For the record, what Nintendo pays is different from what Nvidia pays. Nvidia doesn’t have to pay for their IP, Nintendo would have to pay nVidia for their IP. So I’m not sure what you mean by at cost.
NVidia would be paying the foundry for the node IP and ARM, but for Nintendo they have to pay for the nVidia IP, the ARM IP and the Node IP.
And of course the silicon but that goes without saying really.
Also the packaging process would be there too.
And the software
Etc.
My post wasn't very detailed, sorry about that. What I mean by 'at cost' is that any costs associated with Drake that Nvidia has to pay (so developing NVN2, designing the chip itself, ARM and Foundry IP, and of course fabrication and packaging and logistics) would be reimbursed by Nintendo, while any Nvidia expenditures not exclusive to Drake would be waived (so Ampere technology and the like). The outcome being that Nvidia makes no loss on Drake, but doesn't directly profit from it either. The amount, either per-unit or in total, which Nvidia loses out on by going this route would be considered internally as money invested in expanding adoption of DLSS (and I don't know much about things like Nvenc or RTX, but if those require specific implementation by devs rather than merely being a way to accelerate existing algorithms then this would benefit them as well).
I'm basing a lot of this on the assumption that Nvidia would otherwise be profiting around maybe $5-10 per chip. That would mean 'writing off' (not in the tax sense) maybe $150m a year, which is a lot yeah, so it comes down to if Nvidia believes sacrificing 3-4 percent of their annual profit (though around 1% of revenue) would result in a large enough lift down the line to their consumer GPU sales through wider adoption of DLSS and other proprietary technology. The math here fluctuates wildly depending on how much Nvidia profits per chip they sell to Nintendo, so the exact numbers aren't important. If they profit less per chip my theory is more likely, if they profit more then it's less likely.
Though come to think of it, that all only makes sense if Nvidia felt Nintendo otherwise wouldn't choose a SoC capable of DLSS for their next system, as otherwise there would be no incentive to leave the money on the table in the first place.
And to clarify another point, by 'a decent chance' I only meant something like 25%, maybe as high as 40% if they weren't already doing so well in the dGPU market. I don't think it's the most likely outcome, but I did believe it isn't completely outlandish. Though after my previous paragraph I'm thinking the odds are more like 10:1
On a slightly related topic: I read that Nvidia holds an ARM architectural license. Does anyone know if that license is just for the architecture, or is it a superset of lower licenses? In other words, does Nvidia need to separately license specific core designs/IP such as A78, or are existing implementations included in the architectural license? And either way, is that paid just as a lump sum or is there also a per-chip licensing fee?
I have a feeling the answers I'm looking for will be protected by NDAs, since google is failing me, but if any of them are public info please do share!