if it takes a million years for 1 video card to search, then a million video cards will be found in 1 year. Unlike mining, when it is extracted regularly for a little bit, the process of searching for keys takes a long time, but the reward is more significant.
The average video card would cost you about $200, so that's $200,000,000 in cards alone, it's safe to assume that you will need 30% of that in other pieces of equipment (apparently a GPU isn't a stand-alone device) so that is another $,60,000,000. You will also need a few million dollars in labor and shit.
The average GPU would consume about 120w per hour to do the "guessing" job, that will be 2.9kwh a day, the average electricity cost is 5 cents per kWh, so you will pay $0.145*1,000,000 = $145,000 daily, in other words, $52,925,000 a year.
A million GPUs would need 120 megawatts of power, and perhaps another 10MW to cool that down, you will need to get your own transformers and build your own infrastructure to attempt that kind of shit, you will be looking at close to 1 billion dollars in total by the end of the year including all the other expenses, that is close to 0.5% of the total value of bitcoin, you will spend a 100,000BTC in an "attempt" to hit a collision of some type, only to realize that those funds were moved to a thousand different addresses, or at best case scenario, whatever you may have found isn't worth a 100,000BTC anyway.
On the other hand, if you use those 1M GPUs for mining you make $620,000 in profit (after the power bill of 5 cents per kWh), so in a year you make 1/4 of a billion-dollar that is clean money without the chance of actually "stealing" some else's bitcoin.
One-way hash functions are built in a way that makes them economically less incentive compared to when you harness the same resources to make money else way, so if profit is the only "goal" it makes no sense to attempt that unless you want to gamble.
