This is a difficult question to answer because:
- the network isn’t running yet, so there is no data
- even when it is running, things will change as the network grows and the volume of data storage, communications and access changes
Two key factors will be the rank of your node and the volume of data it is given to store. The higher your rank, the more data you will be trusted to store, and the more data you store the more chances you will have to farm Safecoin. But whenever you get this opportunity, you will be competing with the other vaults who hold that same chunk of data, and only the first to provide the data will be rewarded. So in addition to rank and storage capacity, speed will also help.
So what are the factors that affect rank? Rank is a measure of how useful and reliable a node you are, which assuming you are a well behaved node, comes down mainly to availability over time. There is a FAQ video where David Irvine discusses this if you want to dive deeper.
A final factor is cost. As a farmer, it’s no good earning all the Safecoin if you do it by spending more than you earn, through buying and powering fast hardware and massive disks. So low power consumption could well be an important factor in being a successful farmer.
This is good, because it is unlike bitcoin, where spending lots of money on dedicated mining hardware allows a few companies to corner most of the the income. That’s both inefficient (because it encourages maximum energy use) and unhealthy (because it causes centralisation).
In SAFE much of the network will be provided by users sharing spare capacity on their existing machines. The farming costs for these people are very low: they don’t have to buy extra hardware or spend on energy to power a machine that is already switched on. This immediately limits the potential benefit of being a dedicated farmer - who will have to compete not just on performance, but to do so at low cost in terms of hardware and energy use.
So if you plan to buy and run dedicated hardware, you’ll want to find a sweet spot between setup and running cost, and competing for rewards based on availability, performance (response time) and storage capacity.
I hope you can see why this question is hard to answer! And why you need to do your research and make choices based on that, and on what others suggest - but also realising that no one knows.
Personally I’m going to set up at least a couple of small low power SoC (system in a chip) devices: for this I like the Odroid systems from Hardkernel, and have built the MaidSafe code on these and connected them to testnet2 - but there are other choices as well. I like the Odroids because they are low power, reasonably fast, and quite cheap.
You’ll find discussions about all the issues I’ve mentioned by searching for “farming” or “hardware” over on the forum at http://MaidSafe.org. I’ve also posted a bit there about the ODROID-U3, but hardkernel just released a new model, the XU3-Lite that I might try too.
My advice is to focus on availability, and low cost & power in the beginning. Either use hardware you already have and is already on most of the time, or buy something cheap and low power that you can leave switched on without worrying about the cost. Once the network has been live for a while things will change, and also become clearer, so later on you’ll be better able to decide whether it’s worth spending more on farming hardware, or sticking with a basic setup, perhaps just adding some extra storage.
(The above taken from my answer to a question on reddit 10th November 2014)
I always feel good but end up cash poor, I am sure I win, but it’s certainly not obvious 
