Ricardo,
I know that you've already received much of this information from our great SAS Technical Support team but I'll repeat it for the benefit of other forum participants. In your case you'd want to use the DATA= option of PROC NLP to import the data on parameters. However you'll still face the challenge of building and using a SAS data set with millions of variables. Fortunately there's a better way. PROC OPTMODEL is our new "flagship" procedure for linear, mixed-integer, quadratic, *and* general nonlinear optimization. Due to OPTMODEL's much greater flexibility in working with input data you should be able to use many observations rather than many variables to specify parameter values in a SAS data set. OPTMODEL delivers many other benefits vs. PROC NLP as well, including far greater clarity in its model-building syntax, improved and updated solvers, and greater flexibility in working with the results of the optimization process.
... View more