Arm Holdings Launches Platform for Running AI Applications Without Internet Connectivity

Arm Holdings announced on Monday the launch of its Lumex computing subsystem platform, designed to accelerate artificial intelligence capabilities on mobile devices and to run AI applications entirely offline. The platform represents a major shift toward on-device AI processing, which could enable future Android smartphones to outperform Apple’s iPhones in AI performance, reports Arm Newsroom.

Revolutionary AI Performance Boost

The Lumex platform is based on the new Arm C1 CPU cores and Mali G1-Ultra GPU, both featuring the company’s Scalable Matrix Extension 2 (SME2) architecture. According to Arm, the platform delivers up to 5x faster AI performance, 25% better CPU performance, and 3x greater efficiency compared to previous-generation processors. The Mali G1-Ultra GPU achieves 2x higher ray tracing performance and a 20% graphics performance uplift compared to leading benchmarks.

“AI is becoming increasingly essential for a wide variety of applications, whether for real-time interaction or innovative use cases like AI-powered translation,” said Chris Bergey, Senior Vice President and General Manager at Arm. The technology enables real-time features such as camera neural noise reduction at over 120 frames per second in 1080p resolution, all powered by a single CPU core.

Early Adoption by Leading Chipmakers

According to industry sources, MediaTek’s Dimensity 9500 chipset will be among the first to integrate Arm’s new technology, with its launch expected on September 22. The chip is rumored to feature an Arm “Travis” core clocked at 4.21 GHz and the Mali G1-Ultra GPU, aimed at flagship Android devices launching in 2026.

Samsung is also preparing its Exynos 2600 processor, which will be the world’s first mobile chipset manufactured using a 2nm process and featuring Arm C1 cores. The company confirmed the chip is ready for mass production and is likely to power the Galaxy S26 series, marking Samsung’s return to Exynos processors in its flagship models after the disappointing performance of the Exynos 2500.

A Developer-Ready Ecosystem

Arm has already integrated SME2 support into major development frameworks, including Google’s XNNPACK library for Android, PyTorch ExecuTorch, Alibaba’s MNN, and Microsoft’s ONNX Runtime. This means developers will be able to take advantage of performance improvements without changing their code as soon as Lumex-powered devices reach the market.

According to Android Authority, this could give Android smartphones an AI edge over Apple devices, since Apple’s M4 chip in the iPad supports SME2, but the iPhone lineup does not. The technology promises significant enhancements in text summarization, on-device translation, and voice generation directly on smartphone processors.

The Lumex platform marks Arm’s evolution from an intellectual property provider to a full-fledged platform solutions company, aiming to help device makers bring AI-enabled smartphones to market faster while preserving the power efficiency critical for mobile devices.