Your zero and higher order poles (2nd,3rd) are very close to unity gain crossing of Gain plot resulting no gain margin. In simulation you have problem in real chip you will have much worst cases since simulation model accuracy and over PVT etc.
Your input differential pair may have cascodes on them and their Cds creates a forward path to create zeros (or similar situation where parasitic transistor capacitors causing feed forward) and also high parasitic capacitance's in high impedance nodes bringing 2nd and 3rd poles in.
I suggest first to check above and try to minimize parasitic capacitance's (play with transistor W/L), increase branch currents to reduce resistances. If these do not satisfactory or desired then modify or change your topology.