There's nothing fundamentally difficult about it - it's just a bunch of discrete logic, a bank of DRAM, the CRTC (which is not hard to find) and a character ROM (which is just a matter of writing the right data to an EEPROM or similar). The schematics are in IBM's documentation and not hard to find. The biggest obstacle, I think, is just that it's a very large card with a lot of chips on it, so laying out the PCB and soldering it together would take a while and it wouldn't be the cheapest board to get manufactured. Would you want to be true to IBM's original schematic or would you substitute some modern parts to make it cheaper and easier to build? Putting all the discrete logic in an FPGA might be a good way to do that. The CRTC too (though talk to me before you do that to push me to finish making an accurate model of the CRTC's behaviour first, so that we can be sure the FPGA-based design works with future demos). Indeed, as Zup said, the entire card could probably fit into an FPGA with just the DAC for the composite output being outside it.
I guess the reason that it's never been done is that CGA cards (even genuine IBM ones) aren't that rare or difficult to find, so there hasn't been any demand for a modern replacement. I think it would be an interesting project, though, and I'd be glad to contribute any help I can if you decide to take it on.