Samsung and Nvidia at loggerheads?


Faced with TSMC’s production capacity saturation, Nvidia approached Samsung to continue ramping up its AI chips. A provisional contract had been signed for the supply of HBM chips and more. However, the results of preliminary tests did not seem to satisfy the greens. At the same time, Nvidia’s move to Samsung for the next RTX 5000 was also mooted, to take advantage of the company’s 3nm lead. However, in one fell swoop after another, we learn that Nvidia has decided to drop Samsung for its 3nm chips, HBM memories and interconnect substrates. Under these conditions, this means that the greens will once again be relying totally on TSMC for all their product ranges. The yield of 3nm chips is Samsung’s main concern, and a real issue for the future of the Korean foundry division. Does this mean that TSMC finds itself in a quasi-monopol situation to the point of imposing its conditions on the entire market? Possibly, but this situation may only be temporary, as another player rubs its hands.

Nvidia Samsung 3nm
Jensen Huang, CEO of Nvidia (left) and Jay Y. Lee, President of Samsung Electronics (right), in a Japanese restaurant in Silicon Valley on May 10, 2023

Intel on the lookout to offer its services to Nvidia and others.

The projects under discussion all concern 2024 and 2025. A situation that suits Intel, which is counting on strained relations between TSMC and its partners to enable it to emerge as the ideal solution. While Intel’s current production capacity is extremely limited, the company is engaged in an impressive industrial sprint, with a policy of building or extending its fabs. A colossal effort that will materialize between now and 2025. Capacity is one thing, technology is another, and here again, Intel is in a race. The Intel 20A and Intel 18A processes are ready, and the Intel 16A is under development. The promise is to put a new node into production every year, a hellish pace, but above all a determination to regain the lead in innovation in the sector, the better to sell its capabilities to others…including Nvidia.