After thinking about this a little more, it's not strictly true that
all the other tech would be avoided because of NIH.
The game industry is always several years behind research in some levels, most notably around content creation. A modern, triple-A title has hundreds of artists laboring for years on all of the art for a game. All the tools to export art from Max or Maya or Z-brush or Modo or whatever are custom, and buggy, and one-offs, because every engine (except for Unity, apparently) has their own custom import format instead of using a standard like COLLADA and then baking out the models into a packaging format afterwards.
Just like CGI helped George Lucas eliminate actors, if you can build something that eliminates the need for more artists, you'll be able to sell that.
Things like procedural content generation: even open-source game engine tech like OGRE supports integration with systems like Speedtree, which is a specialized procedural foliage generator, because when you're making an outdoor scene, making trees and bushes and grasses by hand takes a lot of a real person's time: http://www.speedtree.com/
The latest version of UnrealEngine3 supports procedural building generation: you tell it how buildings are made and all of the art assets that make up corners and sides and windows and walls and such, and it'll generate a building out of all the component art, but you still have to build the component art: http://udn.epicgames.com/Three/ProceduralBuildings.html (Also, this is stuff that was in SIGGRAPH papers a decade ago.)
Procedural texture generate for art is also still really in its infancy as far as adoption goes. Besides what was done in Spore, I only know of one tools company that does it: http://www.allegorithmic.com/
Animating characters is expensive. A lot of it's motion captured, which is expensive, and requires a lot of refinement by character animators, and a lot of it's still done by hand. Creating characters and all the outfits they wear all the time is such a chore. I once wanted to use real-time game engines and real-time motion capture to put on theatre productions. If you could get the capture equipment good enough for that, and inexpensive enough to use for children's theatre, that would be a hardware win, but you shouldn't even have to do that.
This is all something that I believe could be done procedurally.
I mean, people all have a certain set of proportions, right? With a fractal variety of differences. Generate bodies and then apply procedural stylistic changes. Let artists define their character style and algorithmically riff off that. Then do the same for clothing and outfit models and textures. Content creation for games should be picking characters out of a line-up and tweaking them by turning knobs and dials, not modeling by hand. Same for monsters and other enemies. There are thousands of years of fashion design that could be boiled down to algorithms and procedural generation.
I believe you could do the same for gameplay. Some Doom map generators were excellent. Gameplay has gotten different and more complex, but so has our understand of what's fun. Generate plausibly fun maps as starting points for content generation based on the gameplay story.
All of this is several steps beyond where any game company thinks, because the development model doesn't support ongoing, repeat development. If you're an engine company like Emergent, who makes the Gamebryo tech, you'll sell to one company, and at some point early on they'll stop taking your updates because they've forked something and don't have time to re-integrate it, and then you'll never hear from them again because their game didn't sell well enough to sustain the studio.
Plus, some of what I'm talking about isn't supported by research yet. There's no minimum viable product that you could build in a Y Combinator three-month timeframe. This is like Masters in graphics thesis-type stuff. :/
(That doesn't change the fact that it's what the industry needs, of course.)