Unofficial sources reveal that the new Exynos 2500 processor to be launched by Samsung may include an AI accelerator developed by Google. This move is a significant step towards improving the artificial intelligence capabilities of mobile devices.
According to information obtained from anonymous user X, known as Connor/콘어/コナー, Samsung will work with Google to integrate a TPU (Tensor Processing Unit) into its next-generation Exynos chips. This TPU, designed by Google, was previously used in Pixel phones and is an integral part of Tensor chipsets, developed in partnership with Samsung.
The TPU is an AI accelerator used by Google products and can even be rented in more powerful variants through Google Cloud for those working with artificial intelligence. Its integration into Exynos 2500 chips would give phones, starting with the Galaxy S25 series, access to Google’s machine learning APIs, allowing AI models to run directly on the device.
Another leak reveals that the Exynos 2500 will include two neural processing units: G-NPU and S-NPU. These represent the evolution of the NPU (Neural Processing Unit) units included in Exynos chips since the 2200 series. The “G” unit is designed for flexibility, being a more adaptable variant of the NPU with fixed functions.
In terms of architecture, it seems that the Galaxy S25 series will integrate both Exynos 2500 and Snapdragon 8 Gen 4 chips, contrary to previous reports. Qualcomm, the manufacturer of Snapdragon chips, is not inferior in terms of AI acceleration, but its architecture differs from that of Google’s TPU. However, the possibility of integrating these two different technologies within Android’s machine learning APIs remains an interesting prospect.
According to unofficial sources, the Exynos 2500 will be manufactured using Samsung’s manufacturing technology based on the company’s second 3nm node (3nm GAP/SF3). The processor will include a Cortex-X5 core at 3.2GHz, three A730 cores with frequencies between 2.3 and 2.5GHz, along with two more A730 cores at a lower frequency and four A520 cores. Although the details of the Xclipse 950 GPU are not yet available, it is expected to deliver outstanding performance.
In comparison, the Tensor G4 (the likely source of the TPU design) will be manufactured using Samsung’s fourth-node 4LPP+ technology, following the breakdown of talks with TSMC. This suggests that the Exynos 2500 could be faster and more efficient than other solutions, including in artificial intelligence tasks, which until now have been Google’s forte.
As information continues to trickle out and precise data is revealed, it’s clear that Samsung remains committed to improving its user experience by integrating the latest AI technologies into its top-of-the-line chips. We look forward to seeing how this evolution will impact the mobile market and user experience.
Tags: Samsung Exynos vision acceleration
