I'm aware, but thanks for posting nevertheless. I've actually read Gentry's thesis. Last I looked into FHE though it was something like 14 times to 100 times as inefficient (either in time or space depending on the scheme) as operating on unencrypted data.

Now things may have changed since then, but I'd imagine it's not yet gotten down to 1.X inefficiency multiplier regardless of the FHE scheme you're using.

That would increase their computation costs by a fair bit, it would be more expensive to run the same amount of computation on their cloud using fully homomorphic encryption, even without taking the engineering costs on your side into account.