TSMC Capacity Tight, Samsung Wins US Cloud Provider's 2nm AI Chip Order via Faraday Through DSP Channel

TSMC Capacity Tight, Samsung Wins nm Cloud AI Order via DSP? Source: Ban Dao Zong Heng Published: 2026-11-03 17:46 Samsung Electronics Foundry generates poster. US cloud vendor switches to DSP partnership route, Faraday wins record-breaking ASIC mega-order, selecting Samsung 2nm GAA process.

Samsung Electronics' foundry business is expected to manufacture 2nm process AI chips for a US cloud vendor through Taiwan IC design service provider Faraday. After directly signing with top customers such as Tesla and Broadcom, Samsung is expanding its high-performance computing (HPC) customer base through the channel of certified DSP (Design Solution Partner) partners, and this order has become the latest landing point of its DSP strategy.

According to industry news on November 3, Faraday disclosed that the company has won a cloud AI chip design project from a US customer. Faraday is an ASIC (Application-Specific Integrated Circuit) design service provider listed on the Taiwan Stock Exchange. It provides chip design and semiconductor IP licensing for fabless customers, and coordinates mass-production supporting work such as wafer stocking and packaging and testing scheduling. It is also a DSP partner certified by Samsung Foundry. Such companies play the role of a "design service middle layer" in the industrial chain: they do not directly manufacture chips, but are deeply involved in full-process coordination from specification definition to mass production and shipment.

This order is regarded by the industry as the largest single project since Faraday's founding, with a contract value of tens of millions of dollars, and the first batch of revenue is expected to be recognized in the first half of 2027. Wafer manufacturing will be undertaken by Samsung Electronics, while Faraday also maintains cooperation with manufacturers such as UMC, and the supply chain arrangement is not tied to a single source. In addition, Faraday is also advancing multiple new 4nm and 2nm projects and plans to complete design tape-outs for four advanced-process ASIC projects in the first half of next year. The company's revenue in the third quarter of this year was NT$3.664 billion (about KRW 150 billion), a year-on-year increase of 13%.

The report said that the US customer initially planned to use the 4nm process, and after comprehensively weighing performance, power consumption and product life cycle, ultimately switched to Samsung's 2nm GAA (Gate-All-Around) process. The 2nm node is regarded as a key watershed in the advanced-process race, with TSMC, Samsung and Intel all committing heavy resources at this node; compared with traditional FinFET, the GAA structure is better in leakage control and power consumption performance, and has therefore become the common choice for each company's 2nm mass-production process. This chip is planned to adopt 3D stacked custom DRAM and hybrid bonding packaging technology: the former increases capacity density through vertical stacking, while the latter replaces traditional bump connections with wafer-level bonding, both of which are mainstream technology routes for high-bandwidth AI chips. However, the amount of non-recurring engineering (NRE) fees, the value of Samsung's subsequent foundry orders and the number of wafers started, as well as the specific identity of the customer, have not been disclosed.

From the perspective of Samsung's customer structure, the acquisition path of this order is clearly different from before. Samsung previously signed a US$16.5 billion foundry contract with Tesla and a memorandum of understanding on memory and foundry cooperation totaling US$200 billion with Broadcom, both of which were direct customer signings. At its third-quarter earnings call on October 29, Samsung said that this year's 2nm order scale had doubled compared with the same period last year, with customers covering leading cloud vendors and AI companies, and predicted that next year the capacity utilization rate of 8nm and below processes will exceed 100%, and HPC revenue will account for more than half of total foundry revenue. In addition to directly signing major customers, covering small and medium-sized design companies through the DSP system is becoming a conventional means for Samsung to complete its order structure.

In Samsung's DSP ecosystem, domestic manufacturers are the main body. According to information on Samsung Foundry's official website, its DSP partners are mainly Korean manufacturers such as ADTechnology, GAONCHIPS, Semifive and CoAsia, and Faraday is currently the only known manufacturer from Taiwan. Most projects of domestic DSPs come from domestic AI chip startups such as DeepX, and the industry generally believes that it is difficult to reach customers of the scale of large US cloud vendors relying only on domestic DSPs. Bringing in overseas DSPs is equivalent to opening an entry point for Samsung to overseas custom chip design demand, which is another reason why Faraday's order is valued at the strategic level.

From the perspective of the industry landscape, this order occurs at a node of structural tightness in advanced-process capacity. TSMC's capacity for 2nm and below advanced processes continues to be tight, and demand from leading design companies and cloud vendors for second suppliers is rising. For Faraday, entering Samsung's 2nm GAA production line means obtaining an optional foundry path in the supply chain of US cloud vendors; for Samsung, it is a key move to penetrate the HPC and cloud AI custom chip market through the DSP channel after directly signing major customers. Some semiconductor industry figures commented that design companies in the US market urgently need alternative supply chains, and Taiwan design service providers choosing Samsung as an alternative may become a supply reference for similar cooperation in the future.

What this order reflects is not limited to the gains and losses at the level of a single company. Demand for cloud vendors' self-developed AI chips continues to expand, and custom chips are gradually growing from a supplementary option beyond Nvidia GPUs into an important pole of computing power supply. Against this background, the collaboration model among foundry, design services and packaging and testing is changing: intermediate roles such as DSP undertake full-process coordination from design to mass production and become a bridge connecting design companies and foundries. For Samsung, the economies of scale of advanced processes require sufficient orders to support them, and the DSP channel can effectively supplement customer sources beyond direct signings; for design service providers such as Faraday, binding to the advanced-process capacity of leading foundries is the key bargaining chip for obtaining orders in the wave of AI custom chips; and for US cloud vendors, one more advanced-process foundry path also means a dual improvement in supply chain resilience and bargaining space.

This article is reproduced from media reports or online platforms and represents the author's personal stance or views. Our reproduction is solely for sharing and does not mean that we approve of or endorse it. If the source attribution is incorrect or your lawful rights and interests are infringed, please contact customer service in a timely manner, and as a neutral platform service provider we will promptly correct, delete or handle it in accordance with the law. Comments: No user comments yet. Latest News More Related Recommendations

Over 50 billion! Semiconductor results released, who is the king of Q3 net profit? 2026-11-03 18:07

DeepSeek's next-generation large model, Ascend chips undertake training 2026-11-03 18:03

Alibaba DAMO Academy XuanTie C950 wins the "China Chip" Annual Major Innovation Breakthrough Product Award 2026-11-03 18:02