AMD reveals Fusion CPU+GPU, to challenge Intel in laptops

You could take advantage of the high-bandwidth, low-latency connection between CPU and GPU to get higher performance, especially in GPGPU applications, but GPGPU hasn't really taken off yet. There's also the limitation that now CPU and GPU are sharing the same heatsink.

You could use it to offer a low-end system with good integrated graphics, because there's no longer a penalty for the GPU sharing the CPU's RAM. However, this approach pretty much requires a new socket, which probably eliminates any potential cost savings from not having to buy a discrete GPU. This approach also sacrifices some in flexibility, as adding a discrete graphics card later on makes the on-die GPU worthless unless you have a good asymmetric SLI/Crossfire system, and even then you're limited to the features supported by the on-die GPU.

The approach taken so far by Intel and AMD is to use it to make a better low-power platform. Intel's obviously been way ahead here. Ever since the Pentium M, AMD's had trouble keeping up in the mobile market. ATI's GPUs haven't been that great for the mobile market, either. It was obvious in 2006 that this was going to be the hardest approach for AMD to pull of, but it also seems to be the only one that might pay off.

Because "adequate" can have vastly different meanings, depending on what you use your GPU for. I use it mostly for eye-candy at the GUI level. I play a bit with Google Earth and that's about all. My Intel GMA can do that just fine.

If, however, I were into heavy gaming, I would find this setup unusable (down to the fact I don't run Windows).

I am not sure how well an Intel GMA would fare if it were used as a number cruncher. I suspect it wouldn't excel, but, compared to x86 vector hardware, even a lowly GPU should hold it's own well.