I don't understand why everyone writes about billions of keys per second. It's really slower than you think, at least for now.
Calculating sha256 is really just a tiny fraction of calculating an ecdsa keypair.
While vanitygen only does 1 EC addition, calculating an ecdsa key from seed needs at least two full blown multiplications of big numbers.
Long story short, CPUs only can generate about 1,000 keys a second, modern GPUs can calculate 20,000 keys a second, tops.
Not billions, thousands. 6 orders of magnitude smaller than billions. 1,000 times slower than vanitygen.