Google Releases ML Drift, GPU Foundation for On-Device AI

Google has released ML Drift, an open-source GPU computing engine designed to run AI models on devices. Provided under the Apache 2.0 license, anyone can freely use and modify it. This tool is designed to efficiently execute AI inference—the process by which trained models produce results—on devices such as smartphones and PCs.

Historically, AI processing has primarily been handled by transmitting data to cloud servers. However, in recent years, there has been growing interest in "on-device AI," which runs AI directly on the device itself. This approach offers advantages including reduced communication latency and better privacy protection, and is increasingly becoming a feature on smartphones and notebooks. Within this context, ML Drift is positioned as an infrastructure-layer tool that facilitates on-device AI implementation.

The challenge ML Drift addresses is the diversity of GPUs (graphics processing chips). Across different devices—Android smartphones, iPhones, Macs, and web browsers—GPU types and the low-level APIs (connection points for programs to interact with hardware) that control them vary significantly, requiring developers to write platform-specific code. ML Drift is designed to abstract this complexity, allowing developers to work without being aware of these differences.

The supported APIs are OpenGL ES, OpenCL, Metal, and WebGPU—four types that cover virtually all major platforms. OpenGL ES targets mobile platforms like Android, Metal is for Apple devices, and WebGPU handles browser-based execution. OpenCL is used for general-purpose parallel computing and can support a wide range of GPU environments.

From a developer's perspective, the value of ML Drift lies in the ability to "write code once and run it on many devices." Previously, device-specific optimization required substantial effort; with ML Drift, this burden is significantly reduced. It has the potential to shorten AI application and model development cycles and make deployment across more devices practically feasible.

The open-source release is also a critical aspect. The Apache 2.0 license has minimal restrictions and permits broad use by companies, individuals, and research institutions alike. By providing AI inference infrastructure as an open platform, Google demonstrates an intent to promote on-device AI adoption across the entire ecosystem.

On-device AI is an area where smartphone manufacturers and semiconductor companies are investing heavily, and in the future, "multi-platform inference frameworks" like ML Drift may become a competitive focal point. As the diversity of devices increases, the value of abstraction layers like ML Drift will only grow, making the adoption rate within developer communities and integration patterns with other frameworks key areas to watch going forward.