Both Ace's Hardware and x-bit have posted some information regarding what we should (and shouldn't) expect from upcoming versions of nForce chipsets. First Ace's reveals the following info from a meeting with nForce's product manager:
Incorporating a GF4 Ti engine was almost impossible, because it was too big to be economically feasable. Socketed GPU's on the motherboard where also a nice idea, but impossible due to space constraints, trace lenghts for the memory, and fact MB's usually only have 4 layers, as opposed to the 12(?) layers an FX Ultra requires.
Expect 120+W (!) GPUs to appear in the near future. Though nVidia was actively researching SOI and Germanium oxide (?) to reduce power requirements, he didnt expect it any time soon, saying the graphics market changes too fast to warrant that sort of investments for now. So nVidia was concentrating on cooling solution to be able to use that sort of power-hogs.
Ouch... certainly 120W have to produce a lot of heat, as a comparison FX 5800 Ultra boards dissipate something like 75W and Radeon 9700s about 54W (taken from the same site) so you can draw your own conclusions, then again, NVIDIA has promised it won't ever use a similar cooling "solution" as the unfamous DustBuster.
X-Bit also has some related information that NVIDIA could be offering a "nForce2-MX" chipset in order to target the mainstream market and gain a few points there. According to the information originally posted here, NVIDIA would cut the dual-channel memory controller from the North Bridge of the chipset, but leave PC2700 memory support as well as 333MHz (and possibly even 400MHz) system bus.
Update: The Inquirer seems to have further details on different models that would supposedly be catered for enthusiast, mainstream, mainstream graphics, entry level graphics and business motherboards; actual products are expected by the third quarter.