According to foreign media, NVIDIA is preparing to produce its next-generation Ada Lovelace GPU on TSMC's N5 node, and this is most likely the new RTX 40 series products. TSMC's advanced process capacity has always needed to be booked in advance, and NVIDIA has paid a lot of fees in advance in order to ensure the smooth launch and delivery of its new generation of graphics cards.
NVIDIA spent $9 billion in advances for inventory and future products in the third quarter. However, for TSMC, which accounts for 53% of the global wafer foundry revenue market share, it has accepted an advance payment of $5.44 billion from at least 10 customers, including AMD, Apple, Qualcomm and so on. So these actions made by NVIDIA still can't ensure the smooth delivery of the RTX 40 series.
Why does NVIDIA love TSMC's 5nm process? According to the data provided by Digitimes, in the 5nm process, TSMC jumped to 173 million units per square millimeter, significantly surpassing Samsung. Only from the transistor density indicator, TSMC's 5nm process and Samsung's 3nm process are almost at a level. Although Samsung's process technology is similar to that of TSMC, there is a big gap in the number of transistors.
Not only is the number of transistors more, TSMC's 5nm process wafer revenue accounts for about 20% of its overall revenue, and the leading nature of this advanced process is unmatched by other foundries.
However, for TSMC's 5nm process capacity, there are also quite a few competitors. The old rival AMD has announced that the Zen4 architecture CPU codenamed Raphael will be built based on TSMC's 5nm process. In addition, it is expected that the graphics chip part of the GPU series based on the RDNA 3 architecture will also use the TSMC 5nm process in 2022, so the 5nm production capacity is tight.
Although TSMC announced in 2020 that it will invest tens of billions of dollars to build a 5nm plant in the United States, it is expected to be put into operation in 2023, when the monthly production capacity will reach 20,000 pieces; however, the factory is still under construction, and employee recruitment and training are also being carried out simultaneously. So when it will actually translate into capacity is still unknown.
For the technology community, the adoption of advanced processes in application industries such as smartphones, HPC, and 5G is a major trend. Advanced processes mean new technology research and development, transistor density superposition and increased investment in production lines, so the corresponding cost increase is also considerable; so large wafer foundries such as TSMC also need to lay out the next generation of advanced processes in advance to meet the long-term strategy of customers. For chip giants such as NVIDIA, early reservation of production capacity is undoubtedly the key to ensuring the smooth listing of new series products.