Why is this so complicated for people to understand?!?!
Apple needs to continue to develop their laptop and convertible business. They need new products in competitive form factors. This product enables them to produce very thin laptops and convertibles in the same markets where MS is moving the surface, surface book and other products. This will give Apple (and potentially MS) the product that they need to enable them to tightly package high performance graphics, high performance CPUs, and large enough batteries to continue to gain overall capabilities. Currently, if you want anything above minimal graphics quality at middle of the road resolutions, you have to have a dGPU in your mobile device. That dGPU takes up a lot of board area in an already tightly packed system. This compacts all of that into a slightly enlarged CPU area and concentrates most of your hot spots into one area, making it convenient to route to the edge of the case for heat exchange. This is perfect for that area.
Why didn't Intel just add an HBM2 stack next to their existing CPU cores with integrated graphics? Intel likely has a disadvantage with respect to IP there given how much less comprehensive their IP portfolio is for graphics. In addition, it can be demonstrated that intel's integrated graphics on their processors are less efficient with respect to raw performance than their AMD counterparts when measured against die area, power draw, and other metrics. To make a chip that would be competitive with this AMD solution, they would likely have to make their die prohibitively large, reducing yields and affecting the rest of their product line (remember, Intel tends to share their actual dies through many different products, disabling various parts as needed. Using a monster die for this solution would make the rest of the stack too expensive to exist in the market). This solution lets them use a die that they already have in use in many other market segments in a product that is targeted in a segment that is outside of their volume areas.
Why does this work for AMD? It works because "money". AMD needs the revenue, especially in the RTG. This moves a custom product (read, expensive) to a market that they wouldn't be in any way. These are sales that AMD would not realize without this relationship. Because of the likely capability gap between this product and anything that AMD could COMMERCIALLY PRODUCE IN VOLUME, it won't be a direct threat to their Ryzen mobile line. That line is expected to be sold in laptops and convertibles that are in the $500-$100 market. This combined product with Intel would be in the $1000+ market (probably well into it). Ryzen mobile is for casual gaming at 1080p in low-to-medium detail. This intel product is for 1080p gaming at maximum detail, and higher resolutions with still high detail levels.
What this also offers is the possibility to produce professional level thin laptops that are aimed at the CAD/CAM/content creation crowd that are fully capable of executing most of the functions that their desktop workstations can do, save for some of the most extreme scenarios.
One thing to keep in mind, this AMD graphics core will likely be produced in Intel's fabs at a current generation process. This means that it will likely be more power efficient than the existing vega chips. I can see this thing being capable of running at up to Vega 56 levels of performance for short bursts with a robust cooling solution (like you might find in a larger, mini-desktop situation) and adequate power delivery. However, since this is targeted at mobile and small portables, it would likely be more of what you would imagine a "Vega 48" might be. The limits on this package will definitely be thermal and power delivery related. Remember, better process for the GPU than desktop vega, and likely a slightly more efficient HBM2 stack as well.
For AMD to have done this on their own, they would have had to develop the package, the custom interposer (as they don't have EMIB), and done a lot of work getting it into the market. With Intel's name behind this package, it will be easier to get it into the market. Ryzen Mobile can't really fight in this market with the current die due to limits on power, thermals, and most importantly, memory bandwidth. HBM2 for them would make their package too large for mobile. The next best thing is going custom die, which would be huge at 14nm, would have to entail a much larger GPU section, would require quad channel DRAM, and would still be unwieldy for manufacturers to implement for a second tier company. Ryzen Mobile has a LONG way to go to get into the market where this will compete.
This is a solid win for RTG. This is a solid win for AMD as a whole.
Now, the big question is, how does this affect Nvidia? This will muscle them out of the lowest end (and highest volume) section of the thin and light market. Their dGPU parts will have to be cheaper to keep volume and entice manufacturers to still produce for them there. Their only viable solution here would to be to broker the same exact deal with AMD's processor division. Can you imagine a Ryzen Zen2 or 3 processor on an interposer with an Nvidia GPU with HBM2 or 3 by then competing against intel here?