Google Finally Explains Why the AICore App Is Eating up Your Storage

Why It Matters

The temporary storage spike can surprise users and affect device capacity, while the redundancy safeguards continuous AI functionality—a trade‑off that shapes the user experience of on‑device intelligence.

Key Takeaways

• AICore may consume up to 11 GB during model updates.
• Google keeps old and new AI models for three days.
• Storage is automatically reclaimed once the new model stabilizes.
• AICore powers Gemini Nano offline AI on Pixel 10 Pro devices.
• No user control to limit temporary AICore storage.

Pulse Analysis

On‑device artificial intelligence has become a cornerstone of modern smartphones, with Google’s Gemini Nano leading the charge on Android flagships. The AICore service acts as the engine behind features like smart replies and notification summaries, running AI inference locally to preserve privacy and reduce latency. However, the models that power these capabilities are sizable, often several gigabytes, and must be updated periodically to improve accuracy and add new functions. When a new model is pushed, AICore downloads it alongside the existing version, creating a temporary storage surge that can reach double‑digit gigabytes.

Google’s explanation hinges on reliability. By retaining both the old and new models for up to three days, AICore creates a fail‑safe that allows an instant rollback if the fresh model proves unstable. This approach mirrors enterprise software deployment practices, where redundancy mitigates the risk of service interruption. For end users, the upside is uninterrupted AI functionality, but the downside is a sudden, unexplained dip in available storage—especially problematic on devices with limited capacity. Competing ecosystems, such as Apple’s on‑device ML framework, typically employ more aggressive pruning, which can lead to brief feature outages but avoids large storage spikes.

The broader implication for manufacturers and developers is a call to balance robustness with user‑centric storage management. While Google’s automatic cleanup alleviates long‑term impact, offering a manual override or configurable retention window could empower power users and prevent surprise storage constraints. As on‑device AI models continue to grow in size and complexity, future Android updates may need smarter compression or incremental patching techniques. For now, users should monitor AICore’s storage footprint during update windows and consider clearing cache if the temporary usage persists beyond the three‑day window.