South Korean semiconductor giant Samsung Electronics (005930-KS) has officially launched a joint development program with chip architecture firm Arm for an on-device artificial intelligence (AI) system-on-chip (SoC), targeting AI computing needs in end devices such as smartphones. Notably, market speculation suggests the ultimate customer for this custom chip may be OpenAI, which has been increasingly active in the AI hardware space.
South Korean media, citing industry sources, reported that Arm approved the project’s initial non-recurring engineering (NRE) costs in late August, and the two parties immediately entered the substantive development phase. Under the division of labor, Arm is responsible for providing the AI accelerator (AIC) architecture and core design technologies such as RTL, conveying end-customer requirements to Samsung, and overseeing overall development progress. Samsung Electronics’ System LSI division handles the specific SoC design work, while the foundry division plans to mass-produce using the most advanced 2nm process.
Currently, Samsung, Arm, and OpenAI have all declined to confirm that OpenAI is the final buyer of this chip, and related claims remain speculation by market and industry insiders. However, if the reports prove accurate, this would represent a critical breakthrough for Samsung in the custom AI chip market.
On-Device AI Rises, Custom Chips Become the New Battleground
The core backdrop of this collaboration is that AI computing models are gradually shifting from cloud data centers to end devices. Traditional AI services rely heavily on cloud servers for inference and data processing, while on-device AI allows smartphones, personal computers, and other devices to complete some AI computing locally. This not only reduces dependence on cloud computing power but also improves response speed, power consumption, and data privacy protection.
As generative AI features accelerate their integration into various consumer electronics, different devices present distinct requirements for AI computing capability, power consumption, latency, and cost—making it difficult for general-purpose chips to fully satisfy all use cases. This has made the custom AI chip model—where AI companies or end-device manufacturers specify requirements, and chip design and manufacturing firms jointly develop solutions—an emerging direction in the semiconductor industry.
From a business model perspective, Samsung and Arm are operating under a Limited Use License (LUL) framework. In other words, the architecture and technology provided by Arm are not for Samsung to freely apply across all product lines, but are restricted to development for specific customers and specific products. Arm primarily plays the role of core technology provider and co-development partner in the project, and will not directly sell this chip to the market.
Samsung, meanwhile, shoulders the primary work from SoC design to manufacturing, delivering the completed chip directly to the end customer and realizing commercial revenue. This gives Samsung the opportunity to generate income from both its System LSI design business and its foundry operations—leveraging its vertically integrated design and manufacturing advantages more effectively than simply taking on foundry orders alone.
2nm Process Track Record: A Critical Litmus Test for Samsung’s Foundry Business
For Samsung, the strategic significance of this project extends far beyond securing a single chip order. The System LSI division handling SoC design and the foundry division handling 2nm mass production can achieve synergy within the same project, forming a complete “design + manufacturing” AI chip solution. If the product ultimately reaches mass production, Samsung will be able to simultaneously demonstrate its SoC design capabilities and its 2nm process manufacturing prowess, accumulating a track record for pursuing more custom chip projects from AI companies and technology firms in the future.
Particularly in the foundry space, actual participation in the design and mass production of an on-device AI chip allows Samsung to accumulate valuable experience serving AI customers with its 2nm process. If successfully commercialized, the results will serve as an important case study for Samsung to promote its 2nm foundry capabilities to other potential customers.
Samsung has been actively positioning itself in AI chips and advanced processes through partners including Arm in recent years. Previously, Samsung’s foundry division collaborated with Arm, ADTechnology, and Rebellions to develop an AI CPU chiplet platform, also utilizing Samsung’s 2nm process and advanced packaging technology—demonstrating Samsung’s efforts to enter different AI computing markets with its 2nm process.
Custom AI Chip Market Growing at High Speed
The growth momentum of the AI custom chip market is evident from chip giant Broadcom’s (AVGO-US) latest earnings report. The company’s third-quarter AI semiconductor revenue reached $16.7 billion (approximately NT$530 billion), with 73% coming from custom chips. The company also expects fourth-quarter AI semiconductor revenue to rise further to $21.7 billion (approximately NT$690 billion), while projecting AI semiconductor revenue to double consecutively in fiscal years 2027 and 2028, reaching $115 billion (approximately NT$3.6 trillion) and $230 billion (approximately NT$7.3 trillion) respectively.
Broadcom is already developing custom chips for AI players including OpenAI, Anthropic, and Google, and has begun shipping the next-generation Google TPU and OpenAI’s Jalapeño chip. Additionally, Reuters previously reported that Anthropic is building a Custom Silicon Team and seeking custom chip partners, further confirming that demand for dedicated AI chips among AI model companies is rapidly heating up.
Against this industry backdrop, the significance of the Samsung-Arm collaboration lies in establishing complete service capabilities for AI custom chips. If the final product reaches mass production, Samsung will have the opportunity to establish a firm foothold in the custom AI chip market and compete with rivals such as Broadcom through its integrated “design + manufacturing” model.
The partnership between Samsung and OpenAI is also deepening. OpenAI announced in June that all Samsung Electronics employees in South Korea and its global Device eXperience (DX) division employees will use ChatGPT Enterprise and Codex—one of OpenAI’s largest enterprise deployments to date. Samsung plans to apply these tools across R&D, manufacturing, marketing, product development, and software.
The collaboration remains in the development stage, and whether OpenAI ultimately becomes the customer, along with the chip’s specific specifications, shipment timeline, and production scale, have yet to receive official confirmation. The market will need to await further product development progress and customer finalization. If all goes smoothly, Samsung will not only secure an on-device AI chip project but potentially establish a complete track record of serving AI customers with its 2nm process—opening new growth curves for its System LSI and foundry businesses amid the wave of AI computing demand spreading from data centers to end devices.