News reports that TSMC is about to finalize future 3nm and 2nm customers
According to industry insiders, due to the increased difficulty of process technology and TSMC's one-stop service including advanced backend packaging, customers of TSMC's 3nm and 2nm processes are unlikely to transfer orders.
![](/upfile/images/17/20231212232543171.jpg)
Sources say that TSMC is about to finalize its future 3nm and 2nm customers. In 2025, the pure wafer foundry will begin producing 2nm chips, while its 3nm chip production will increase quarterly in 2024.
In addition to Apple, AMD, NVIDIA, Broadcom, MediaTek, and Qualcomm are also customers of TSMC's 3nm and 2nm chips. These major customers are unlikely to reduce TSMC's production of 3nm and 2nm wafers before 2027.
Although Nvidia CEO Huang Renxun and CFO Colette Kress, as well as executives from AMD, Qualcomm, and MediaTek, have previously expressed the possibility of collaborating with other wafer foundries, their primary goal is to negotiate prices with TSMC or ease the pressure on the US government to urgently promote domestic semiconductor manufacturing.
Recently, there have been rumors that companies such as Nvidia, Qualcomm, MediaTek, and AMD are interested in issuing 3nm and 2nm chip orders to Samsung foundries and Intel foundries.
Sources say that Nvidia has placed all orders for GeForce RTX 40 series graphics cards and AI GPUs with TSMC, with a focus on storage chips in cooperation with Samsung. Nvidia has not yet confirmed whether to introduce Intel foundry.
Due to TSMC's CoWoS production capacity being in short supply, it is reported that Samsung is working hard to secure some advanced packaging orders from Nvidia, followed by orders for processes below 7nm. However, sources have revealed that Nvidia's 2024 roadmap indicates that the company is still striving to acquire CoWoS production capacity from TSMC and has no plans to transfer some orders to Samsung.
In addition, given Intel's focus on marketing CPUs and GPUs to take advantage of the huge financial opportunities brought by the boom in artificial intelligence (AI) chips, Nvidia has no reason to limit its cooperation with TSMC and may transfer orders to Intel's foundry business.
It should be pointed out that the degree of separation between Intel's design and foundry business is limited to internal factors and far from achieving complete separation between AMD and Global Foundries. Therefore, even if Intel offers a lower price, the likelihood of Nvidia turning to Intel is low.
Sources say that if such an event occurs, it is almost certainly due to coercion or trade demands from the US government.
AMD's situation is similar to that of Nvidia, but it is more difficult for AMD to transfer orders from TSMC. AMD has previously paid a large fee to Global Foundries to gain greater autonomy in establishing high-performance product roadmaps with other foundries, allowing it to utilize TSMC's technology to produce products below 7nm. The tactical decisions involved are crucial for AMD's operational recovery.