So what you're saying is this: Someone buying an ASIC will pay more than the value of what the ASIC will mine.
Where on earth did I say that? I would expect that the majority of miners will recover their initial hardware investment but the amount of time that takes will increase as more and more power ASIC power is thrown at the network, so miners will either need to buy additional hardware to mine a given amount of Bitcoins each month or see the amount of Bitcoins they mine each month diminish if they don't invest in more hardware.
By contrast, once their R&D costs have been recovered, those selling the shovels (ASICs) are making a known amount of profit on every unit they sell and making that profit instantly. If they're half-way competent, they know exactly at what point the market will be saturated and they'll need to drop prices to stimulate further sales, but miners themselves are going to need to go on purchasing additional hardware just to tread water for as long as the overall amount of network power continues to increase.
In a sense,
every computing technology which hits the market is already obsolete in that plans are usually already in place for the next generation. It's not a field in which you rest on your laurels for a few years and then think about designing something new after your product is no longer adequate for its intended purpose. I seem to recall Josh saying he expects that people will still be able to use their BFL ASICs in 5 years but that doesn't mean an individual ASIC will be bringing in a return which covers its power costs at that point - 5 years is a
very long time when it comes to technology.
Hell, in the 18 months I've been on this forum we've gone from people being able to mine using the CPU on pretty much any computer they already owned to the development of ASICs - it would be foolish to try to predict where Bitcoin mining will be in 18 months time, let alone 5 years, but one thing you can be certain of is that there'll always be a demand for new shovels.