ML Drift: Next-Gen GPU AI/ML Inference at the Edge

ML Drift is Google AI Edge's high-performance, universal GPU compute framework for accelerating on-device AI and ML inference. It's the next-gen GPU delegate for LiteRT (TensorFlow Lite's successor), built to extract real performance from the GPU hardware already sitting in every phone and laptop.

Across edge and workstation environments alike, peak throughput only tells half the story: memory footprint dictates what workloads can actually run concurrently. In Gemma benchmarks, ML Drift achieves up to 12% lower memory overhead than other frameworks, freeing memory for other local tasks.

Battle-tested at scale

ML Drift is already running in production across millions of devices daily, powering features across Google's ecosystem:

• YouTube Shorts migrated its segmentation-based effects to ML Drift — up to 40% reduction in average frame latency on both Android and iOS, letting creators apply high-fidelity visual effects in real time without dropping camera frames.
• Google Photos integrated ML Drift across computational photography and segmentation pipelines — up to 2-second speedup versus the legacy GPU delegate, making photo enhancements feel instantaneous.
• Google Chrome integrated ML Drift into its AI runtime to power native hardware-accelerated Gemini Nano models for the Built-In AI APIs (Prompt, Summarize, Writer) across desktop platforms.

At the frontier of on-device capability it also powers Google AI Edge apps like Gallery, Foresight, and Eloquent.

The partner ecosystem

Beyond Google products, developer partners have adopted ML Drift:

• Adobe Lightroom and Photoshop upgraded key AI features — Select Subject, Select Sky, Adaptive Portrait — with ML Drift, achieving up to 30% faster on-device performance for pro-quality photo editing on mobile.
• Snap integrated LiteRT to power on-device inference on GPU across Android.

Bottom line

ML Drift is the GPU story for on-device AI in 2026: lower memory overhead, genuinely faster on-device inference, and already proven at Google's own scale (Chrome, YouTube Shorts, Photos, Meet). For anyone shipping on-device AI/ML — from camera effects to built-in browser AI — it's the backend that makes "runs locally" actually fast, not just possible.