Nvidia has never had a ton of success with their Tegra chips. Tegra 2 wasn't very popular and was pretty seriously flawed by lacking NEON extensions, and it also arrived too late on the market when dual core chips weren't such a big surprise anymore. Then came Tegra 3, which was a bit more successful, but failed to win many high-end devices, and it was quickly relegated to being the chip to get for $200 devices.
Tegra 4 took a step backwards again, because it also seems to have come pretty late to market to compete with the "next-gen" chips, but probably its fatal flaw is that it's not a chip that can be used in smartphones. In my book, if a "mobile" chip can't be put into a smartphone, then it's not really a mobile chip. This left an even bigger opening for Qualcomm to take over the market, especially since Samsung has been having problems with its own Exynos chips lately, too, and can't even use them in too many of their smartphones.
Nvidia failed to understand a long time ago that if they want to be the "Intel of mobile chips", then they needed to be at every level of the smartphone market, just like Intel tried to cover every level of the PC chip market, from Atom to the Core i7 line, and even to server chips. But Nvidia has always seemed more interested in only competing at the very high-end of the mobile market (even though they haven't always been very successful with that).
The problem with that strategy is that Qualcomm is doing it for them, and OEM's prefer them for high-end, too, because they are already working with Qualcomm for lower-end devices. Nvidia can't say the same, because they don't have low-end chips, and Tegra 4i seems like a last-minute decision to try to compete in smartphones at the high-end, but I doubt they'll be very successful with that, either.
Nvidia has also tried to make game developers dependent on their GPU technology, but that hasn't worked very well for them either, because their chips haven't been very popular. So now they are going to try plan B – licensing out their Kepler-based GPU technology, the same one that's supposed to arrive in the next-gen Tegra next year, to other chip makers, such as Samsung or others. That's Imagination and ARM's strategy, with the PowerVR and Mali line of GPUs, and so far it has worked pretty well for them.
Samsung has been having a GPU crisis lately, and they seem undecided between Mali and Imagination GPU's. If Nvidia can convince Samsung that their next-gen OpenGL 4.3 GPU's are good enough, that would put them back in the game pretty quickly, even if the next Tegra on its own won't be very successful.
Kepler is supposed to be an efficient architecture, but that's in terms of PC efficiency, with this latest generation of Graphics Card taking less power than before – and less than their AMD competitors. We don't know yet just how efficient it will be for mobile devices. I have no doubt it's going to be very powerful, probably the most powerful mobile GPU architecture, but at what cost in power consumption? Nvidia really needs to avoid another Tegra 4 failure, and not go overboard anymore in performance at the cost of efficiency.
Again, they should think of this in terms of smartphones. If that chip can't be used in a smartphone, then they've failed. If it can be used in a smartphone, then it will be much easier to simply scale it up a bit more for tablets and whatever else, just like Apple is doing with its tablet GPU's, by making them 2x stronger than their iPhone GPU's, simply by doubling the number of GPU cores.
So if it's a good GPU, then it could quickly become a winner, simply by getting one of the major OEM's to adopt it, like Samsung or Huawei. But for now, that's a pretty big if.