We are on the eve of CES 2022 where NVIDIA’s latest addition: the RTX 3090 Ti is due to be presented, but the rumors are not focused on it, not even its two minor variants that will come later, but the RTX 40. And it is the case that the eyes are fixed on them, so a new rumor has ended up being confirmed and it is nothing more than a very curious casuistry which directly concerns NVIDIA: the payment in advance to manufacture the RTX 40.
There is a saying that ensures the following: “pay first, then ask and if anything complain.” Well, that’s precisely what they told NVIDIA directly, which was surprising because normally contracts work the other way around in the industry and are binding: first the number of wafers and chips for each of them is agreed and they are paid when they are manufactured and both parts are compliant. With the RTX 40 TSMC It’s a game changer and it’s been an awesome amount of money from NVIDIA.
Upfront payment is there for NVIDIA
The industry cannot cope, the investments run into billions of dollars and the companies cannot properly manage their assets as they are overflowing and applying for loans is losing money with the banks and interest which are not optimal for profits. The solution to this problem comes with pre-purchase or pre-order financing, something that could become all the rage in the semiconductor industry.
TSMC’s requirements reportedly changed completely following the company’s 6nm node high performance or N6, since it is the last with “common” contracts. And it is that for the “new” customers in large volumes, the Taiwanese have a way of proceeding which has changed: it will be necessary to pay in advance what has not yet been manufactured in order to reserve the volume.
It’s odd, as it wouldn’t apply to customers who booked wafers and chips years in advance.
Apple, Mediatek and AMD exempt from this policy
While already paying in advance is surreal for the industry, having exclusions from the same system for some customers is already a funding problem for many. It seems that everything has to do with the requested volume and in this case NVIDIA has a little problem, although it should be called “problem”.
TSMC asked the people of Huang for a deposit of $ 1.64 billion for the third quarter of this year and no less than 1.79 billion for this fourth quarter, where we do not know the figure for the first of 2022 either. What has been disclosed is that the total figure at long term amounts to $ 6.9 billion to be able to manufacture RTX 40 GPUs at 5nm.
This appears to be a maneuver to secure funding for FAB, production, and customer loyalty despite the fact that sales don’t live up to expectations (Nvidia doesn’t of course). The most striking thing about this is that Apple, Mediatek and AMD don’t have to do this because they are customers who have requested volume well in advance, so they are “rewarded” with typical contracts, while NVIDIA comes and goes with different needs depending on whether Samsung can provide them. Does this set a precedent for Taiwanese, Koreans and Americans? Intel and Samsung could take note of the current funding need with costs skyrocketing at every node.
