Why Your Mac’s Storage Is Hitting 90% in the AI Era
If your Mac used to stay comfortably under 70% full and suddenly lives at 85–95%, you’re not imagining it. The storage crisis of 2025–2026 is real, and it has almost nothing to do with photos, videos, or old iTunes libraries.
The culprit is the AI-coding stack that now sits on almost every developer and power-user Mac.
The new junk is heavier and stickier
Classic Mac clutter was relatively polite: browser caches, app support folders, a few leftover installers. The AI era introduced entirely different classes of data:
1. Local model weights Ollama, LM Studio, Hugging Face Hub, and similar tools download multi-gigabyte (sometimes 20–40+ GB) model files. People keep several quantizations of the same model “just in case.” Unused models from two months ago still sit there because nothing automatically prunes them.
2. AI coding-agent residue Cursor, Claude Code, Windsurf, and similar tools leave behind:
- Massive state.vscdb and workspaceStorage folders (reports of 50–100 GB growth are no longer rare)
- Session transcripts and agent logs that accumulate for weeks
- Checkpoint / local-history packs that never get cleaned
- MCP server logs and temporary worktrees
These tools are designed for continuous, high-volume interaction. They are not designed to be tidy.
3. Apple Intelligence and on-device models macOS itself now ships generative models, visual models, and enhanced Siri assets. These routinely claim 7–14 GB (sometimes more with multiple languages or accessibility features) and live in system locations that Finder never shows clearly.
4. The usual developer suspects, amplified node_modules graveyards from abandoned side projects, Xcode DerivedData, Docker images and build caches, Homebrew and npm caches—all still exist, but they now share the drive with the AI payload above. The combination pushes many 512 GB and even 1 TB machines into the red.
Why 90% is the danger zone on a Mac
macOS does not handle near-full SSDs gracefully. Once free space drops below roughly 10–15%, you start seeing:
- Slower file operations and Spotlight indexing
- Failed software updates
- Aggressive swap and memory pressure (especially painful when local models are loaded)
- Occasional mysterious “System Data” inflation that is hard to diagnose
The OS will not protect you the way it should. Many users only notice when an update fails or the machine starts feeling sluggish under AI workloads.
Traditional cleaners were built for a different decade
Most Mac cleaners still look for the junk of 2015: old iPhoto libraries, language files, and generic caches. They have no concept of:
- Which Ollama models haven’t been touched in 60 days
- Which Cursor workspace packs are safe to remove
- What Claude Code transcripts are regenerable versus what contains unique context
- The difference between a regenerable Hugging Face cache and something expensive to re-download
They either miss the biggest offenders or treat them as opaque “System Data,” forcing you to delete by hand (and risk breaking something).
A cleaner that understands the AI-coding era
This is exactly why tools built specifically for the current workload matter. Sweep was designed around the actual files that AI developers and power users accumulate in 2026: Claude Code caches and transcripts, Cursor and Windsurf editor data, unused Ollama models, Hugging Face hubs, node_modules graveyards, and the rest.
It classifies findings as Safe (regenerated automatically) or Review first, never touches credentials or settings, moves everything into a local 7-day undo vault instead of permanently deleting, and explains every byte in plain English before anything is touched.
Scanning stays free. You see exactly what is eating the drive before you reclaim it.
If your storage bar has been creeping toward 90% and the usual suspects no longer explain it, the AI stack is almost certainly the reason. The good news is that most of that space is recoverable once you can see it clearly.
Your Mac doesn’t have to live at the edge of full.
Comments