as I know at 23andMe. Also not shown is the Z106 SNP, this is a subset of L160+ which probably occurs in about 1/3 of Spanish L160+ and in a much smaller percentage of English and other northern L160+. The L277 SNP is placed in quotes because FTDNA has not been able to sequence the L277 area, our information on L277 comes mostly from 23andMe. Finally, there is yet another SNP called L707 which shows the same pattern so far as L672 and L673, we are currently testing all of these in the L277+ group.
(Note: in the final line of the tree picture "happened in position A or B" should be "B or C")