Nvidia always said that would make PhysxAPI available to other GPU.
AMD working with Intel for Havok Support with the CPU's, now if they work with Nvidia to get Physx on the GPU as well then covers all the bases.
AMD and INTEL will have gpu's intergrated on every cpu in 1-2 years when we hit the 16 core level maybe(possible 8 but unlikely). that means each and every last cpu sold in 2 years will support HAvok out of the box, Physx is dead, completely and utterly. No one will buy a 2nd gfx card to not use in SLI/Crossfire, Physx actually had a better chance as a 3rd party card. I might shell out £50 for a card in another slot if it was any good(it wasn't). But while its using a 2nd gfx card, if i buy a 2nd gfx card considering 99% of games I'll be using sli/crossfire, I wouldn't ever remotely bother to switch it to PPU mode and lose framerate and the expense of playing a couple crap levels and some ridiculous and not realistic effects.
Offered the choice of a 2nd gfx card for, graphics, or as a PPU, 99.9% of people would use it for graphics, so they failed massively. Except they didn't, Nvidia bought out Physx to let them die, rather than spend 10 times the cost competing with them for the next decade. Business wise its much much cheaper, its as simple as that.
Intel vs Nvidia, intel are about 7 zillion times bigger, Nvidia can not compete, and they've been irritating Intel lately aswell so Intel are even more opposed to helping Nvidia and more likely to do things to screw them.
Physx was dead when it started, Anandtech suggested a LONG time ago that the people behind Physx had this gameplan the whole time, become irritating enough and slightly embedded into the market so someone buys them out for a decent profit, and that was their goal the whole time. Thats why the marketing campaign was always promising things the hardware could not do, the appearance that Physx was fantastic was more important than it being fantastic, as long as people "thought" it would be useful, Nvidia/Ati would need to compete, thats what led to Nvidia buying them out.
Every single useful thing you can do with real physics, you can do with estimated cpu friendly physics. How walls blow up has NOTHING TO DO WITH THE PHYSICS ENGINE. Every wall that can be destroyed must be designed to do so, a wall that explodes in 5 pieces takes 50 times as long to design as a wall that doesn't destruct, because clipping issues, the various parts, making it explode to look decent, making it explode decently, making the parts fall to the ground and not dissappear and a whole lot of other things. Levels are fully destructable as games would take 5 years rather than 2-3 to make. This is why we've only ever seen 1-2 "physx/highly destructable" levels per physx supported title, there is no time to make more.
New engines, new design tools, and things to speed up the level building process is what will bring more destructable levels and more realism to games, nothing more, nothing less.
ALso, Nvidia will still look for ways to make back some cash on it, even if its never used getting ATI or anyone else to buy a licence still means money in their pocket, having people use the basic physics engine saves others time and puts a little more money in their pocket so Nvidia will still "market" it as useful, till its finally dead.