There have been some rumors lately that this was going to happen. It perhaps started at HardOCP last December:http://www.hardocp.com/news/2016/12/06/amd_licensing_radeon_graphics_to_intel63
And now, he says that it's not merely going to happen, but will happen this year:https://hardforum.com/threads/from-ati-to-amd-back-to-ati-a-journey-in-futility-h.1900681/page-72#post-1042797289
So does this make sense? If it's a one-time thing, I don't think it does. But if it lets Intel shut down their GPU division entirely, that's a different story entirely.
Intel has spent a ton of money over the years to develop their own GPU architectures. It's not just fabricating the chips that is the problem. It costs a lot of money to design the architecture, make particular chips using it and fix problems, license patents from the major GPU vendors, and write drivers. Intel has spent a lot of money doing this, with results ranging from rather bad to shockingly awful.
If they license someone else's GPU, all those development costs go away. They probably get a much better product, and provided that they pay substantially less than their own development costs would have been but substantially more than nothing (so that whoever sells the GPU makes money, too), it can be a win-win. Intel did this in the past with one generation of Atom CPUs licensing an Imagination GPU.
Plenty of cell phone chip manufacturers license a GPU, generally from ARM or Imagination. And that's in the same SoC as the CPU, even. The rumor here is a multi-chip module, with an Intel CPU and an AMD GPU as separate chips in the same package. Intel has done this in the past with Clarkdale, and also had multi-chip modules for pure CPU products, most notably the Core 2 Quad. AMD has also done multi-chip modules with a number of CPUs, generally all of their high end server chips from Magny-Cours onward.
So why license AMD's GPU in particular? The markets Intel CPUs target tend to ask for higher GPU performance than cell phones provide. AMD and Nvidia are the only two proven GPU vendors for high performance GPUs, though Imagination would probably claim that they could offer the performance Intel needs, too. AMD and Nvidia also happen to conveniently write drivers for Windows and Linux, so there's virtually no additional driver creation cost. So why AMD and not Nvidia? I don't know, but it could plausibly be that AMD offered a better price.
So why would AMD do this? Don't they want a GPU advantage to drive sales of Raven Ridge? Suppose that you're AMD and you have a choice between making $50 in profit on every CPU you sell and nothing on every CPU Intel sells, or $50 on every CPU you sell and $5 on every CPU Intel sells? Even if the deal means you end up with 12% market share rather than 15% as some of the people who care about the integrated GPU buy Intel, you still come out ahead. This could also greatly reduce the risk to AMD, and keep them in business even if their CPU side has another hiccup. AMD has been actively looking for ways to monetize their GPU IP, and this is one.
Does Intel want to be dependent on AMD for their CPUs? Of course not, but even with this deal, they wouldn't be over the long term, even if they shut down their own GPU division. A multi-chip module means that you could license an AMD GPU one generation and Nvidia the next--or ARM or Imagination or Qualcomm. Game consoles have no problem with bouncing between GPU vendors from one generation to the next. Apple does the same thing.
Another possibility that I'd like to introduce as my own speculation is that it could be Apple driving this. Apple has long been unhappy with Intel GPUs and AMD CPUs, for obvious enough reasons. If they threatened to ditch x86 in favor of ARM unless something like this happened, it could provide the impetus to get something done. Of course, if that's what happened, the chip might end up being Mac-exclusive.
It's all speculation and rumors at this point, so I don't know if this is actually going to happen. But it will be interesting to see if it does.