The nature of Arellano-Bond estimation is to create a huge number of
instruments, each of which is a separate variable in Stata terms. Try
constraining the lags allowed for GMM instruments to, say, lags 3
through 5. That should drastically reduce the number of instruments
and make your problem feasible.
A-B is really made for "small T, large N"; you presumably have 50
states and about 33 quarters per state. That is a huge number in
terms of A-B, which often is used for T=7 or 8.