I encountered a similar situation in which the level-1 intercept I was studying did not significantly vary across level 2 units but I detected a cross-level interaction and had a theoretical reason to do so. Snidjers and Bosker discuss exceptions to the precondition of significant random intercept variance on pg. 106 of their book "Multilevel Analysis" (2nd ed.).

Given your a priori hypothesis regarding the presence of an interaction, there is no need to observe a large ICC to proceed with your cross-level interaction test. For more detailed information on this issue, see the following article available at http://mypage.iu.edu/~haguinis/pubs.html

Aguinis, H., Gottfredson, R. K., & Culpepper, S. A. 2013. Best-practice recommendations for estimating cross-level interaction effects using multilevel modeling. Journal of Management, 39: 1490-1528.