Android users who have pushed the limits of their device storage will recognize the sudden surge that accompanies an AICore update. Google’s on‑device AI engine, Gemini Nano, lives inside the AICore app, and when it pulls in a new model, the service stores both the old and new binaries for a short period. The temporary footprint can hit 11GB.

To guard against corrupted downloads, Google keeps the dual‑model stack for up to 3 days. This buffer allows an instant rollback if the new model fails, sparing users from a forced reinstall of several gigabytes. Once the update passes stability checks, the obsolete files vanish automatically, restoring freed space.

The practice explains why several Pixel 10 Pro owners reported storage spikes after the latest Gemini upgrade. While the extra data load feels hefty, the fail‑safe design ensures uninterrupted AI features like smart replies and notification summaries. Users need not intervene; the cleanup happens behind the scenes.

Google’s clarification demystifies a previously frustrating symptom of on‑device AI growth. By transparently outlining the temporary storage policy, the company reassures consumers that the spike is normal and self‑reverting. The move also signals a broader industry trend toward robust, self‑repairing AI models that prioritize reliability over immediate storage efficiency.