In the past, a mobile site had to be built using carrier-specific markup to properly function for that carrier. However, now it is possible to use XHTML to build a cross-carrier site, as all Japanese 3G handsets support some form of it [1]. Unfortunately, it is not the same dialect of XHTML: Docomo use i-mode xhtml (based on XHTML mobile profile), au uses XHTML basic, and SoftBank uses XHTML Mobile Profile 1.2. Nevertheless, although there are some differences between these versions, they all share a common base, making it possible to make a site that will function more or less the same across all carriers.

Given that not all handsets support XHTML, if you were to build a mobile site using XHTML, what percentage of handsets would it support? I've scoured the web for the answer to this question, but haven't been able to directly find the answer anywhere. However, as the number of handsets subscribing to 3G vs 2G plans are available, if we assume a handset supports XHTML if and only if it is 3G, we can use these numbers to find the answer.

In total, over 95% of Japanese handsets natively support some form of XHTML. Almost all au handsets support XHTML natively and furthermore the au gateway handles conversion of XHTML to HDML (the markup that older au handsets used). As SoftBank will discontinue its 2G service on March 31, 2010, current subscribers will need to migrate to the 3G service (and handsets that support XHTML). Finally, although Docomo has, relative to the other carriers, a large number of handsets that are not XHTML compatible, Docomo claims that i-mode HTML (which older Docomo handsets use), and i-mode XHTML is mostly compatible. Therefore, I would suggest that if you are building a site for the Japanese market, you do so using XHTML.

[1] Although many Japanese web sites say this is the case, it does not appear to be technically true.  According to Docomo, handsets from the FOMA  2051V, 2002, and 2001series do not support XHTML.  However, we can assume these handsets are few enough to make this generalization.