It always makes me very sad when abandoned next generation Amiga hardware design gets mentioned.

The AAA chipset was in development right from 1989 to 1993 and it is rumoured that some working chipsets did exist in development versions of A3000s. By 1993 it is rumoured that the chipset was close to completion, but still contained many bugs that needed to be ironed out. However instead of spending some money to complete the Alpha and Beta phases of the chipset, Commodore instead completely cancelled their development in favour of a newer chipset design called Hombre.

The sad thing is that the AGA chipset was also waiting in the wings long before it was actually launched in the A4000. Some rumours suggest its design was completed before the AAA chipset development begun in 1989, meaning the AGA chipset was ready 3 years before it was finally used.

Had the AAA chipset been completed (I can't see why it wouldn't have been ready in 1994) then a new big box Amiga (A5000?) and a new version of the A1200 (A1400?) could have easily been launched in 1994/5.

The troubled finances of Commodore US were probaby the real reason for the cancelling of the AA chipset development in 1993, rather than to concentrate on different, newer, chipset development.

The Hombre chipset, supposedly replacing the AAA development was however very interesting and really shows what the Amiga could have become by the late 1990's had it continued.

Hombre was to be a true 64bit multimedia system with a 3D graphics chipset and fully RISC based architecture. It was being developed with HP and estimated to be completed in just 18 months.

The chipset was to feature a DMA engine with blitter and 3D texture mapping (plus gouraud shading), a 16-bit sound processor with 8 voices, and a Retargetable graphics system supporting 1280x1024 32-bit resolutions.

You can read more about Hombre here and here.

An interesting quote from Tom MacDonald, the designer of the AAA Monica chip, answers your question the best:

"We were on Rev 2 of most chips by the time the project was shut down. Much of Andrea had been characterized; we ran high resolution displays, copper and blitter, 24-bit modes, etc. On the other hand, only five of the 16 bitplane pointers worked, Monica's colour lookup table was all wrong, and a bug in Andrea prevented the testing of any read registers in the other chips. There was more work to do."