Not to mention H.264 was designed with hardware decoding in mind, while Ogg Theora was not.
From reddit: http://www.reddit.com/r/programming/comments/8yhi6/decoding_...
Theora is derived from VP3 and neither were designed with hardware implementations in mind. VP3 uses funky high precision unsigned multiplication because it was fast on a Pentium II MMX back in the day. Theora carries this over and it makes hardware implementation a pain in the arse and virtually unimplementable in some cases ... H.264 was specifically designed with hardware implementation in mind and is thus pure 16-bit mathematics, at least through to Main Profile.
It would seem that for the embedded market, Theora is not going to happen anytime soon.
For starters "designed with hardware implementations in mind" is a double edged sword. It seems to me that H.264 used this as a crutch to a degree that maybe only makes sense if selling decoding chips is part of your revenue stream.
But it appears that even "hardware" implementations of H.264 don't even use ASICs anymore. They just use a seperate, but generic, chip with DSP. It's sometimes treated as a hardware black box by the rest of the system but inside it's software:
http://lwn.net/Articles/337101/
There's an OMAP version of Theora being built right now, (sponsored by Mozilla I tihnk) but I can't find the link right now. That should reset people's expectations somewhat in this regard.
http://archive.netbsd.se/?ml=mozilla-dev-planning&a=2009...
There's a summary post at the very end which says, Theora on portable hardware with DSP acceleration is very doable, and that they have a suitably qualified volunteer to do it.
"On performance:
* Right now we're getting 1-2fps with decoding video (wasn't sure what size, though?)
* Some people on the call had done testing and it's possible to do 24fps at VGA resolution of Theora decoding, not including YUV -> RGB conversion and without the painting overhead. This was on a 200mhz ARM chip. So we know it can go fast.
* It's very possible to port that decoding code to a DSP chip - someone did it at broadcom and was able to get 720p decoding at full speed on a 125mhz DSP chip. So we know this is possible.
* In theory because of the design of Theora we should be able to do better than H.264 in terms of performance and be able to match on power consumption as well. "
Here is a mail confirming that the work is underway: http://www.mail-archive.com/[email protected]/msg15730...
So, I don't necessarily agree with this reasoning, but I'd suspect it goes like this:
The hardware point is quite valid; Apple doesn't manufacture all the hardware in, say, an iPhone (they buy lots of stuff off the shelf and assemble overseas), and would probably like to avoid the expense involved in not only finding and/or designing a Theora decoder but also integrating it into a space- and power-constrained device. And they've invested too much in H.264 to just drop it, so they'd have to make both work, which is even tougher.
The patent concerns, meanwhile, are probably less about minimizing exposure to possible patent claims and more about ensuring that Apple doesn't have significantly more exposure than any other major player. They have that situation right now with H.264: everybody's on a level playing field with respect to the known licensing, and if submarine claims pop up, Apple's not the only company vulnerable, so they wouldn't suddenly be at a huge disadvantage.
With Theora that situation doesn't exist yet; nobody's yet taken the big plunge into wide-scale deployment of Theora in the market, and Apple (and others, including Nokia) are skittish about being the first to do so, since first to market may end up being first to get sued, and then nobody else will want to use Theora, leaving them all alone exposed to patent claims. I suspect that if somebody else steps up and does Theora in a big way first, Apple will feel more comfortable following, since then they'll once again be maintaining rough parity of exposure.