I personally do not understand who would pay so much for such an FPGA. I think going for an ASIC is better instead.
Even several of those FPGAs is cheaper than ordering a few ASICs only to find out a design error made them useless. The biggest market for FPGAs of that level are emulating ASICs so that errors can be found, fixed, and retested on very short notice. Only once the FPGA emulation is running flawlessly do they go to the step of ordering some prototype ASICs to test.
Interestingly, low to mid range FPGAs have been replacing ASICs in certain applications, one good example is motor control ASICs in electric and hybrid cars. Some factors include being able to make more substantial fixes after the product is out to market, the high cost of making ASICs on smaller process nodes (greatly cutting away at the cost advantage, especially for smaller chips), and having more flexibility in reusing designs for future products. We're also starting to see microcontrollers like PSoC and some PICs gain some FPGA-like programmable logic blocks.