In a new blog post, Arm has confirmed Samsung’s upcoming Exynos 2600 SoC will support its SME2 instruction set out of the box. It will allow the chip to process on-device AI tasks, such as object detection, faster.
Samsung’s highly anticipated Exynos 2600 SoC is set to launch alongside the Galaxy S26 series on February 25. Earlier leaks suggested that the chip could rival—and in some cases outperform—Qualcomm’s Snapdragon 8 Elite Gen 5 for Galaxy. Now, Arm has officially confirmed that the Exynos 2600 will support SME2 (Scalable Matrix Extensions), offering a glimpse into its hardware capabilities. According to Arm, SME2 can boost AI-powered applications, like object detection, by up to 70%.
For context, Arm introduced SME2 alongside its latest CPU cores: Lumex C1-Ultra, C1-Premium, C1-Pro, and C1-Nano. This strongly indicates that the Exynos 2600 will use a combination of these cores. Geekbench listings reveal a 10-core CPU configuration, which likely includes one C1-Ultra core, three C1-Premium cores, and six C1-Nano cores—consistent with previous Exynos designs. Arm’s Stefan Rosinger further supports this theory, stating:
“As on-device AI becomes central to the mobile experience, efficiency and responsiveness are increasingly critical. Built on Arm compute subsystems with SME2-enabled C1-Ultra and C1-Pro, Exynos 2600 leverages SME2 to expand the potential of CPU-based AI, reducing the latency associated with offloading to discrete accelerators and making it well-suited for short, interactive, and real-time AI workloads.”
Interestingly, Rosinger’s statement only mentions the C1-Ultra and C1-Pro cores by name. This could imply that the Exynos 2600 might rely solely on these cores, potentially excluding the C1-Premium cores. Such a decision would make sense, as the C1-Premium cores could strain the chip’s thermal budget. Samsung is likely to play it safe, as the Exynos 2600’s performance will be critical for the future of Samsung Foundry’s SF2 node and its ability to attract major clients like Qualcomm and Nvidia.
