The most popular reply was:

It is not guaranteed that purchasing mining equipment will generate in its lifetime more than it cost to purchase. That depends on the future of BTC price and the difficulty, both of which are hard to predict.

"Those who believe, for whatever reason, that it will indeed be profitable, will purchase devices. But most companies selling mining hardware are in the business of designing hardware, not of speculation and running datacenters - in terms of both risk profile and expertise. So long-term they should just sell the hardware."

and also

"In the gold rush the people who made the most safe/constant returns where not the miners but the people who sold them the shovels"

This question has been answered an immeasurable amount of times.

Several groups organized ASIC production projects. Avalon, BFL, and a couple others I can't remember. None of these initial groups had the capital to finance the NRE (Non-Recurring Engineering) for the ASIC, nor were traditional venture capital sources willing to do so. The ASIC production process has a relatively a high NRE, but low incremental cost. So, Bitcoin folk organized and participated in pre-sales to finance the NRE for these ventures. They had to sell the things or they would never have been able to afford to produce the initial units.

Additionally, even if some group would've had the capital to spin their own ASIC, it would have been a bad bargain for them to monopolize ASIC mining when competing against FPGA miners. Their capacity would have been limited to some fraction of the much smaller total network capacity in the FPGA era. Also, a single party swamping the network would've earned enmity from all of the other miners and Bitcoin users who could possibly have forced them out of the network, rendering their investment useless.

Depending on your evaluations on those things, it'd make sense to sell none of the miners you manufacture, or all of them, or some of them to hedge your bets.

Also, short-term liquidity. You need to pay for manufacturing in hard currency, and you may want to hold on to your own BTC or have other reasons for not wanting to pay for manufacturing with USD converted from BTC.

Also, BFL will benefit from the share of users that will not use their equipment to the max for whatever reason and they may not have the equity required to do the upfront hardware investment.

Anyone building a bunch of good mining rigs could certainly have a decent amount of bitcoins - but lots of people don't want to hold a large amount of a currency that is risky, volatile, possibly illegal, difficult to exchange for their local currency (that their taxes and rent need to be paid in), impossible to pay their vendors in, etc.

If you want gold, mine gold. If you want dollars, sell shovels.

In reality though, the ASIC companies are probably doing a mix of both - selling the hardware until they've reduced their exposure to BTC/USD to their desired levels, then keeping the remaining miners for "testing."