Put your tin foil hats away. This doesn't "crack" NTLM, it brute forces at a very high rate. The NSA has more money to spend, but are similarly limited by the hardware available at any given point.
The NSA has access to their own chip fabrication facilities. I do not know if they own their own plant, or just have secure fab space at some other company's plant.
So they could have easily fabbed something like this, or a tuned architecture specifically designed for the purpose.
It's very probable that they have and if so, it's almost certain that it would involve specialized hardware implementations (ASIC, FPGA, whatever) rather than commodity graphics hardware which is burdened with expensive and useless stuff like onboard memory and would be power inefficient
considering that the entire purpose of NSA in the first place is to provide SIGINT and encrypt or decrypt signals, it's almost a given that they're trying to the best of their ability to crack stuff.