There is one way in which this is worryingly similar to the Conroe situation: AMD failed to win the performance crown in spite of having a new technology that the competitor lacked. In this case, it's HBM. In the case of the K10, it was the on-die memory controller (Conroe still used a classic front-side bus). When Intel integrated the memory controller in Nehalem (which could basically be described as Conroe+IMC), they took another big step ahead of AMD, and AMD was never really able to catch up. What will happen when Nvidia starts using HBM2? In that case, AMD has to compete on a raw architecture basis, but die-shrunk GCN 1.2 doesn't seem like it is going to be competitive enough with Pascal (Maxwell+HBM2).
That said, AMD's CPU business still managed to do OK for a while even after Conroe's release. It was a serious blow, but not an insta-kill. Thuban was a decent budget alternative to Nehalem, especially if you did lots of multi-threaded work, and it didn't do too badly on single-threaded tasks either. It was Bulldozer that really torpedoed AMD's CPU division. Billions of dollars of R&D spent on a product that in many ways was objectively worse than its own predecessor. AMD would have been far better off with a straight die-shrink of Thuban followed by incremental tweaks.