THe problem is, both Nvidia and TSMC have released multiple news reports in the past 6 months talking about 40nm being fixed, being ramped up massively, Fermi being in full production, Fermi yields NOT being an issue at all, etc, etc, etc.
Its called, appeasing the analysts who somewhat control stock prices, which have dipped since their financial results. AS I said, while they appear positive, a bit of profit and increase in revenue, they are far worse increases than Intel/AMD for the same period and they've increased because they went down so significantly last year. Lots of analysts are rating Nvidia badly, trying to get the word out stock won't be an issue when they finally launch, well, thats a very important goal, for TSMC aswell, if its true or not, who the hell knows.
THe problem with rules changing with a MAJOR 40nm fix being found, is that Nvidia clearly will not be able to apply such a massive change, it would be incredibly unlikely such a change would be fixable in a simple stepping change, even then thats a 2 month respin process and another 6 weeks for volume production. So Fermi, is smegged in that sense. THeir lower end derivitives might have better success, they've not been sent to have the first spins made, so they are still at the point they can change the masks and do a full redesign of the layout before they send the design in, though that still takes time and after they do so, they'll still be 3 months away from getting any chips back with the tape out and volume.
The low and mid end will still be 60% larger than AMD, so still have lower yields(its theoretically possible, but the bigger the chip the lower the yield has always been true in reality), because the size of the core increase will stay similar most likely, with a 192-256 mid end, and a stupid low end. SO they again will be more expensive with not great yields in comparison.
Apart from the likelyhood of complete BS from TSMC about yield fixes, the most likely ones are simply giving better design rules out, to compensate for issues, and I wouldn't be surprised if they basically said, here's what AMD did to negate are utter uselessness, go do it yourselves. IE leave more space for our variable size transistors to work, and use 2-4via's per transistor, rather than one. AS I said though, Fermi needs a proper respin to use those idea's, not a small quick respin. Its highly unlikely they've got their immersion lithography suddenly working more accurately. THe equipment they've brought lately is for 22nm and beyond, not 40-28nm.