Next year, Samsung will be focusing on incorporating AI capabilities into its devices, particularly with the Exynos 2500 chipset, as reported by X poster Connor/콘어/コナー. Just like how Samsung utilizes an AMD-designed GPU in its top-of-the-line chips, it will also be integrating a Google-designed TPU.
The TPU, or Tensor Processing Unit, has been a pivotal component in Pixel phones for quite some time, and it will be a key feature in the Tensor chipsets that are being jointly developed by Samsung. This particular type of AI accelerator is extensively utilized in various Google products, and larger versions of the TPU can even be rented from Google Cloud for individuals working on AI projects.
This development will grant devices powered by the Exynos 2500 – such as the Galaxy S25 series – access to Google’s machine learning APIs for local model processing. Presently, Galaxy AI heavily relies on cloud resources, but this move signifies a shift towards more robust AI capabilities directly on the device itself.
As per Connor’s post, the new flagship chipset is also set to include two Neural Processing Units – G-NPU and S-NPU. The “G” variant is a more adaptable version compared to the fixed-function NPU and has been a staple in the Exynos lineup since the 2200 series.
A recent report indicates that the Galaxy S25 series will utilize both the Exynos 2500 and Snapdragon 8 Gen 4 chips, contradicting an earlier rumor. While Qualcomm has also been enhancing AI capabilities in its processors, the differing approaches between the TPU and Qualcomm’s AI Engine raise questions that Android’s machine learning APIs may need to address.
Leaked details suggest that the Exynos 2500 will be manufactured using Samsung’s second generation 3nm node (3nm GAP/SF3), featuring a Cortex-X5 clocked at 3.2GHz, three A730 cores running at 2.3-2.5GHz, additional A730 cores at lower frequencies, and four A520 cores. While details on the Xclipse 950 GPU are scarce, it is expected to be a key component of the chipset.
On the other hand, the Tensor G4 chip (which likely influenced the TPU design) will be manufactured on Samsung’s 4LPP+ node after negotiations with TSMC fell through. This suggests that the Exynos 2500 could offer superior speed and efficiency, especially in AI-related tasks, an area where Google has traditionally excelled.
