So, let's talk some business here.
AMD owns no fabrication facilities. AMD basically owns the home console market. AMD is somewhere between 20% and 25% of the discrete GPU market. AMD is behind Intel in fabrication technology, because the fabricators they contract out to are a generation or two behind Intel. These are all facts we can agree to, because they are either incontrovertible, or have statistical backups.
Now, let's consider AMD's position. They are a publicly traded company, who has functionally no labor costs. They design chips, test them out, and then have their IP produced by someone else. This means you've got significantly less of your resources into physical components, but it also makes you beholden to whomever can make your stuff.
As AMD, I'd look to do two things. Keep control of the console market, to act as a cost absorption sponge and never turn a profit.
Those with tons of skills are probably looking at me, and getting ready to call me an idiot. Hear me out first.
Absorption costs on IP are generally going to be huge. Running small batch fabrication (testing) has very large costs, and justifying the small army of indirect labor would make any company look bad. If you book all of that cost to a high volume, low margin, business segment the costs vanish. Sell 1,000,000 units, and that $10,000 expense is only $0.01 per unit. Yeah, the AMD costs are much higher, but $0.10 per unit is almost a rounding error on a $100 chip. Turning a profit is fantastic, if you want to remain exactly where you are. Surprisingly enough, accountants can rather easily hide money as other expenses. Can you say R&D expenditures kiddies? Invest the money into research, and suddenly that huge profit was a marginal loss. Assuming the R&D budget isn't spent, it can be rolled forward into new ventures. Who's to say that R&D doesn't need to produce a small scale fabrication facility, designed to pump out only the highest end GPU chips? Investors don't have to be any the wiser, and AMD gets to be profitable without ever being profitable on paper.
So, is AMD in trouble? Financially speaking, they've gotten into a pickle. At the same time, pickles are pretty tasty with potato chips. Chapter 11 restructuring and bankruptcy could be a useful tool. The US has allowed some crappy airlines, and criminally guilty bankers, to get a pass on this sort of thing. It's reasonable that the DoD would arrange something to keep AMD afloat, given AMD supplies them hardware. AMD could shed its debt, restructure, and wait out their foundries being too far behind Intel. Heck, an 18 month protected restructuring would suck for your average employee, but just imagine the triumphant return of AMD to the CPU game with Zen, a processor one generation behind Intel and designed for heterogeneous computing. Nothing quite like a phoenix rising from the ashes to inspire Intel to stop popping incrimentaly improved CPUs, with more and more die space cannibalized by a meh iGPU. At around the same time Arctic Islands successor could be introduced, with the supposed gains of DX12 actually being demonstrable.
AMD isn't out of the fight because Nvidia GPUs are uniformly better. AMD is out of the fight because their management is "special." All they need is someone with some chutzpah to show why they are great. Despite what has been said, Intel cannot release a GPU right now. Yes, IRIS is great for some things, but even when upped to a full card it'd perform crappy (gotta love GPU patented technologies). Half of the benefits to IRIS come from being on a CPU. There's no processing of data over a bus, minimal buffering between CPU and GPU, and no reason to believe that a GPU sized IRIS chip would offer better performance than a similarly sized GPU from Nvidia or AMD. Just because IRIS doesn't suck, doesn't mean it could be scaled up into something more useful. Sometimes scaling up something small that works leads to a much larger problem. If that's an annoying concept to grasp, then tell why Intel doesn't just triple buffer sizes in their chips to allow more data to be readily accessed.