What really hurt AMD in the middle part of the last decade was lack of manufacturing capacity.
Intel chose to increase fab spending even as the dotcom bubble and Y2K upgrade cycle bubble had deflated. Very controversial at the time.
Even as AMD had a pretty good part--and Intel was forced by customers to respond with its own variant of the 64-bit extensions that AMD had pioneered--it lacked sufficient manufacturing power and yield (so I heard at the time from some with insider knowledge) to exploit its edge.
Getting major design wins, e.g., having a major PC maker adopt the line as the basis of its PCs, requires more than just a good chip design. An Apple, for instance, considering adopting the AMD64 line for its Macs, would be looking very carefully at whether AMD had the output to fulfill its expected needs.
We certainly know that IBM did this kind of careful scrutiny of Intel vs. Motorola when it was deciding between going with the smaller 8086/8088 versus the much larger, but more elegant, 68000. One of the issues I heard about at the time was that Boca Raton was impressed that Intel had the x86 in CAD, whereas the 68K was still on rubylith (red sheets cut up with X-acto knives!).
IBM was concerned that Moto wouldn't be able to meet its needs for a high-volume part, nor would be able to proliferate needed variants quickly enough. (Indeed, it took longer for Moto to switch to better design tools and then to introduce variants like the 68010 and the important 68020 and 68030. Intel had a good head start plus a lot of fab capacity at the then-cutting-edge lithography node. (As a detail, Intel's EPROMs were both paying the bills and generating the profits and also driving the fab capacity machine. We used to be amazed that things like the 2716 and variants were accounting for nearly all of Intel's profits.)
By now, only 3 major fab companies really are in the forefront: Intel, TSMC, and Samsung. The Japanese are mostly out of the picture, especially for very large CPU-type chips. Micron and a few others are in the niche memory/flash category. GloFo has the aging Dresden facility and one or two others, plus something in the pipeline funded by bureaucrats in Albany and oil sheiks. Time to write them off, I think.
Maybe something will emerge out of China.
