For a while now, all the rumors have been converging towards Nvidia abandoning Samsung Foundry for its next cards (RTX 40 and pros). But to change cremerie, Nvidia (like Intel) should comply with the requirements of TSMC. The Taiwanese is relatively “flowing” for its historical customers like Apple, MediaTek or AMD. On the other hand, it seems that for new entrants, who request wafers and especially demand a large volume, TSMC requires a substantial deposit to ensure a place on its production lines.
Nvidia (but also Intel) finances the increase of TSMC’s capacities
According to Nvidia’s figures, a down payment of US$1.64 billion has already been made in the third quarter. In total, before receiving the first 5nm chip, Nvidia will have to pay a sort of “super down payment” of 6,9 billion US dollars In fact, TSMC’s new customers allow it to finance its investments to increase its production capacity. Recently, Intel had to comply with the same requirements, even if communications on the subject are more discreet.
TSMC’s expertise far ahead of Samsung?
The Taiwanese company is charging for its expertise and seems to want to take advantage of the situation. A blatant example of its technological advance was revealed with the confrontation of 4nm TSMC vs Samsung for the latest generation ARM SoCs. The MediaTek Dimensity 9000 was compared to the Snapdragon 8 Gen1. Both SoCs use exactly the same configuration for the CPU, but by using TSMC’s 4nm the MediaTek SoC can achieve a higher speed (+16% performance) and above all consume 14% less power compared to Samsung Foundry’s 4nm. Given the arrival of the new AMD GPUs and Intel’s entry into the market, Nvidia had to put all the chances on its side.
