The big problem with MRAM last time I visited the subject was mass production and stability of the memory. These chips have been pumped for years but always seem to get delayed.
Cypress has a product spec sheet for a 256K chip on their web site, was claiming to begin shipping, and two years and nothing has shipped.
IBM has had prototypes for years. They hit a snag when their production run had a yield of zero.
Cypress also now claims that the bits in their MRAM aren't very stable and have some elaborate error correction fix called soft error. I have no idea if that is a feasible or infeasible idea.
Haven't followed the subject much since. Gee, maybe this time it really is different.