ASIC is booming, and tech giants are competing for the 3-nanometer project! Morgan Stanley: Taiwan Semiconductor will participate in all production, and World Semiconductor's growth potential is even greater

Morgan Stanley believes that despite facing competition from NVIDIA's GPUs, the AI ASIC (Application-Specific Integrated Circuit) market will continue to expand, with 3nm projects becoming key competitive areas.

Morgan Stanley analyst Charlie Chan stated in a research report on December 15 that the ASIC market size will continue to grow, with the AI ASIC market expected to increase from $12 billion to $30 billion between 2024 and 2027, where 3nm projects will become key competitive areas.

Analysts noted that several key 3nm ASIC projects are nearing finalization, including Amazon AWS's 3nm Trainium project, Google's 3nm TPU project, Microsoft's 3nm Maia200 project, and Meta's 3nm AI network project. Taiwan Semiconductor is expected to continue benefiting from this, but considering business exposure and short-term catalytic upside potential, 世芯电子 (Sicong Electronics) is more worthy of investors' attention.

世芯电子， Marvell, and other manufacturers are major players

In the AWS 3nm Trainium project, 世芯电子 and Marvell are engaged in fierce competition, with Marvell increasing its CoWoS allocation.

Morgan Stanley's senior electronics analyst Joe Moore stated that Marvell's CoWoS allocation has increased to about 70,000 units, which means approximately 1.2 million Trainium chips, including Trainium 2.5, which is migrating to HBM for greater memory density.

Analysts believe that the Trainium 2 market size is continuing to expand (growing by 30%), which is a good sign for the future market size of Trainium 3 for 世芯电子。 The strong execution of 世芯电子 makes it likely to be responsible for 3nm backend design services. The next round of 2nm projects (Trainium 4) will witness a new round of competition between 世芯电子 and Marvell, but the awarding of this project will not be determined until the second half of 2025.

In the Google 3nm TPU project, MediaTek is making progress.

Currently, the Google TPU v7 3nm project is experiencing delays due to customer design plans, with mass production expected to begin in 2026. Given its high performance, this project focuses more on training. Meanwhile, Broadcom will launch another TPU version focused on inference in 2026, which may be more efficient in inference.

In the Microsoft 3nm Maia200 project, Marvell has an advantage in the enhanced version of this project. Although Creative Electronics has seen increased demand for cryptocurrency mining chips, its share in this project has decreased.

Broadcom and 世芯电子 are simultaneously competing for the Meta 3nm AI network project The MTIA v3 artificial intelligence accelerator is expected to migrate to HBM and CoWoS by 2026, continuing to use the RISC-V IP from Andes Technology. Industry research indicates that Meta is not only customizing switch networks (links) for AI data centers but also customizing RDMA-type network chips. Given Broadcom's strong SerDes IP, Morgan Stanley believes it should be the preferred choice.

Preferred stocks shift from TSMC to ASE Technology

Morgan Stanley believes that although TSMC continues to benefit from the 3nm project, ASE Technology may be a better choice.

Analysts state that TSMC currently produces all of the aforementioned ASIC chips. TSMC will allocate 27% of its CoWoS capacity for ASICs in 2025, which has higher profit margins. By 2026, total revenue from AI semiconductors will account for 30% of its revenue.

However, for investors, ASE Technology may be more worthy of attention in the short term (stronger short-term catalysts and better upside potential).

Supply chain investigations show that TSMC will allocate CoWoS - R packaging capacity for Annapurna Lab in the second half of 2025, which may be related to ASE Technology's 3nm chip production. This aligns with Amazon AWS's recent announcement that it will launch Trainium3 by the end of 2025.

Analysts emphasize that the 3nm project tape-out in early 2025 will serve as a strong catalyst for stock prices — it helps alleviate investor concerns about competition from Marvell in the 3nm project. With the expansion of global AI capital expenditures and ASE Technology's solid partnerships with TSMC, IP suppliers, and AI customers, its strong growth momentum should be restored in 2026