Samsung Electronics and Arm have kicked off a joint project to build an on-device AI system-on-chip (SoC) using Samsung's 2nm process, with market speculation pointing to OpenAI as the potential end customer. The move places Samsung directly into the fast-growing custom AI chip market, where rivals like Broadcom are already shipping silicon for AI companies and seeing revenue surge.
The deal structure and what each party brings
Arm approved the initial non-recurring engineering costs in late August, and the two companies moved into substantive development immediately afterward. Arm provides the AI accelerator architecture and RTL-level core design technologies, relays the end customer's requirements to Samsung, and oversees development progress. Samsung's System LSI division handles the full SoC design, while its foundry division will mass-produce the chip on the 2nm node.
The collaboration operates under a Limited Use License model. Arm's architecture is restricted to development for a specific customer and product - Arm won't sell the chip directly to the market. Samsung owns the design-to-manufacturing pipeline and delivers the finished chip to the end customer, collecting revenue from both its design business and its foundry operations.
Samsung, Arm, and OpenAI have all declined to confirm that OpenAI is the final buyer. The connection remains industry speculation.
Why on-device AI is pushing custom silicon forward
AI workloads are shifting from cloud data centers to edge devices. Running inference locally on smartphones and PCs cuts latency, reduces power draw, and keeps sensitive data on the device. But different devices have different constraints around compute, thermals, and cost - a one-size-fits-all chip rarely hits the sweet spot for all of them.
This fragmentation is driving demand for custom AI chips, where the AI company or device maker sets the requirements and a semiconductor partner handles design and manufacturing. Broadcom's latest earnings show the scale of this shift. The company reported $16.7 billion in AI semiconductor revenue for the third quarter, 73% of which came from custom chips. It expects that figure to climb to $21.7 billion in the fourth quarter, with projections of $115 billion and $230 billion in fiscal 2027 and 2028 respectively.
Broadcom is already developing custom silicon for OpenAI, Anthropic, and Google, and has begun shipping the next-generation Google TPU and OpenAI's Jalapeño chip. Reuters has also reported that Anthropic is building an internal custom silicon team and seeking chip partners.
Samsung's 2nm track record and foundry ambitions
For Samsung, the project is about more than a single chip order. Successfully designing and mass-producing an AI SoC on its 2nm process gives the company a reference case to show other AI firms. The integrated "design + manufacturing" model lets Samsung pitch a complete solution rather than competing on foundry services alone.
Samsung has been building toward this. Its foundry division previously worked with Arm, ADTechnology, and Rebellions on an AI CPU chiplet platform, also using the 2nm node and advanced packaging. The OpenAI connection, if confirmed, would add a high-profile name to that 2nm portfolio and strengthen Samsung's position against Broadcom in the custom AI chip segment.
The relationship between Samsung and OpenAI already extends beyond chips. In June, OpenAI announced that all Samsung Electronics employees in South Korea and its global Device eXperience division would use ChatGPT Enterprise and Codex - one of OpenAI's largest enterprise deployments to date. Samsung plans to apply the tools across R&D, manufacturing, marketing, and software development.
Why this matters for IT and development professionals
The shift toward on-device AI and custom silicon changes the compute substrate your code runs on. As inference moves from the cloud to the edge, developers will need to optimize for heterogeneous hardware - chips purpose-built for specific model architectures, power envelopes, and latency targets. Samsung's entry into this market, alongside Broadcom's existing work for OpenAI and Anthropic, signals that custom AI silicon is becoming a standard procurement path for large AI companies. For AI for IT and development teams, understanding the chip supply chain and the performance profiles of these custom SoCs will increasingly inform deployment decisions, toolchain choices, and model optimization strategies.
Your membership also unlocks:
