Apple always said: "Those who are serious about software, make their own hardware"

I've been wondering since Charlie's SemiAccurate article when someone was going to put this together. Doesn't mean it's real, or really going to happen, but there are a whole lot of pieces to this puzzle that seem to fit.

If they really want to differentiate their products, Mac & handheld, what better way than standards-based GPUs that blow the doors off of everyone else's and can't be bought off of the shelf by the other vendors. It's non-proprietary since it's standards-based, so they aren't trapped by the 'proprietary' label. But, the only way anyone can compete is to do something similar, which has serious cost, time, and talent constraints.

Toshiba pull out laptops with a special CELL processor to render HD video. The CELL as you know, is based in the PowerPC core, the same core that PA Semi had license to IBM to build the PA Semi Processors.
So, If the PowerPC has already prove that they can do graphics like a typical GPU, what can stop Apple from sue the PowerPC core as a graphics engine? Imagine a chip specialized in Core Animation running OpenCL and Grand Central at the same time (a multicore PowerPC CPU).
Now, not only clones will be left behind, but increased performance will have no match in the PC world. Reduce that to 32nm process and you have our iPhone's SOC.

Didn't Apple try this years and years ago with the Pegasus CPU project or something?

Toshiba used the Cell in the SpursEngine, which is a co-processor, not a GPU (the laptops they launched it on even came with a discrete 9600M GT). There's a big difference. In this case, the SpursEngine operates alongside the GPU and the CPU to handle specific graphics-related tasks, but it is not a general purpose graphics processor. It handles things like upscaling videos or transcoding, but just as CPUs make poor GPU stand-ins, so too would one of these co-processors. Toshiba also admits that the technology will chew your battery up and spit it back out (my words, not theirs, of course, but what else can you call 40 minutes of battery life? Citation: http://www.reghardware.co.uk/2008/06/17/toshiba_rebrands_spursengibe/), and since Apple has been all about performance-per-watt in recent years, I seriously doubt they would turn to Cell. The Power line, and from that the Cell line, have never been designed with energy efficiency in mind.

Papermaster worked on Power architecture most recently. I think he was hired more because of his experience and capability, and not because of any inside knowledge (plus, IANAL, but I'd imagine IBM would have a real good case for appeal in the recent court proceedings if it turned out he used inside knowledge from IBM). You're not going to have a VP laying down paths for logic flow of these new circuits on a daily basis, after all.

It seems odd that they would push through OpenCL, a new standard for processing on GPUs, and get it adopted by Intel, ATI, and nVidia, only to turn its back on all three and develop its own hardware. I can only think of two reasons to want those companies to adopt the technology: 1) because you'll be using their hardware and want the technology available; 2) because you want the adoption rate high so that your fledgling hardware will still have people developing software for it even if they aren't developing specifically for your hardware. While #2 is certainly possible, it seems like a rather roundabout way of doing things.

To me, there seem to be a few options here that do make sense, all things considered. There's the iPhone route, where we consider them developing either additional co-processors, which makes LOTS of sense, both in terms performance-per-watt (so long as you're not using the power-hungry Cell architecture) and raw computing capability. They might also be developing discrete GPUs for the iPhone or an iTablet, and while that probably makes sense in the long-term, right now, there doesn't seem to be a need for it since the iPhone hasn't converged on the laptop market enough for that to matter yet. They could also be developing co-processors for things like the AppleTV, or the other computers they have. I'm not sure of the likelihood of that route, so I won't comment.

Yeah, I would say that's a streeeeeeeeetch.

Unless PA Semi is REALLY working on some cool stuff behind the scenes then I don't see Apple dumping Nvidia despite Nvidia's manufacturing problems. Perhaps Apple has some inside information on intel's larrabee roadmap and likes what they see.

Doesn't seem likely. Virtualization is one key to getting OS X adopted in the enterprise workplace. Does Apple really want to be in the business of writing proprietary graphics drivers for Windows and Linux in perpetuity? I doubt it, and you can bet your life that Microsoft won't do it. But yeah, stranger things have happened.