All the software infra and processing was built years ago for Google Plus, so the ongoing engineering costs are minimal. The compute might be more than Backblaze, but it's predictable (the majority happens on upload) and it's a tiny tiny fraction of the cost of daily search indexing, YouTube, etc.
The costly thing for Google is storage and bandwidth to the storage (since it's hard to predict what can be kept cold-- but recognizing faces and objects can help). And those costs, as we see, are a drop in the bucket for Google. That's why I conjecture this move is much more about marketing and company culture and less about real technical constraints.
