If each AI waits the previous one to finish, it only can be a sequential algorithm. But, it most cases, it is not necessary. Supose that each AI works from the initial state of the world. Most results should be correct, but some not, some units would be on the same title, on war times, some units would be easily destroyed, etc... ok... supose that, a second, this time sequential algorithm, is applied only over the conflictive cases assuming the non conflictive as good. The AI would be nearly as smart, and much faster.