Arm collaborates with Samsung on AI chip project for 2nm process

Arm Holdings and Samsung Electronics are teaming up to build next-generation AI chips using Samsung’s 2nm manufacturing process. Arm approved the non-recurring engineering fees for the project at the end of August 2026, officially kicking off the joint R&D effort.

Arm’s stock climbed roughly 2.5–3% on the news.

What they’re actually building

The collaboration is focused on developing system-on-chips, or SoCs, purpose-built for on-device AI. The division of labor is straightforward. Arm brings its AI accelerator architecture and register-transfer level design expertise. Samsung’s System LSI division handles SoC design and integration using Arm’s intellectual property. Samsung’s Foundry division takes care of manufacturing, advanced packaging, and commercialization.

The chips will be fabricated on Samsung’s 2nm Gate-All-Around, or GAA, process node. GAA is the successor to FinFET transistor architecture, which has dominated chipmaking for over a decade. The basic idea: instead of the gate wrapping around three sides of a transistor channel, GAA wraps around all four. That gives chipmakers better control over electrical current, which translates to lower power consumption and higher performance at smaller scales.

Why on-device AI matters

The customized chips coming out of this collaboration are specifically targeting smartphones and other terminal devices. The goal is to give device manufacturers a way to offer powerful AI features without requiring constant cloud connectivity, reducing latency and eliminating the need to send personal data off-device.

Building on existing groundwork

This isn’t the first time Arm and Samsung have collaborated on advanced chip technology. In February 2024, the two companies worked together to optimize Arm’s Cortex-X and Cortex-A processor cores for Samsung’s 2nm GAA manufacturing nodes.

Then in October 2024, the pair announced a joint effort to develop a multi-party AI CPU chiplet platform. The claimed result from that chiplet work was efficiency gains of 2-3x for large generative AI models.

The current SoC collaboration builds on both of those foundations. The optimized cores give the chips raw processing muscle. The chiplet architecture provides flexibility and efficiency. And the 2nm GAA process delivers the transistor density needed to pack all of that into a form factor that fits inside a smartphone.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.