This can vary -massively- and has almost no correlation with the amount of bandwidth you're seeing on an interface. A single 50M flow (which would account for one of your devices) could be in constant use and would, as a result, generate flows only when you hit the natural timeouts (so every 5 mins or so - 288 flows a day). At the same time a DoS attack could be generating 50Mbps of 64 byte packets, each with a different src/dst L3 and L4 tuple and you'd be staring down 1M flows per second.
Obviously these are extreme examples but the point is that the nature of the network and the type of traffic you're carrying will make a big difference. The other variables are the Netflow version - old style V1/V5/V7 vs V9, the latter of which has capabilities for aggregation, summarization and customization. The other potential point is the use of Netflow sampling (on the higher end / SP side of Cisco).
What I can advise is that scaling Netflow is usually pretty linear. Your best bet by far would be just gathering some empirical data about what's being pushed right now and sizing your infrastructure to accommodate a healthy margin over that. I can tell you that I've been in enterprise environments on the smaller side of medium that were pulling from a similar number of devices and were seeing ~120-150 million flows per week.
I can tell you that in many environments where Netflow is being pulled from a similar number of devices in fairly diverse enterprise networks (on the smaller side of medium) that I've been in shops where I was