Q: How to store HUGE matrices?

So your RAM is no longer enough? It might help to use shorter datatypes (SINGLE, INT32, INT16, INT8) that use (4,4,2,1) bytes per element, compared to 8 bytes for double (PC platform):
  Name      Size         Bytes  Class
  A       100x100        80000  double array
Grand total is 10000 elements using 80000 bytes
  Name      Size         Bytes  Class
  A       100x100        10000  int8 array
Grand total is 10000 elements using 10000 bytes
There are two caveats, though. First, no arithmetic operations are defined for non-double classes:
  ??? Error using ==> +
  Function '+' not defined for variables of class 'int8'.
Reshaping, subscripting and transposing still work, and you can always convert a subset of your matrix back to double for computations.
Second, the standard way of pre-allocating, say, INT8 array is
This, however, creates a full-size double array first, defying the whole purpose (it is slow and it might not fit at all)! Here is a couple of workarounds (© Steve Eddins, eddins@mathworks.com):
A(1e6,1) = int8(0); % assuming "A" has not yet been defined
 - or - 
A = repmat(int8(0),1e6,1);