NVIDIA’s next-generation GeForce RTX 4090 graphics card based on the Ada Lovelace architecture was released on October 14. It adopts TSMC’s customized 4nm process, coupled with new technologies such as the third-generation RTX optical tracking architecture and DLSS 3, which greatly improves performance. After the first round of sales, most of the GeForce RTX 4090 graphics cards on the market are out of stock, and you need to make an appointment to buy them.
According to TomsHardware, Nvidia has moved the GeForce RTX 4090 production capacity to the H100 computing card based on the Hopper architecture, which may be related to some recent new measures in the United States, which led to Nvidia’s decision to transfer production capacity. Nvidia’s current export license allows it to continue exporting A100 and H100 computing cards, with a temporary deadline of March 1, 2023. It is estimated that Nvidia will sell A100 and H100 computing cards as much as possible before this time point.
For TSMC, this is not a difficult task. After all, Nvidia’s customized 4nm process still belongs to the 5nm process node, and there is no problem in transferring production within a specific distribution range. From Nvidia’s point of view, however, things are a little different, which means that a change in sales direction will ultimately affect R&D as well. From the perspective of ordinary players, it may mean that the supply of the GeForce RTX 40 series in the market will decrease, there will be more shortages in the market, prices will remain high, and the deployment of new product lines will slow down.
Due to the high inventory of the GeForce RTX 30 series, Nvidia can also appropriately delay the shipment of mainstream models of the Ada Lovelace architecture to give partners more time to ship. From a financial point of view, Nvidia also needs to ensure as much as possible the revenue of the data center business to reduce losses due to the decline in revenue from the gaming business and new measures.