In the previous event at TSMC's Arizona factory in the United States, including Apple CEO Cook, AMD CEO Su Zifeng and NVIDIA CEO Huang Renxun, all attended the event. As TSMC TOP customers, this has shown the ambiguous attitude of the three technology giants towards American factories in the future.

In an exchange with the US commercial and financial media a few days ago, Huang Renxun made it clear that he would adopt GPU chips directly from the Arizona factory.

According to the planning of TSMCThe total investment of the two phases of the Arizona plant in the United States will be as much as $43.5 billion (about 300 billion yuan), of which the first phase will produce N4 (4nm) chips in 2024, and the second phase is expected to produce 3nm chips in 2026, with a total production capacity of more than 600000 pieces and a product value of more than $40 billion.

You know, when TSMC initially planned the fab in 2020, it envisioned a $12 billion plant with a monthly production capacity of 20, 000 5nm wafers.

In the exchange, Huang Renxun did not forget to praise ChatGPT. After all, behind the beta version, there are as many as 10, 000 pieces of NVIDIA GPU to provide training and arithmetic support.

According to Huang Renxun's on-site display, the motherboards they shipped to the ChatGPT project had eight GPU each at a price of $200000. It is estimated that ChatGPT's latest model requires 30000 NVIDIA GPU, or $750 million in GPU sales for NVIDIA.