I could give lots of reasons. A big one is AMD is still at 32nm process. Intel has been using 22nm and is moving to 14nm (it's not easy producing 14nm, even for Intel). This hits you square in the die size, the number of transistors you can pack in, and the power and cooling requirements.
Piledriver: 1.6 billion transistors, 319 mm2 die size, 125W at 4.1-4.2 GHz (8350), 220W at 4.8-5.0 GHz (9590).
Haswell: 1.4 to 1.6 billion transistors, including the GPU, 177mm2 die size, 84W at 3.5-3.6 GHz (4770K).
But that's not the full explanation. You can see from those numbers that Intel is doing more with a lot fewer transistors (many of those transistors on Haswell are in the useless GPU) at a much lower clock speed.
You can find many other technical explanations. The 8 Piledriver cores are not independent, and on loads not optimized for their architecture, they behave like 4-core CPUs. Haswell's 8 bogocores can perform better than 4 cores on ordinary loads, because their hyperthread scheduler is pretty good at scheduling work onto idle resources. Piledriver has a bigger L2 cache, but its performance is dreadful. Haswell can do an L2 cache read in one cycle. Piledriver takes 20 cycles.
But I think the biggest reason is creativity. Intel has Ron Friedman, and AMD, well, doesn't. Ron Friedman is legendary in the CPU business. He led the coup by Intel Israel that took over the CPU division and replaced the Pentium 4 with the Pentium M and then the Core series. He became head of the CPU division and held that position for many years; Intel Israel developed the Sandy Bridge and Ivy Bridge CPUs. Since then, he's been kicked upstairs and is Intel's general manager for intellectual property, according to a recent org chart.
One man leading one group of like-minded engineers, with a commitment to making Moore's Law a reality and a work ethic like Michael Jordan's, really can make a difference. A huge, company-saving, market-dominating difference.